124
SmartMiner INTERNAL 2017-09-07 Huawei confidential. No spreading without permission. Page 1 of 124 SmartMiner Quick Start Common Operation Pages Page for Editing a Process Page for editing a process A process editing page is comprised of the following parts: Toolbar: Place the pointer over an icon on the toolbar, and you can view the matching function. Click the icon, and you can perform the corresponding operation. Node panel: To add a node, select the node in the node panel and click in the process editing area. Process editing area: Connect nodes based on your service requirements to complete process configuration. Process exception information area: If an exception occurs when the process is saved, the system will display exception details and a matching solution in the area.

developer.huawei.comdeveloper.huawei.com/ict/files/en/include/Universe/pdf/data-mining.pdf · SmartMiner INTERNAL 2017-09-07 Huawei confidential. No spreading without permission

Embed Size (px)

Citation preview

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 1 of 124

SmartMiner

Quick Start

Common Operation Pages

Page for Editing a Process

Page for editing a process

A process editing page is comprised of the following parts:

Toolbar: Place the pointer over an icon on the toolbar, and you can view the matching function. Click

the icon, and you can perform the corresponding operation.

Node panel: To add a node, select the node in the node panel and click in the process editing area.

Process editing area: Connect nodes based on your service requirements to complete process

configuration.

Process exception information area: If an exception occurs when the process is saved, the system will

display exception details and a matching solution in the area.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 2 of 124

Page for Viewing Data

Page for viewing data

You can view source data files and result files in the Data directory of a project.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 3 of 124

Page for Viewing Evaluation Results

Page for viewing evaluation results

You can view model evaluation files in the Evaluation directory of a project.

Page for Viewing Models

Page for viewing models

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 4 of 124

You can view model files in the Model directory of a project.

Page for Viewing Analysis Results

Page for viewing analysis results

You can view analysis result files in the Analytic directory of a project.

Service Application

Node Description

Source Nodes Source nodes include the TextImport, FolderImport and DatabaseImport nodes.

ImportText Node

Function

The ImportText node reads data from a text file containing variable-length fields record by record. The

number of fields in a text file is fixed, and fields are separated by fixed separators. The number of

characters in a field is changeable.

Restriction

The corresponding project and process have been added.

Parameter Description

Parameters

Parameter Description

Data File

File Click Select, the Select page is displayed. By default, the first 100 rows of a

data file can be previewed. The options are as follows:

● File System Type:

– Local: Select the data files in the Projects/Data directory of the

HDFS.

For example, select data files in the

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 5 of 124

Parameter Description

${smart_project_home_dir}/Projects/ONE/Data/ directory.

${smart_project_home_dir}: set the parameter in the

${SmartMiner_HOME}/conf/smartminer.properties file.

ONE: project name.

– HDFS: Select a data file from the Hadoop distributed file system

(HDFS).

NOTE

If the file system is HDFS, the ImportText node exports files from each project directory in /smartminer/inputdir of the HDFS to the mapping project directories in /smartminer/outputdir.

You can configure the directory by setting the following parameters in ${SmartMiner_HOME}/conf/smartminer.properties as a SmartMiner user. Restart the SmartMiner for the settings to take effect.

smart_hdfs_input_dir=/smartminer/inputdir

smart_hdfs_output_dir=/smartminer/inputdir

– FTP: Select a data file from the FTP server.

NOTE

You need to enable the FTP service before selecting a data file. For details, see Configuring the FTP Service.

● File: Enter a file name in the text box and click Query. All files that

meet the search criteria are displayed on the page. By default, all files are

selected. A file name can contain an expression, for example,

sm_user_retain_#date(yyyyMMddHHmmss)#.csv.

File Uploads a local file to the node server.

Read field names from

the file

Specifies whether to read field names. The parameter is selected by default.

● If the parameter is selected, the TextImport node will read the first row of

the text file as field names.

● If the parameter is not selected, the TextImport node will generate field

names automatically, for example, FIELD1 and FIELD2.

Metadata

Field Name Field name in a data file.

● If the TextImport node reads field names, the first row in the data file is

read as field names.

● If the TextImport node does not read field names, the node generates

field names automatically.

Filter

NOTE

After the field name is modified, the mapping field names on subsequent nodes also need to be modified.

Click Export, and the system will export a metadata file that contains values of the fieldName, dataType, and format fields.

Click Restore, and metadata, such as fieldName and dataType, will be restored to its factory default.

Select a Metadata File Select a metadata file and click Import, and the system will import values of

the New Field, DataType, and Format fields. If the TextImport node has a

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 6 of 124

Parameter Description

different number of fields from the metadata file, the system notifies you

that the number of fields in the TextImport node does not match that in the

metadata file.

● If the TextImport node has more fields than the metadata file, the system

notifies you that the number of fields in the TextImport node does not

match that in the metadata file after reading the file.

● If the TextImport node has fewer fields than the metadata file, the system

notifies you that the number of fields in the TextImport node does not

match that in the metadata file after setting the ImportText node with the

imported data.

ImportFeatureLibrary Node

Function

The ImportFeatureLibrary node combines corresponding fields in feature files based on a specified

prediction field and multiple specified feature fields to generate sample data required for data mining. Files

involved in field combination must have the same primary key, for example, user ID. Invalid data is filtered

out during field combination.

Restriction The corresponding project and process have been added.

Feature files and features have been created.

Parameter Description

Parameters

Parameter Description

Prediction Field You can set the forecast field in either of the

following ways:

● Click the text box:

– Click Select Field: select a field in the

feature list on the page that is displayed.

– Click Customize Field: select a field in

the feature list as a reference field, and

then customize a forecast field based on

the expression.

● Click Importing External Tag Data. The data

to be imported must have two fields: primary

key and forecast field.

Input Field Click the text box and select input fields on the

page that is displayed.

● On the Input Field tab page, you can query

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 7 of 124

Parameter Description

features in the feature query area. The query

results are displayed in a list. Features selected

as input fields are displayed in a list at the

bottom of the tab page.

● On the Auto tab page, you can configure

analysis counters and click Recommendation

Similarity. The system then automatically

analyzes the correlation between the prediction

field and input fields, selects features based on

the correlation evaluation counter threshold,

and displays the selected features in a feature

list. Features selected on the Auto and Input

Field tab pages are combined.

Prediction Periods The default value is 1. The value of this parameter

must be equal to or less than the value of

Maximum storage duration (months) minus one.

For example, if the maximum storage duration is

three months, the maximum value of this

parameter is 2.

Sampling Conditions Click the text box. The page for editing feature

conditions is displayed.

Sampling Periods The default value is 1. The value of this parameter

must be equal to or less than the value of

Maximum storage duration (months) minus the

maximum value of Prediction Periods. For

example, if the maximum storage duration is three

months and the maximum value of Prediction

Periods is 2, the value of the parameter is 1.

Input Sampling After you click Month, the input month source

and forecast month source of the combined sample

data are displayed at the bottom of the page.

ImportFolder Node

Function

The ImportFolder node imports folders and displays folder and text information.

Restriction

The corresponding project and process have been added.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 8 of 124

Parameter Description

Parameters

Parameter Description

File Folder Selects a folder where the file to be uploaded is

located.

NOTE

If FTP is used to upload folders, set the encoding code for FTP uploading to the same value on the Linux server. If the encoding codes are different, Chinese characters in folder or file names are displayed as garbled characters.

Folder Displays details about a selected folde.

Character Set Encoding method of a data file. The default value

is UTF-8.

ImportDatabase Node

Function

The ImportDatabase node extracts data from database tables and views.

Restriction The corresponding project and process have been added.

The corresponding database has been configured and the database can be connected successfully.

Parameter Description

Parameters

Parameter Description

Data Source

Database Database name. Select a currently available

database from the drop-down list box. Oracle and

DB2 databases are supported.

Schema Table mode in the database. The default value of

the table mode is the name of the created schema.

For example, if database user U1 creates tables T1

and T2 and user U2 creates table T3 in the

database, the options of the table mode in the

database are U1 and U2. When you select a mode,

only the tables of the selected mode are displayed.

Table Database table name. Select a value from the

drop-down list box.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 9 of 124

Parameter Description

Select condition Criteria for filtering extracted data, which is similar

to the where expression in a SQL statement.

NOTE

The criteria does not need to contain the keyword where.

Sample The default value is No.

Indicates whether to extract 2 million records when

the number of records exceeds 2 million. When the

total number of records is less than 2 million, all

records are extracted.

Meta Data

NOTE

After the field name is modified, the data on the following nodes also needs to be modified.

Delimiter

NOTE

The delimiter for separating data stored when a process is executed is set on this tab page. The delimiter cannot be a special character that exists in the stored data.

Field Nodes

A Field node bins, partitions, fills, or filters source data.

1.1.1.1 Type Node

Function

The Type node specifies the data role, direction, and missing value for each field in a data set, and verifies

that field types are valid.

Restriction

The Type node follows a ImportText, ImportFeatureLibrary, ImportFolder, or ImportDatabase node or

Application node.

Parameter Description

Parameters

Parameter Description

Read Value Reads the values of Role and Value Range from the

data audit file.

Clear Clears the values of Role and Value Range.

Role Role type.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 10 of 124

Parameter Description

The options are as follows:

● No: The value type is not specified.

● Range: Specifies a value range, for example, 0 to

100. The range can be an integer, real number, or

date/time range.

● Sign: The value has only two options, for

example, 0 and 1, or y and n.

● Set: The value has multiple options, for example,

high, middle, and low, or type1, type2, and type3.

Value Range Value range. Set this parameter when Role is not set

to No.

● If Role is Range, the following parameters are

required:

– Lower Limit: lower limit of a range

– Upper Limit: upper limit of a range

● If Role is Sign, the following parameters are

required:

– Flag Value 1: use the first value

– Flag Value 2: use the second value

● If Role is Set, set the Set Value parameter. You

can click New to add options.

Default Value Default value used for replacement. When Check is

Modify and data is missing or out of range, the data

is replaced with the parameter value.

Check Checks all values in a field to verify that all values

are correct. Using this method, you can manage the

data sets and reduce the data sizes conveniently.

The options are as follows:

● Close: Not check a field. This value is the default

value.

● Modify: Check all values in a field and correct

incorrect values. If the default value is not set,

values will be modified according to the following

rules:

– For a field of the Set role, the method

changes all unknown values to the first value

in the data set.

– For a field of the Sign role, the method

changes all unknown values to the first value

in the data set.

– For a field of the Range role, the method

changes values greater than the upper limit to

the upper limit, changes value less than the

lower limit to the lower limit, and changes

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 11 of 124

Parameter Description

null values to the middle value in the range.

● Discard: Check all values in a field and delete

incorrect values.

Anonymize Specifies whether to anonymize a field. For example,

sensitive customer information displayed in models

need to be anonymized in actual use to protect

privacy.

The options are as follows:

● Yes: Replace values in a field based on the field

type. The replacement complies with the

following rules:

– For a field of the Range role, the range is

changed to another range to anonymize

sensitive data. The replacement rule is as

follows: Final value = Conversion factor x

Actual value + Offset. The default conversion

factor is 3 and default offset is 9.

– For a field of the Sign or Set role, values in

the field are changed to the following

character string:

– prefix_Sn: prefix is a character string defined

by users. The default value is anon. n is an

integer greater than 0. Therefore, by default,

unique values in a field is changed to values

such as anon_S1 and anon_S2 in sequence.

– For a field of other roles, values in the field

are changed to the following character string:

prefix_S0: prefix is a character string defined

by users. The default value is anon.

● No: Not anonymize a field.

● Define: Customize a field. When the option is

selected, you can customize a value to replace

sensitive data.

Direct Field direction that specifies the role of a field during

the modeling process, for example, an input field or

an output field.

The options are as follows:

● No: ignore the field

● Primary key: primary key field

● Input: self-learning input field (forecast variable

field)

● Output: self-leaning output field or object (field to

be forecasted)

● Two-way: input/output field to be used by the

Apriori node. Other modeling nodes will ignore

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 12 of 124

Parameter Description

the field.

● Partition: field to be partitioned into samples for

training, test, and verification

NOTE

When the value of Direct is Partition, you should choose Sign of Role. The relationship between Default Value and Partition is:

● Flag Value 1:Training Partition

● Flag Value 2:Tset Partition

Bin Node

Function

The Bin node divides the attribute value range of fields of the Range role into segments and assigns a value

to each segment. This reduces the number of attribute values. The Bin node can create a field of the Set role

based on one or more values of range segments. For example, the node can change the customer income

range into a set of income groups or a set of differences from the average income.

Restriction

The node must follow a source node (ImportText, ImportFeatureLibrary or ImportDatabase) and a Type

node.

Parameter Description

Parameters

Parameter Description

Fixed width Fixed binning width.

The maximum and minimum values in a data set

are calculated. Then the binning method is defined

based on the minimum and maximum values and

Bin width.

For example, if the minimum value is 10, the

maximum value is 30, and Fixed width is 10, the

range is binned into [10,20) and [20,30].

Width or Amount This parameter is valid only when Binning Type

is set to Fixed width.

● Bin width: binning width. The default value is

10.

● Number of bins: number of bins. The default

value is 10 and the value cannot exceed 100.

Fixed depth Bins a range at a fixed depth.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 13 of 124

Parameter Description

The number of values in a data set is calculated

and values are sorted in ascending order. Then data

is binned based on the value of Bin depth, that is,

the number of values in a set.

Depth or Amount This parameter is valid only when Binning Type

is set to Fixed depth.

● Bin depth: binning depth. The default value is

10.

● Number of bins: number of bins. The default

value is 10.

Standard Deviation Bins data based on the standard deviation. Values

are compared with the standard deviation and

binned based on the differences.

Offset Bins data based on the average value and

deviation. This parameter is valid only when

Binning Type is set to Standard Deviation.

The options are as follows:

● +/- 1 Standard Deviation. The range is as

follows:

– [-∞,average value - deviation)

– [average value - deviation,average value +

deviation)

– [average value + deviation,+∞]

● +/-2 Standard Deviation. The range is as

follows:

– [-∞,average value - 2 x deviation)

– [average value - 2 x deviation,average

value - deviation)

– [average value - deviation,average value +

deviation)

– [average value + deviation,average value +

2 x deviation)

– [average value +2 x deviation,+∞]

● +/-3 Standard Deviation. The range is as

follows:

– [-∞, average value - 3 x deviation)

– [average value - 3 x deviation,average

value - 2 x deviation)

– [average value - 2 x deviation,average

value - deviation)

– [average value - deviation,average value +

deviation)

– [average value + deviation,average value +

2 x deviation)

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 14 of 124

Parameter Description

– [average value + 2 x deviation,average

value + 3 x deviation)

– [average value +3 x deviation,+∞]

Frequency Bins data based on the frequency of each value.

Frequency Range List Enter a list of integers separated with commas (,),

for example, a1,a2,a3,...an.an cannot exceed

999999999, and n must be an integer less than 10.

Example: 2,4,5

Gain Ratio of the difference between the type

distribution of input attributes (for example, B) and

the corresponding type distribution of output

attributes (for example, A) to B. As a result, the

gain expression is as follows:

p=(A-B)/B

When the method is used, you need to configure a

field whose Role is Set or Sign and Direct is

Output.

Binning Number Beforehand Number of bins for pre-binning specified fields.

Irregular Binning Threshold If the value of Gain is less than the value of

Irregular Binning Threshold, records that exceed

the threshold are put into the irregular bin. The

default value is 0.2.

Partition Node

Function

The Partition node generates partition fields. It partitions data into subsets and samples for the training and

test phases in the modeling process. During the modeling process, a sample is used to generate a model and

another sample is used to test the model. In this way, the system can check the forecast accuracy deviation

of the model on large size data sets similar to the data samples.

The Partition node generates fields of the Sign role. Only fields of the Sign role can be defined as partition

fields on the Type node.

Restriction

The node must follow a source (ImportText, ImportFeatureLibrary or ImportDatabase) node and a Type

node.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 15 of 124

Parameter Description

Parameters

Parameter Description

Partition Type ● Random: separate data to training partition and

testing partition according to the rate you

entered.

● Stratified: separate the data to some floors, then

separate the data from every floor to training

partition and testing partition according to the

rate you entered.

● Condition: separate data using expressions.

Stratified Field Field based on which data is separated to multiple

floors in stratified partitioning.

Partition Field Partition field name. The value must be unique in

the data sets.

Training Data Rate Percentage of a training data set to the input data

set. The default value is 0.5

NOTE

The sum of Training Data Rate and Test Data Rate cannot exceed 1.

If the sum is less than 1, the system will discard records that are not contained in the two sets. For example, if a user has 10 million records, and Training Data Rate and Test Data Rate are 0.05 and 0.1 respectively, after the partition node is executed, about 0.5 million training records and 1 million test records are generated and other records are discarded.

Training Data Partition Condition Click the text box. The dialog box is displayed for

you to configure the partition condition for training

data.

Training Data Flag Flag of a training data set. The default value is 1,

unchangeable.

Test Data Rate Percentage of a test data set to the input data set.

The default value is 0.5.

Test Data Partition Condition Click the text box. The dialog box is displayed for

you to configure the partition condition for test

data.

NOTE

The partition condition for training data and that for test data cannot be both empty. If one of them is empty, the two partition conditions are complementary by default.

If a data record meets both the partition conditions for training data and test data, it will be used as training data.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 16 of 124

Parameter Description

Test Data Flag Flag of a test data set. The default value is 2,

unchangeable.

Fill Node

Function

The Fill node replaces the field values in the input data set.

Restriction

The node must follow a source node (ImportText, ImportFeatureLibrary or ImportDatabase) and a Type

node.

Parameter Description

Parameters

Parameter Description

Value Range Value range. By default, the setting on the Type node is used. You need to change the

parameter value after Conversion setting is complete.

Conversion

Setting

Conversion type. Click Configure. Then the Conversion Setting Window page is

displayed.

Conversion types include missing, exception, data normalization, function, expression,

and virtual variable. The conversion types that can be used for data of the set type

include missing, expression, and virtual variable. The conversion types that can be used

for data of the value type include missing, exception, data normalization, function, and

expression. Multiple conversion types can be used for a same attribute at the same time.

The system will execute the conversion in the sequence that the types are configured.

0 describes the parameters on the Conversion Setting Window page.

Conversion setting page

Parameter Description

Missing

Conversion Mode The options are as follows:

● Fill with the cumulative rate: Select this option for fields of the range type.

● Fill with the modal number: Select this option for fields of the set type.

Exception

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 17 of 124

Parameter Description

Check Method The options are as follows:

● Standard deviation method

● Five-number summary

● Walsh test(If Walsh test is used,the number of records must be greater than

5500)

Conversion Mode The options are as follows:

● Discard

● Fill with the average value

● Fill with the maximum or minimum value

Data Normalization

Conversion Mode The options are as follows:

● Extremum method

● Standard deviation method

● Percentile method

Function

Base Parameter corresponding to the function. This parameter is dimmed if the parameter

corresponding to the function does not exist.

Virtual Variable

Field Name Source fields to be selected, based on which new fields will be generated.

Select a value from the drop-down list box.

NOTE

You can select multiple fields but cannot select all.

Virtual Variable

Name

Name of the field generated after conversion.

Expression

Condition

Expression

Condition expression.

Enter the expression in the text box on the lower left corner of the page.

NOTE

To enable the field and function association mode, click the text box and press Alt+/.

if statement must have a corresponding else statement.

Else may not be displayed because of inappropriate screen resolution. In this case, set the resolution to 1280 x 960.

You can use the following resources to configure the expression:

– Available fields

– System functions

– User-defined functions

When using a function to configure the expression, you can select only the functions whose return values are of the boolean type.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 18 of 124

Filter Node

Function

The Filter node filters fields based on the correlation information of the analysis field and the forecast field

or filters specified fields.

Correlation filtering indicates that fields are filtered based on Error Decrease Rate in the analysis field

and the forecast field. The system retains fields whose Error Decrease Rate is greater than a specified

threshold according to the configured Max. Number of Retained Fields. This method is used when a large

amount of data needs to be filtered.

Users can also filter fields manually. The filtering effect of this method is similar to that of Filter on the

TextImport node.

Restriction

The node must follow a source(ImportText, ImportFeatureLibrary or ImportDatabase) node.

To filter fields by correlation, you need to use Correlate to analyze the correlation of the source data.

Parameter Description

Parameters

Parameter Description

Threshold Value threshold of Error Decrease Rate. This

parameter is valid only when a correlation analysis

file is selected by clicking the Select Correlation

Analysis File button.

When you click Filter, the system will filter out all

fields whose Error Decrease Rate is less than the

threshold.

Max.Retained Fields Maximum number of fields that are retained after

correlation filtering. This parameter is valid only

when a correlation analysis file is selected by

clicking the Select Correlation Analysis File

button. When you click Filter, the system will sort

fields by Error Decrease Rate in descending

order and retain the first Max.Retained Fields

fields.

Field Field to search.

Filtering Field Filtering field.

If you use the correlation filtering method, the

system automatically selects filtering fields when

you press Filter. To deselect all filtering fields,

click Reset.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 19 of 124

Record Nodes

A Recode node samples or selects source data.

1.1.1.1 Sampling Node

Function

The Sampling node can extract samples from records for analysis. The Sampling node has the following

advantages:

You can evaluate models based on sample analysis results to improve model performance. Models

improved based on sample analysis results can generate accurate forecast results. In addition,

improved models can provide more methods that can further improve the models.

The Sampling node can extract records that share specified features, for example, all items in a

shopping cart or all attributes of related objects.

The node can randomly extract samples in a specified unit or with a specified attribute and check them.

In this way, service quality is verified, fraud is prevented, and security is ensured.

NOTE If you only need to partition data into a training data set and a test data set, use the Partition node.

The Sampling node supports the following sampling modes:

Random: The Sampling node extracts data at a specified ratio. For example, if a user has 10 million

records and the sampling ratio is 0.5, the node will extract 5 million records.

Equidistant: The Sampling node extracts a record from every N records. For example, if a user has 10

thousand records, N is 10, and the maximum sample size is 100, the Sampling node will extract 100

records.

Cluster: The Sampling node extracts records from a group with a specified field at a specified ratio.

For example, if the sampling field is school and the sampling ratio is 0.5, the Sampling node will

extract 50% of the records from the school group.

You can set multiple sampling fields, and the node will extract records across the specified groups.

For example, field A and field B are specified, and the sampling ratio is 0.5. Field A is of the Set role,

and the options are a and b. Field B is a string character, and the options are c and d. Then the

Sampling node will extract all data from two of the following sets:

− a,c

− b,c

− a,d

− b,d

Stratified: The Sampling node extracts records at a specified ratio from each sampling group specified

by a sampling field.

For example, field A and field B are specified, and the extraction ratio is 0.5. Field A is of the Set role,

and the options are a and b. Field B is a string character, and the options are c and d. Then the

Sampling node will extract 50% records from each of the following sets:

− a,c

− b,c

− a,d

− b,d

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 20 of 124

Balance: Balance sampling balances discrete fields. The Sampling node returns the extracted records

to the input data sets for next sampling, so that the value types in the final extracted records are

balanced.

Restriction

The node must follow a source(ImportText, ImportFeatureLibrary or ImportDatabase) node and a Type

node.

Parameter Description

Parameters

Parameter Description

Random

Sampling Rate Ratio of randomly selected records to total records.

Equidistant

Value of N Extracts a record from every N records. The

default value is 2.

Cluster

Sampling Field Sampling field. Enter one or more fields of the

sign or set type. A maximum of three sampling

fields are allowed.

Stratified

Sampling Field Sampling field. Enter one or more fields of the

sign or set type. A maximum of three sampling

fields are allowed.

Fixed Rate Fixed sampling ratio. The default value is 0.5,

indicating that the Sampling node will extract 50%

records from each sampling group.

User-defined Rate Customized sampling ratio. You can set the

parameter for each sampling group.

For example, you can set the parameter to 0.4 for

sampling group A and set the parameter to 0.8 for

sampling group B.

NOTE

The parameter is valid only when the Sign and Set roles have been configured on the Type node and the number of values in a set does not exceed 3.

Balance

Sampling Field Select a field of the set or sign type as the

sampling field.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 21 of 124

Select Node

Function

The Select node selects records with a specified feature, for example, Salary>2000, from data streams.

Restriction

The node must follow a source(ImportText, ImportFeatureLibrary or ImportDatabase) node and a Type

node.

Parameter Description

Parameters

Parameter Description

Edit Expression

NOTE To enable the field and function association mode, click the text box and press Alt+/.

Verify Authenticates the selected expression type and entered expression.

Click the check button.

SelectFeature Node

Function

The SelectFeature node filters out invalid or indistinct attributes based on filter criteria.

Restriction The node must follow a source(ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

The SelectFeature node cannot be configured in a process that contains the Bin node. The

SelectFeature node can only be used for Naive Bayes models.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 22 of 124

Parameter Description

Table 1 Parameters

Parameter Description

Max. Missing Value Rate Maximum field loss ratio. The FeatureSelection

node filters out attributes whose field loss ratio is

higher than the value of Max. Missing Value

Rate.

For example, if a user has 100 thousand records,

the number of lost records of the X attribute is 40

thousand, and Max. Missing Value Rate is 0.3,

the FeatureSelection node will filter out the

attribute.

Max. Repetition Rate Maximum field repetition ratio. The

FeatureSelection node filters out attributes whose

field repetition ratio is higher than the value of

Max. Repetition Rate.

For example, if a user has 100 thousand records,

the number of fields whose value is 1 of the X

attribute is 50 thousand, and Max. Repetition

Rate is 0.3, the FeatureSelection node will filter

out the attribute.

Max. Category Rate Maximum different field ratio. The

FeatureSelection filters out attributes whose

different field ratio is higher the value of Max.

Category Rate.

For example, if a user has 100 thousand records,

all records of the X attribute are unique, and Max.

Category Rate is 0.9, the FeatureSelection node

will filter out the attribute.

Field Confidence Attribute confidence. The FeatureSelection node

calculates the confidence of input and output

attributes based on the chi-square test and filters

out attributes whose confidence is lower than Field

Confidence.

Bin Count The FeatureSelection node filters attributes based

on the field filtering condition, and bins the

remaining attributes of the Range role at a fixed

binning depth.

Modeling Nodes

Models are composed of rules, expressions, or equations. You can use models to forecast output results

based on input values or variables.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 23 of 124

1.1.1.1 NaiveBayes Node

Function

NaiveBayes classifier is a classification method in statistics. The NaiveBayes node forecasts the class

membership probabilities, for example, the probability that a sample belongs to a specified class.

The NaiveBayes node can build models to forecast event probability by analyzing event attributes based on

the system's cognition towards reality and obtained records.

Restriction The node must follow a source (ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

A data mining process containing the NaiveBayes node must meet the following conditions:

− The node must contain a minimum of one input field and only one output field.

− For input fields, Role must be Sign, Set, or Range.

− For the output field, Role must be Sign or Set.

The computing framework (Hadoop or Spark) on which the NaiveBayes node runs can be configured

in the ${SmartMiner_HOME}/conf/smartminer.spark.nodes file. Parameters in Precision

Improving Method vary depending on the selected computing framework. For details about the

parameters, see Parameter Description.

An example of the configuration file is as follows:

NaiveBeyes=hadoop //The node runs on the Hadoop framework.

NaiveBeyes=spark //The node runs on the Spark framework.

Model Input Example

USER_ID,SERV_NUMBER,STATIS_DATE,AREA_CODE, ...,OUT

001,18936897385,20100606,0371,...,yes

002,18936897386,20100607,0371,...,no

Fields are separated by commas (,).

On the Type node:

USER_ID: Primary key

Modeling feature field: Input

Forecast field: Output

The role and value range of each field need to be configured.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 24 of 124

Parameter Description

Parameters

Parameter Description

Model File Model file name. Model files are stored in the Project

name/Model directory.

By default, MOD files are exported from the

${SmartMiner_HOME}/smartminer/Projects/test/Model

directory.

NOTE

In the preceding directory, test indicates the name of the project where the process is located.

Precision Improving Method Specifies whether to use the precision improving method.

● Empty: The precision improving method will not be used.

● Bagging: The system randomly extracts 50% data for three

times and generates three temporary models for calculation.

● Boosting: The system classifies data and extracts data based on

the weight of each class to generate model A. Then the system

applies model A, analyzes the error rate based on the

application result, and extracts data again to generate model B.

During the extraction, more data of the class whose error rate

is high is extracted. Model C is generated and applied the same

way as model B. Then the system converts the error rate of the

three models into weights, calculates the weights and classes,

and generates the class whose weight is the highest as the

result.

NOTE

The Bagging and Boosting options are valid only when the NaiveBayes node runs on the Hadoop.

Use Partition Specifies whether to use only the training data set to build models

if the Partition node is configured.

● Yes: Use only data in the training data set.

● No: Use data from both the training and test data sets.

Ignore Missing Value Specifies whether to ignore missing values.

● Yes: Ignore records containing missing values.

● No: Not ignore records containing missing values. The system

skips only missing values in records. Normal values are

calculated.

Select Features Specifies whether to filter features.

● Yes: Filter out feature fields whose confidence is lower than

the value of Variable Confidence.

● No: Not filter feature fields.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 25 of 124

Parameter Description

Variable Confidence Confidence value. The fields whose confidence is lower than this

parameter value will be filtered out. The default value is 0.95.

NOTE

When the value of Select Features is yes, you should configure this parameter.

Visualization Input and output attributes of the model

Click the NaiveBayes model file. The tree structure of the NaiveBayes model is displayed, including

the input and output attributes of the model, as shown in 0

NaiveBayes model file structure

NOTE If Bagging and Boosting are used, the SmartMiner will generate multiple models. The model structure tree displays multiple models, among which the root node is ModelSet, the Boosting model weight is calculated based on the error rate, and the default Bagging model weight is 1.

Input attribute node information

− When the inputs attribute node of the set type is clicked, the statistics table and probability table

are displayed to the right of the structure tree, as shown in 0.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 26 of 124

Set type

− When the inputs attribute node of the range type is clicked, the average table and deviation table

are displayed to the right of the structure tree, as shown in 0.

Range type

DecisionTree Node

Function

The DecisionTree node can develop a classification system. Using this system, you can forecast results or

classify records based on predefined decision policies.

Restriction The node must follow a source (ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

A data mining process containing the DecisionTree node must meet the following conditions:

− The process must contain one input field and one output field.

− For the input field, Role must be Sign, Set, or Range. The number of set value types cannot exceed

10.

− For the output field, Role must be Sign or Set.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 27 of 124

Model Input Example

USER_ID,SERV_NUMBER,STATIS_DATE,AREA_CODE, ...,OUT

001,18936897385,20100606,0371,...,yes

002,18936897386,20100607,0371,...,no

Fields are separated by commas (,).

On the Type node:

ID: Primary Key

Modeling feature field: Input

Forecast field: Output

The role and value range of each field need to be configured.

Parameter Description

Parameters

Parameter Description

Model File Model file name. Model files are stored in the Project

name/Model directory.

By default, MOD files are exported from the

${SmartMiner_HOME}/smartminer/Projects/test/Model

directory.

NOTE

In the preceding directory, test indicates the name of the project where the process is located.

Precision Improving Method Specifies whether to use the precision improving method.

● Empty: The precision improving method will not be used.

● Bagging: The system randomly extracts 50% data for three

times and generates three temporary models for calculation.

● Boosting: The system classifies data and extracts data based on

the weight of each class to generate model A. Then the system

applies model A, analyzes the error rate based on the

application result, and extracts data again to generate model B.

During the extraction, more data of the class whose error rate

is high is extracted. Model C is generated and applied the same

way as model B. Then the system converts the error rate of the

three models into weights, calculates the weights and classes,

and generates the class whose weight is the highest as the

result.

Select Attribute Method NOTE

If this parameter is set to Gini or F-Score and the role of the input fields is Set, the number of set value types cannot exceed 10. If input fields of the set type must be used, a data processing node is required to process the fields first. For example, you can use a Filler node to combine data or convert a set to multiple fields.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 28 of 124

Parameter Description

Use Partition Specifies whether to use only the training data set to build models

if the Partition node is configured.

● Yes: Use only data in the training data set.

● No: Use data from both the training and test data sets.

Ignore Missing Value Specifies whether to ignore missing values.

● Yes: Ignore records containing missing values.

● No: Not ignore records containing missing values.

Use Pruned Branch Specifies whether to enable the pruned branch function. When the

function is enabled, the system does not analyze attributes that

cannot affect decision results.

● Yes: Enable the pruned branch function. Set the parameter to

Yes when the training data set contains abnormal data or when

data amount in the training data set is too small to generate

practical functions.

● No: Disable the pruned branch function.

Prune Confidence Factor Prune confidence factor for pruning a field. The value must be

greater than 0 and less than 1. The system ignores fields whose

confidence is lower than Prune Confidence Factor.

NOTE

This parameter is valid only when Use Pruned Branch is set to Yes.

Min. Leaf Nodes Minimum number of records on a leaf node. Set the parameter to a

positive integer. The default value is 2. The system ignores

attributes whose number of fields is less than Min. Leaf Nodes.

NOTE

It is recommended that you use the value obtained by dividing the number of records in the raining data set by 2L. L indicates the number of input fields in the training data set.

Visualization Model Structure

Click the DecisionTree model file. The system displays information about the model, in which, the

model structure is displayed on the left.

NOTE When viewing the DecisionTree model for the first time, the model structure tree on the left displays a maximum of

10,000 nodes. You can click the structure tree to display the hidden nodes.

If Bagging and Boosting are used, the SmartMiner will generate multiple models. The structure tree displays multiple models, among which the root node is ModelSet, the Boosting model weight is calculated based on the error rate, and the default Bagging model weight is 1.

Node information

The information about the node is displayed on the right, as shown in 0.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 29 of 124

DecisionTree model file information

Display Tree Diagram

Click DecisionTree_normal next to Display Tree Diagram in 0. The model file structure is displayed

in a tree diagram, as show in 0.

Figure 2 Displaying the model file structure in a tree diagram

NOTE When the attribute is empty, the category cannot be specified. As a result, the current category ratio is used. For

example, the sex attribute has the male and female options, and assume that, among the football lovers, 60 are male and 40 are female. Accordingly, among 4 football lovers whose gender is unknown, 2.4 of them are male and 1.6 of them are female.

If the flowchart involves a large number of components and you need to display the nodes more clearly to improve operation experience, use the full screen function by clicking the Full Screen button above the process.

Only Firefox 10 and later versions support the function. Internet Explorer does not support the function.

Models cannot be displayed in the tree diagram mode if containing more than 700 nodes.

Display Path

Click DecisionTree_normal next to Display Path in 0. The model file structure is displayed in paths.

Click a path, and the details about the path are displayed, as shown in 0.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 30 of 124

Figure 3 Displaying the model file structure in path

Logistics Node

Function

The Logistics node determines the cause-effect relationships between variables, sets up regression models,

and checks the correlations between symptoms and the correlation directions and levels.

Restriction The node must follow a source (ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

A data mining process containing the Logistics node must meet the following conditions:

The process must contain only one output field and one or more input fields.

If the input field is of the character string type, the field must be of Sign, Range or Set role. If the

input field is of the numeral type, you do not need to set the Role parameter.

The output fields must be of Sign role.

The computing framework (Hadoop or Spark) on which the Logistics node runs can be configured in

the ${SmartMiner_HOME}/conf/smartminer.spark.nodes file. Parameters to be configured vary

depending on the selected computing framework. For details about the parameters, see Parameter

Description.

An example of the configuration file is as follows:

Logistics=hadoop //The node runs on the Hadoop framework.

Logistics=spark //The node runs on the Spark framework.

Model Input Example

USER_ID,SERV_NUMBER,STATIS_DATE,AREA_CODE, ...,OUT

001,18936897385,20100606,0371,...,yes

002,18936897386,20100607,0371,...,no

Fields are separated by commas (,).

On the Type node:

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 31 of 124

ID: Primary Key

Modeling feature field: Input

Forecast field: Output

The role and value range of each field need to be configured. For output fields, Role must be set to

Sign.

Parameter Description

Parameters

Parameter Description

Model File Model file name. Model files are stored in the

Project name/Model directory.

Precision Improving Method Specifies whether to use the precision improving

method. The options are as follows:

● Empty: The precision improving method will

not be used.

● Bagging: The system randomly extracts 50%

data for three times and generates three

temporary models for calculation.

● Boosting: The system classifies data and

extracts data based on the weight of each class

to generate model A. Then the system applies

model A, analyzes the error rate based on the

application result, and extracts data again to

generate model B. During the extraction, more

data of the class whose error rate is high is

extracted. Model C is generated and applied the

same way as model B. Then the system

converts the error rate of the three models into

weights, calculates the weights and classes, and

generates the class whose weight is the highest

as the result.

NOTE

The Bagging and Boosting options are valid only when the Logistics node runs on the Hadoop.

Use Partition Specifies whether to use only the training data set

to build models if the Partition node is configured.

● Yes: Use only data in the training data set.

● No: Use data from both the training and test

data sets.

Max. Iterations Maximum number of iteration times during the

computing process.

Iteration End Threshold The parameter is valid only when the Logistics

node runs on the Hadoop.

Specifies the threshold for stopping iteration. An

iteration process ends if the coefficient change of

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 32 of 124

Parameter Description

the iteration algorithm is less than the value of this

parameter twice.

Select Features The parameter is valid only when the Logistics

node runs on the Hadoop.

Specifies whether to select features. You can use

the parameter to define feature selection criteria

based on the probability associated with fields.

Include Significance Threshold The parameter is valid only when the Logistics

node runs on the Hadoop.

This parameter is valid only when Select Features

is set to Select Features. The default value is 0.05.

In the model iteration process, the ¦Α value of the

chi-squared distribution is used to display the

associated confidence of the statistical probability.

When ¦Α is less than the value of Include

Significance Threshold, the system adds the field

to the model.

Exclude Significance Threshold The parameter is valid only when the Logistics

node runs on the Hadoop.

This parameter is valid only when Select Features

is set to Select Features. The default value is 0.1.

In the model iteration process, the ¦Α value of the

chi-squared distribution is used to display the

associated confidence of the statistical probability.

When ¦Α is greater than the value of Exclude

Significance Threshold, the system deletes the

field.

NOTE

The value of Include Significance Threshold must be less than the value of Exclude Significance Threshold.

Step Size The parameter is valid only when the Logistics

node runs on the Spark.

Specifies the coefficient weight change of each

iteration.

Regularization The parameter is valid only when the Logistics

node runs on the Spark.

Regularization refers to a process of introducing

additional information to solve an ill-posed

problem or to prevent overfitting. In linear algebra,

ill-posed problems are defined by a group of linear

algebraic equations and the linear algebraic

equations come from ill-posed inverse problems

that have large condition numbers. Large condition

numbers will seriously affect the computing result

due to rounding errors or other errors.

Regularization parameters are used to define

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 33 of 124

Parameter Description

parameter conversion of linear equations.

Mini Batch Fraction The parameter is valid only when the Logistics

node runs on the Spark.

Defines an iteration factor (proportion of samples

used for computing for each iteration).

Include Constant Specifies whether the model contains constants.

Base Category Model forecast field. The value is the same as that

of the output field on the Type node.

This parameter is valid only when Include

Constant is set to Yes.

Visualization

Click the Logistics model file. The system displays the model information, as shown in 0. The structure tree

is displayed on the left, indicating the attributes and values. The significance of the attributes is displayed

on the right.

Logistics model file information

NOTE If Bagging and Boosting are used, the SmartMiner will generate multiple models. The model structure tree displays multiple models, among which the root node is ModelSet, the Boosting model weight is calculated based on the error rate, and the default Bagging model weight is 1.

Kmeans Node

Function

The Kmeans node groups data sets into different cluster centers (or clusters). This method defines a fixed

number of clusters, classifies records to clusters in iteration mode, and adjusts the cluster center until the

model can no longer be optimized.

The Kmeans node is a non-monitoring learning mechanism. It finds hidden patterns behind input data sets

instead of forecasts results.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 34 of 124

Restriction The node must follow a source (ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

The data mining process in which the Kmeans node is configured must contain a minimum of one

input field. You need to set the role and range for the input fields.

Model Input Example

id,age,sex,children

ID12101,48,FEMALE,1

ID12102,40,MALE,3

ID12103,51,FEMALE,0

ID12104,23,FEMALE,3

ID12105,57,FEMALE,0

Fields are separated by commas (,).

On the Type node, Direction of ID is set to Primary Key, and Direction for other fields is set to Input.

Parameter Description

Parameters

Parameter Description

Model File Model file name. Model files are stored in the Project

name/Model directory.

By default, MOD files are exported from the

${SmartMiner_HOME}/smartminer/Projects/test/Model

directory.

NOTE

In the preceding directory, test indicates the name of the project where the process is located.

Max.Iterations Maximum number of iteration times for the Kmeans modeling.

Use Partition Specifies whether to use only the training data set to build models

if the Partition node is configured.

● Yes: Use only data in the training data set.

● No: Use data from all data sets.

Visualization View the overall table.

Click the Kmeans model file. The overall table is displayed, as shown in 0. The overall table shows all

the clusters and their input fields. The value of a field of the Range type is the average value of this

field in the cluster to which it belongs. For a discrete field, only three values of the highest weight are

displayed, by weight in descending order. A maximum of 10 fields are displayed and sorted by

importance in descending order. If fields have the same importance, they are sorted by index ID in

ascending order.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 35 of 124

Figure 1 Overall table of the model

View the cluster table.

Click a cluster node in the navigation tree on the left. The cluster table is displayed, as shown in 0. The

cluster table displays the importance and role of all fields in the cluster, as well as the link to the field

distribution map and histogram.

Cluster table

View the field distribution map/histogram.

Click next to a field in 0, or click a field under a cluster node in the navigation tree on the left.

The field distribution map/histogram is displayed, as shown in 0. The field distribution map/histogram

shows how the fields are distributed in the cluster. The system uses the distribution map for fields of

the Range type, and the histogram for discrete fields. In the distribution map for fields of Range type,

the x-axis indicates the median value of the binning, and the y-axis indicates the weight of the field; in

the histogram for discrete fields, the x-axis indicates the value of the field, and the y-axis indicates the

weight of the field.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 36 of 124

Attribute distribution map

EM Node

Function

The EM node groups data sets into different cluster centers (or clusters). The EM node assumes that the

sample complies with multidimensional normal distribution and analyzes hidden classifications of the

sample using the expectation maximization method to implement automatic clustering. This method defines

a fixed number of clusters, calculates the probability that each record belongs to a cluster, and updates the

probability iteratively until the probability change is less than the preset Iteration End Threshold or the

Maximum number of iteration times is achieved.

Restriction The node must follow a source (ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

The data mining process that contains an EM node must comply with the following requirements: the

process contains at least one input field, the input data must be of the numeric type (data of the string

type is not supported), and the role and value range of the input data must be set.

Model Input Example

5.1,3.5,1.4,0.2

4.9,3.0,1.4,0.2

4.7,3.2,1.3,0.2

4.6,3.1,1.5,0.2

5.0,3.6,1.4,0.2

5.4,3.9,1.7,0.4

4.6,3.4,1.4,0.3

5.0,3.4,1.5,0.2

4.4,2.9,1.4,0.2

4.9,3.1,1.5,0.1

5.4,3.7,1.5,0.2

4.8,3.4,1.6,0.2

4.8,3.0,1.4,0.1

4.3,3.0,1.1,0.1

5.8,4.0,1.2,0.2

......

The input fields are a set of IRIS data, indicating the calyx length, calyx width, petal length, and petal width

of the flower-de-luce respectively. The fields are separated by the commas (,).

On the Type node, Direction of all fields is set to Input.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 37 of 124

Parameter Description

Parameters

Parameter Description

Model File Model file name. Model files are stored in the Project

name/Model directory.

By default, MOD files are exported from the

${SmartMiner_HOME}/smartminer/Projects/test/Model

directory.

NOTE

In the preceding directory, test indicates the name of the project where the process is located.

Cluster Name Prefix Name prefix of a clustering field. This field is displayed in the

output result as a new field.

Cluster Count Defines the number of clustering results in the output result.

For example, if this parameter is set to 5, Cluster_0-Cluster_4

will be generated in the clustering result.

Iteration Times Maximum number of iteration times for the EM modeling

algorithm. The model training ends when the number of iteration

times reaches the value of Iteration Times.

The value ranges from 1 to 100.

Default value: 20

Iteration End Threshold Iteration end threshold. If the maximum likelihood estimate

between two iterations is less than the value of this parameter, the

iteration ends.

If the parameter value is between 1.0E-1 and 1.0E-5, the

clustering calculation result precision increases sequentially.

Both Iteration End Threshold and Iteration Times can be used

to end iterations. If the input data is multidimensional or high

clustering precision is required, you are advised to increase the

value of Iteration Times.

Use Partition Specifies whether to use only the training data set to build models

if the Partition node is configured.

The options are as follows:

● Yes: Use only data in the training data set.

● No: Use data from all data sets.

Default value: No

Retain Only Elements on Diagonal

Line

Specifies whether to retain only the elements on the diagonal line

in the covariance matrix calculated in the iteration process.

The options are as follows:

● Yes: Retain only the elements on the diagonal line.

Convergence is fast.

● No: Retain all elements. Convergence is slow.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 38 of 124

Parameter Description

Default value: Yes

Output Result Example

Model Input Example shows the input text, and 0 describes the process configuration.

EM node process

0 shows the configuration of the Segment node.

StayPointAnalysis node configuration

The following shows the output of the StayPointAnalysis node. The fifth field cluster_n indicates the

clustering result.

5.1,3.5,1.4,0.2,cluster_0

4.9,3.0,1.4,0.2,cluster_0

4.7,3.2,1.3,0.2,cluster_0

4.6,3.1,1.5,0.2,cluster_0

5.0,3.6,1.4,0.2,cluster_0

5.4,3.9,1.7,0.4,cluster_0

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 39 of 124

4.6,3.4,1.4,0.3,cluster_2

5.0,3.4,1.5,0.2,cluster_0

4.4,2.9,1.4,0.2,cluster_2

4.9,3.1,1.5,0.1,cluster_0

5.4,3.7,1.5,0.2,cluster_0

4.8,3.4,1.6,0.2,cluster_0

4.8,3.0,1.4,0.1,cluster_0

4.3,3.0,1.1,0.1,cluster_2

5.8,4.0,1.2,0.2,cluster_0

......

Apriori Node

Function

The Apriori node analyzes and mines data associations to obtain valuable information for the decision

process.

Restriction The node must follow a source (ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

If the data format is of sparse matrix on the Apriori node, the node must comply with the following

rules:

− In any of the following scenarios:

At least one input field and one outfield

At least one input field and one bidirectional field

At least one output field and one bidirectional field

All bidirectional fields

− Input and output fields are of the integer type.

− Input and output fields are of the Sign role, and the value options are 0 and 1.

If the data format is Key-Value pairs on the Apriori node, the node must comply with the following

rules:

One key field and one bidirectional field in the condition of model node, one key field and one input

field or bidirectional field in Apply Node.

Model Input Example 1 (Sparse Matrix)

cardid,fruitveg,freshmeat,dairy,cannedveg,cannedmeat,frozenmeal,beer,wine,softdrink,fish,confe

ctionery

39808,0,1,1,0,0,0,0,0,0,0,1

67362,0,1,0,0,0,0,0,0,0,0,1

10872,0,0,0,1,0,1,1,0,0,1,0

Fields are separated by commas (,).

On the Type node:

The Direction parameter for cardid is set to None, and the Direction for other fields is set to

Two-way.

The role and value range of each field need to be configured.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 40 of 124

Model Input Example 1 (Key-Value Pair)

cardid,goods

39808,freshmeat

39808,dairy

39808,confectionery

67362,freshmeat

67362,confectionery

10872,cannedveg

10872,frozenmeal

10872,beer

10872,fish

28935,fruitveg

28935,frozenmeal

41792,fruitveg

41792,fish

Fields are separated by commas (,).

On the Type node:

cardid: Primary Key

goods: Two-way

Parameter Description

Parameters

Parameter Description

Model File Model file name. Model files are stored in the Project

name/Model directory.

By default, MOD files are exported from the

${SmartMiner_HOME}/smartminer/Projects/test/Model

directory.

NOTE

In the preceding directory, test indicates the name of the project where the process is located.

Use Partition Specifies whether to use only the training data set to build models

if the Partition node is configured.

● Yes: Use only data in the training data set.

● No: Use data from both the training and test data sets.

Min. Support Support degree for retaining a rule in the rule set. Support degree

indicates the percentage of records whose conditions are true in

the training data set. If the rule you have obtained is applicable to

data subsets of a small size, increase the value of the parameter.

Min. Confidence Minimum confidence of records forecasted by a rule. Confidence

indicates the percentage of true results forecasted by a rule to the

total forecast results. The SmartMiner discards rules whose

confidence is lower than Min. Rule Confidence. If you have

obtained too may rules, increase the value of the parameter. If you

have obtained few rules, decrease the value of the parameter.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 41 of 124

Parameter Description

Max. Antecedents Maximum input records of a rule. You can use the parameter

along with efficient index modes to reduce the search scope based

on the information theory.

Visualization Model Details

Click the Apriori model file. The information about the model is displayed on the right, as shown in 0.

Apriori model file

To generate a new model, select check boxes of required rules and click Generate Model.

Apriori Model Rule Picture

Click a rearsitem parameter to open the Apriori Model Rule Picture view. In 0, the wine rearsitem

parameter is clicked to display the Apriori Model Rule Picture view.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 42 of 124

Apriori Model Rule Picture

Click or right-click, and hold down and move the mouse to form a rectangle that can cross the selected

connection line. Then the correlation information between the two parameters is displayed, as shown

in 0.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 43 of 124

Correlation information

TimeSeries Node

Function

The TimeSeries node finds rules in sequence data, that is, a trend that the data changes over time to forecast

the future value.

The time series modeling mode assumes that history repeats itself. Therefore, decisions applicable to future

events can be made by analyzing historical records. For example, to forecast the sales volume of next year,

you can use the SmartMiner to find the trend that the sales volume changes over time by analyzing the sales

volume of the past few years. A time sequence is a set of records obtained at scheduled times.

Restriction The node must follow a source (ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

A data mining process containing the TimeSeries node must meet the following conditions:

− The process must contain only one input field and only one output field, and the fields must be of

the value type.

− The input field must be arithmetic series.

− The process cannot have missing value in output or input field.

Model Input Example

YEAR,GDP,AVR_GDP,POPULATION,AVR_WAGE,CS_INDEX

1990,18547.9,1634,212548,2140,103.1

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 44 of 124

1991,21617.9,1879,201548,2340,103.4

1992,26638.1,2287,198452,2711,106.4

1993,34634.4,2939,178456,3371,114.7

1994,46759.4,3923,165874,4538,124.1

1995,58478.1,4854,154846,5500,117.1

1996,67884.6,5576,141548,6210,108.3

1997,74462.6,6054,130254,6470,102.8

Fields are separated by commas (,).

On the Type node:

YEAR: Input

GDP: Output

Other fields: None

Parameter Description

Parameters

Parameter Description

Model File Model file name. Model files are stored in the Project name/Model

directory.

By default, MOD files are exported from the

${SmartMiner_HOME}/smartminer/Projects/test/Model directory.

NOTE

In the preceding directory, test indicates the name of the project where the process is located.

Variable Confidence Confidence interval of the autocorrelation between the forecasted

value and residual. The default value is 95.

ACF and PACF Delay Time delay in the autocorrelation and partial autocorrelation

coefficient. This parameter is used to evaluate models. The default

value is 24.

Test Outlier Specifies whether to detect outliers. The options are as follows:

● Yes: Detect outliers automatically based on data types.

● No: Not detect or build models for outliers. The value is the

default value.

MinHash Node

Function

The MinHash node analyzes the similarity between two data sets quickly.

Restriction The node must follow a source (ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 45 of 124

The data mining process in which the MinHash node is configured must contain only one primary key

field and only one input field.

Model Input Example

phone_num,color_ring

13978965412,A

13978965412,B

13978965412,C

13978965412,D

13978965412,E

13945632178,A

13945632178,B

13945632178,D

Fields are separated by commas (,).

On the Type node:

phone_num: Primary Key

color_ring: Input

Parameter Description

Parameters

Parameter Description

Min. Items Minimum number of data entries in a data primary

key or field. The default value is 5, indicating that

a primary key or field contains a minimum of five

data entries.

Min. Cluster Elements Minimum number of data subsets on a cluster. The

default value is 5, indicating that a cluster contains

a minimum of 5 subsets.

Linear Node

Function

The Linear node determines the cause-effect relationships between variables, sets up regression models,

and checks the correlations between symptoms and the correlation directions and levels.

Restriction The node must follow a source (ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node, and cannot follow a Bin node.

A data mining process containing the Linear node must meet the following conditions:

− The Type node must contain only one output field and one or more input fields.

− If the input field is of the character string type, the field must be of Sign or Set role. If the input

field is of the numeral type, you do not need to set the Role parameter.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 46 of 124

− The output fields must be of Range role.

− The Linear node must contain a maximum of one partition field.

Model Input Example

id,AD,RD,commission,GDP,VOS

2012,1302,1851,723,57080,4970

2011,1204,1704,675,26461.5,4521

2010,1105,1557,629,48235.1,4124

2009,1002,1404,576,49495.9,3692

2008,903,1255,523,31377,3269

2007,802,1109,475,23339.1,2846

Fields are separated by commas (,).

On the Type node:

id: Primary Key

AD, RD, commission, and GDP: Input

VOS: Output; Role must be set to Range.

Parameter Description

Parameters

Parameter Description

Model File Model file name. Model files are stored in the Project name/Model

directory.

By default, MOD files are exported from the

${SmartMiner_HOME}/smartminer/Projects/test/Model

directory.

NOTE

In the preceding directory, test indicates the name of the project where the process is located.

Use Partition Specifies whether to use only the training data set to build models if

the Partition node is configured.

● Yes: Use only data in the training data set.

● No: Use data from all data sets.

Select Feature Specifies whether to select features. You can use the parameter to

define feature selection criteria based on the probability associated

with fields.

Threshold for Including

Significance

This parameter is valid only when Select Feature is set to Yes. The

default value is 0.05.

When the probability of a field is less than the value of Threshold

for Including Significance, the system adds the field to the model.

Threshold for Excluding

Significance

This parameter is valid only when Select Feature is set to Yes. The

default value is 0.1.

When the probability of a field is greater than the value of

Threshold for Excluding Significance, the system deletes the

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 47 of 124

Parameter Description

field.

NOTE The value of Threshold for Including Significance must be less than the value of Threshold for Excluding Significance.

Include Constant Specifies whether the model contains constants.

Visualization

Click the Linear model file. Information about the model is displayed, as shown in 0. The structure tree is

displayed on the left, indicating the attributes and values of the model, and the significance of attributes is

displayed in a column bar on the right.

Linear model file information

NOTE If Bagging and Boosting are used, the SmartMiner will generate multiple models. The model structure tree displays multiple models, among which the root node is ModelSet, the Boosting model weight is calculated based on the error rate, and the default Bagging model weight is 1.

GBDT Node

Function

The GBDT algorithm is an iterative DecisionTree algorithm. It consists of multiple decision trees. The

regression trees from each iteration are merged based on their weights. The algorithm is used to solve

regression and dichotomy problems.

Restriction The node must follow a source (ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

A data mining process containing the GBDT node must meet the following conditions:

− The Type node must contain at least one input field of the numeral or string type, and the Role

must be Sign, Set, or Range.

− The Type node must contain only one output field, and the Role must be Sign or Range.

The GBDT node follows a Type, Binning, Partition, Filler, Filter, Sampling or Select node, and is

followed by a GBDTApply, TextFileExport or DataBaseExport node.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 48 of 124

Model Input Example

range1,set1,sign1,output

10,AA,yes,-3.9

21,BB,yes,6.9

5,AA,yes,5

7,CC,yes,6

8.6,BB,no,9

35,CC,no,12

Fields are separated by commas (,).

On the Type node:

range1, set1, and sign1: Input

output: Output

Parameter Description

Parameters

Parameter Description

Model File Model file name. Model files are stored in the Project

name/Model directory.

By default, MOD files are exported from the

${SmartMiner_HOME}/smartminer/Projects/test/Model

directory.

NOTE

In the preceding directory, test indicates the name of the project where the process is located.

Tree Depth Maximum layer of nodes (root node excluded) in a decision tree,

that is, depth of a decision tree.

The value ranges from 1 to 8.

Default value: 5

Step (Decrease) It determines the merging weight of each decision tree.

The value ranges from 0.01 to 1.

Default value: 1

Round It determines the max number of the decision trees.

Default value: 3

Use Partitioned Specifies whether to use only the training data set to build models

if the Partition node is configured.

The options are as follows:

● Yes: Use only data in the training data set.

● No: Use data from all data sets.

Default value: No

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 49 of 124

PCA Node

Function

The PCA node transforms multiple indexes to few comprehensive indexes that are not correlated.

Restriction The node must follow a source (ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

The PCA node can contain only fields of the value type. To generate the model for factor analysis or

principal component analysis, one or more input fields are required. The node ignores output,

bidirectional, and nondirectional fields.

Model Input Example

ID,FIELD1,FIELD2,FIELD3,FIELD4,FIELD5,FIELD6,FIELD7,FIELD8

1,40.4,24.7,7.2,6.1,8.3,8.7,2.442,20

2,25,12.7,11.2,11,12.9,20.2,3.542,9.1

3,13.2,3.3,3.9,4.3,4.4,5.5,0.578,3.6

4,22.3,6.7,5.6,3.7,6,7.4,0.176,7.3

5,34.3,11.8,7.1,7.1,8,8.9,1.726,27.5

6,35.6,12.5,16.4,16.7,22.8,29.3,3.017,26.6

Fields are separated by commas (,).

On the Type node, Direction of ID is set to None, and Direction and Role of other fields are set to Input

and Range respectively.

Parameter Description

Parameters

Parameter Description

Model File Model file name. Model files are stored in the Project

name/Model directory.

By default, MOD files are exported from the

${SmartMiner_HOME}/smartminer/Projects/test/Model

directory.

NOTE

In the preceding directory, test indicates the name of the project where the process is located.

Use Partition Specifies whether to use only the training data set to build models

if the Partition node is configured.

● Yes: Use only data in the training data set.

● No: Use data from all data sets.

Extract Principal Component By Method for specifying the number of principal components and

extracting principal component factors from input fields.

The options are as follows:

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 50 of 124

Parameter Description

● By Component Count

This method retains a specified number of factors or

components sorted by feature value in descending order. By

default, the node retains five factors or components with the

highest feature values.

● By Min. Eigen

Feature values measure a factor's or component's capability for

describing the deviation of the input field set. This method

retains factors or components whose feature value is greater

than the specified threshold. Ensure that the threshold is not

too large to retain any principal component. The default feature

value threshold is 1.

Rotation Method You can rotate the retained factor set to improve the practicability

of the factors and make the factors easier to describe.

The options are as follows:

● Do not rotate: default value

● Varimax: This method minimizes the number of fields that are

overloaded for each factor, which facilitates the description of

the factor.

● Quartimax: This method minimizes the number of factors for

describing a field.

● Equamax: This method combines the Varimax and

Quartimax methods

NOTE

To rotate principal factors is to perform orthogonal transformation for changing the linear coefficient of principal factors. In this way, the relationships between principal factors and original factors are clearer, but the analysis results are not affected.

CF Node

Function

The CF node analyzes the similarity between users or items, and provides personalized offers to users

based on the similarities.

Restriction The node must follow a source (ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

The data mining process in which the CF node is configured must contain only one primary key field

and a minimum of one input field.

The computing framework (Hadoop or Spark) on which the CF node runs can be configured in the

${SmartMiner_HOME}/conf/smartminer.spark.nodes file. Parameters in Similarity Parameters

vary depending on the selected computing framework. For details about the parameters, see

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 51 of 124

Parameter Description. If the CF runs on the Spark, Harden Trust is not supported for Calculate

Similarity By user.

An example of the configuration file is as follows:

#[Parameter Group]: nodes implementation way

#[Parameter]: CF

#[Description]: CF node implementation way.

#[SetGuide]: Set this parameter based on the way to implement CF node.

#[Default]: hadoop

#[Range]: hadoop

CF=hadoop //The node runs on the Hadoop framework.

CF=spark //The node runs on the Spark framework.

Model Input Example 1 (Item Similarity)

ID,POINT,THINGS

1001,5.2,shoes

1001,7,football

1001,9,mineralwater

1002,7.3,basketball

1002,5.5,shoes

Fields are separated by commas (,).

On the Type node:

− ID: Primary Key

− POINT: Input

− THINGS: Input

On the CF node:

− Calculate Similarity: By item

− Item field: THINGS

− Rating field: POINT

Model Input Example 1 (User Similarity)

user,item,score

uu1,ii1,5

uu1,ii4,4

uu2,ii1,3

uu2,ii2,2

Fields are separated by commas (,).

On the Type node:

− user: Primary Key

− score: Input

On the Type node:

− Calculate Similarity: By user

− Item field: item

− Rating field: score

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 52 of 124

Parameter Description

Parameters

Parameter Description

Similarity Parameters

Harden Trust Specifies whether to consider item popularity when calculating user similarity.

For example, when this parameter is used in item recommendation, the item

popularity is considered in addition to the intersection of items used by

different users.

NOTE

This parameter is valid only when Calculate Similarity is set to User Similarity and the CF node runs on the Hadoop.

Item Field Object rated by users.

Rating Field Score of an item. The system calculates the similarity between users or items

based on users' scores on an item or several items.

Min. Items The parameter is valid only when the CF node runs on the Hadoop.

Minimum number of rating times of a user. The rating history of a user is used

as the input metadata only when the number of rating times of the user reaches

the value of the parameter.

Min. Cluster Members The parameter is valid only when the CF node runs on the Hadoop.

Minimum number of data subsets on a cluster. The default value is 5,

indicating that a cluster contains a minimum of 5 subsets.

A larger value indicates less calculation workload and lower accuracy. A

smaller value indicates more calculation workload and higher accuracy.

Max. Neighbors The parameter is valid only when the CF node runs on the Hadoop.

Maximum number of records to be selected from a cluster to compare with the

existing records for calculating similarity after records are clustered using the

MinHash clustering algorithm.

Similarity Threshold The parameter is valid only when the CF node runs on the Spark.

Only projects or users whose similarity values are greater than the value of this

parameter are generated in result.

Max similarity items The parameter is valid only when the CF node runs on the Spark.

Maximum number of output similar projects of users.

Computing resource

number

The parameter is valid only when the CF node runs on the Spark.

It is recommended that this parameter be set to spark.executor.instances

multiplied by spark.executor.cores (the two parameters are defined in the

${HOME}/conf/smartminer.spark.properties/smartminer.spark.properties

file on the SmartMiner server). Ensure that the value of this parameter is less

than or equal to the number of virtual cores in the Hadoop (which is specified

by the VCoresTotal parameter).

You can check the value of VCores Total in the Yarn service on the Hadoop

Manager GUI.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 53 of 124

Parameter Description

Use Partition Specifies whether to use only the training data set to calculate similarity and

conduct collaborative recommendation if the Partition node is configured. The

options are as follows:

● Yes: Use only data in the training data set.

● No: Use data from all data sets.

Collaborative Filtering Parameters

Collaborative Filtering Specifies whether to use the collaborative recommendation mode. You can

click the check box next to the parameter to select it.

When this parameter is used, the system recommends items that a user may

have interest in to the user. If items are recommended, the system generates the

possible score that the user may rate.

Max.

Recommendations

Maximum number of items that can be recommended to a user.

NOTE This parameter is valid only when Collaborative Filtering is set to Yes.

Max. Ratings Maximum number of items rated by a similar user.

When recommending items to a user, the system needs to refer to the items

rated by other similar users. A similar user may have rated a large number of

items. To improve the calculation efficiency, the system filters some rated

items for calculation based on this parameter value. A larger value indicates

lower calculation complexity and lower recommendation accuracy.

If this parameter is left empty, the maximum number is not restricted.

NOTE This parameter is valid only when Collaborative Filtering is set to Yes.

Max. Similarities Maximum number of users (or items) similar to a user (or an item).

When recommending items to a user (or an item), the system needs to refer to

the items rated by other similar users (or similar rated items). There may be a

large number of similar users (or similar items). To improve the calculation

efficiency, the system filters some similar users (or similar items) for

calculation based on this parameter value. A larger value indicates lower

calculation complexity and lower recommendation accuracy.

If this parameter is left empty, the maximum number is not restricted.

NOTE

This parameter is valid only when Collaborative Filtering is set to Yes.

SNSRS Node

Function

The SNSRS node uses the SNS topology to build models and obtain the recommendations that are hidden

behind the network.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 54 of 124

Restriction The node must follow a source (TextImport, ImportFeatureLibrary or DatabaseImport) node and a

Type node.

The SNSRS node must contain a minimum of one input field that functions as the user field and a

minimum of one input field that functions as the item field.

Model Input Example 1 (Heat Spreading)

A,B

U1,I2

U1,I3

U2,I2

U2,I4

U2,I5

U3,I3

Fields are separated by commas (,).

On the Type node, Direction of all fields is set to Input.

On the SNSRS node:

− ITSF

− User field: A

− Item field: B

− Recommendation algorithm: HeatSpreading

Model Input Example 1 (Probability Spreading)

A,B

U1,I2

U1,I3

U2,I2

U2,I4

U2,I5

U3,I3

Fields are separated by commas (,).

On the Type node, Direction of all fields is set to Input.

On the SNSRS node:

− User field: A

− Item field: B

− Recommendation algorithm: ProbabilitySpreading

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 55 of 124

Parameter Description

Parameters

Parameter Description

Recommendation Algorithm Recommendation algorithm. The options are as follows:

● HeatSpreading

Heat of an item indicates users' acceptance degree

towards the item. The heat of rated items is 1 and that of

unrated items is 0. Heat is spread from high heat to low

heat. In the spreading process, the system spreads heat

from rated items to users and then to unrated items.

Unrated items with higher heat are recommended first.

● ProbabilitySpreading

Probability indicates the likelihood that a user accepts an

item. The probability of rated items is 1. Assume that

probability can be spread on the SNS network. In the

spreading process, the system spreads probability from

rated items to users and then to unrated items. Unrated

items with higher probability are recommended first.

Constringency Factor Constringency factor. A greater Lambda value indicates a

higher probability for recommending unpopular items.

Use Partition Specifies whether to use only the training data set to build

models if the Partition node is configured.

● Yes: Use only data in the training data set.

● No: Use data from all data sets.

PersonalTag Node

Function

The PersonalTag node analyzes the initial preferences, preview history, and features of previewed contents

of users, and recommends offers to users accordingly.

Restriction The node must follow a source (ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

The data mining process in which the PersonalTag node is configured must contain a minimum of

three input fields.

Model Input Example

USER;MOVIE;SCORE;ITEM1;ITEM2;ITEM3

U1;M1;5;I11|I12;I21|I22|I23;I31

U1;M2;4;I11;I21|I23;I32|I33

U2;M3;4;I12|I13;I22|I23;I31|I32

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 56 of 124

U2;M4;2;I11|I13;I23;I31|I33

U3;M2;3.5;I11;I21|I23;I32|I33

Fields are separated by commas (,).

On the Type node, Direction of all fields is set to Input.

On the PersonalTag node:

− User field: USER

− Item field: MOVIE

− Rating field: SCORE

Parameter Description

Parameters

Parameter Description

Parameters

User Field Choose a field as the user attribute.

Item Field Choose a field as the item attribute.

Rating Field Choose a field as the rating attribute.

NOTE

The value of User Field, Item Field, and Rating Field cannot be the same.

Number of User Preferences Number of user preferences to be reserved during

calculation.

The value ranges from 3 to 15.

Default value: 5

Coverage Weight Coverage weight used during feature calculation.

The value ranges from 0 to 1.

Default value: 0.8

Statistics Times Weight Weight of the number of statistics times used

during feature calculation.

The value ranges from 0 to 1.

Default value: 0.2

NOTE

The sum of Coverage Weight and Statistics Times

Weight must be 1.

Recommendation List Threshold Recommendation result threshold. If the

recommendation result of an offer in a

recommendation list is greater than the value, the

offer will be reserved. Otherwise, the offer will be

discarded.

The value ranges from 0 to 10.

Default value: 0

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 57 of 124

Parameter Description

Recommendation List Length Maximum number of offers that a recommendation

list can contain.

The value ranges from 5 to 100.

Default value: 50

Generate User Preference The options are as follows:

● Yes: Export user preferences to a file. You can

customize the file name by configuring User

Preference.

● No: Not to export user preferences.

Default value: No

User Preference Name of the file to which user preferences are to

be exported.

It is available when Generate User Preference is

set to Yes.

Generate Feature Set The options are as follows:

● Yes: Export the feature set to a file. You can

customize the file name by configuring

Feature Set.

● No: Not to export the feature set.

Default value: No

Feature Set Name of the file to which the feature set is to be

exported.

It is available when Generate Feature Set is set to

Yes.

NOTE

The name of the user preference file must be different from that of the feature set file.

Score Weighted Value Weight of the item score.

The value ranges from 0 to 1000.

Default value: 0

Statistics Times Weighted Value Weight of the number of times statistics is

collected for an item.

The value ranges from 0 to 1000.

Default value: 1

NOTE

Score Weighted Value and Statistics Times Weighted

Value cannot both be 0.

Use Partition Indicates whether to partition the data source if a

Partition node is configured.

The options are as follows:

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 58 of 124

Parameter Description

● Yes: partition the data source into the training

data set for recommendation and the test data

set for evaluation.

● No: not partition the data source and use full

data for recommendation and evaluation.

Default value: No

Multi-Attribute Configuration

NOTE At least one Multi-Attribute needs to be configured.

Contain Multiple Values The options are as follows:

● Yes: The attribute contains multiple values.

● No: The attribute contains only one value.

Weight The weight of the field.

The value ranges from 0 to 100.

Default value: 1

NOTE

The weights of the fields cannot be all 0.

Multi-Attribute Separator Separator for separating multiple values in an

attribute.

NOTE The value of Multi-Attribute Separator cannot be the same as that used on the source node.

DiscriminationTree Node

Function

The Discrimination node provides recommendations to new users based on the existing user group

information as follows: The system asks a new user questions, uses the answers to find a matching user

group for this user, and recommend preferences of the user group to the user. (A recommendation can be a

movie that has the highest score or is most frequently watched.)

Restriction The node must follow a source (ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

The data mining process in which the PersonalTag node is configured must contain a minimum of

three input fields.

Model Input Example

USER,MOVIE,SCORE

U1,M29,2

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 59 of 124

U1,M10,2

U1,M1,5

U1,M27,4

U1,M16,1

U1,M20,2

Fields are separated by commas (,).

On the Type node, Direction of all fields is set to Input.

On the Discrimination node:

− User field: USER

− Item field: MOVIE

− Rating field: SCORE

Parameter Description

Parameters

Parameter Description

User Field Choose a field as the user attribute.

Item Field Choose a field as the item attribute.

Rating Field Choose a field as the rating attribute.

NOTE

The value of User Field, Item Field, and Rating Field cannot be the same.

Tree Count Number of Discrimination trees, for example, the

number of questions displayed to users in one

round.

Height Depth of a Discrimination tree, for example, the

number of rounds in which questions are displayed

to users.

Min. Records on Leaf Node Minimum number of records that is required in a

leaf node. If the number of records is less than the

value of this parameter, no more leaf nodes will be

created.

Preference Threshold User preference threshold. Items with a value

greater than the threshold will be considered as

user preferences.

Score Pruning Coefficient Formula correction value introduced in case that

there are few rating records for an item in a leaf

node. The value 0 indicates no correction will be

made. A greater threshold indicates a more

obvious correction will be made to the

recommendation rating.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 60 of 124

Parameter Description

Recommendation Novelty Threshold Formula correction value introduced to improve

recommendation novelty. The value 0 indicates no

correction will be made. A greater threshold

indicates a more obvious correction will be made

to the recommendation rating.

Use Partition Indicates whether to partition the data source if a

Partition node is configured.

The options are as follows:

● Yes: partition the data source into the training

data set for recommendation and the test data

set for evaluation.

● No: not partition the data source and use full

data for recommendation and evaluation.

Recommendation List Length Max number of recommendations that will be

provided to a user.

SimilarFeature Node

Function

The SimilarFeature node calculates the similarity of contents based on the features and the feature weight.

This node supports incremental feature similarity calculation. After a full calculation, you can import only

new, deleted, and updated records, and then the system can calculate feature similarities based on only the

imported records and combine the calculation result with the last analysis result. This function can save

computing resources because it does not calculate the similarities between existing records.

Restriction The node must follow a source (ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

The data mining process in which the SimilarFeature node is configured must contain only one

primary key field and a minimum of one input field.

The analysis result is exported using the ExportText node.

In the incremental calculation mode, the BDI is required. After a full analysis, the BDI periodically

loads incremental data and invokes the SmartMiner for incremental analysis.

In the incremental calculation mode, the input data must contain a flag field that indicates data

resetting, creation, update, or deletion. This field can be user-defined. On the Type node, you must

comply with the following rules when defining the field:

− Set Role of the field to Set, and values of the set must contain r (resetting), n (creation), u (update),

and d (deletion).

− Set Direction of the field to Input.

Model Input Example (Full Calculation Mode)

TONEID,SINGER,SINGERSEX,TONELANGUAGE,TONEINFO,PRICE

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 61 of 124

77954,Football Songs,1,2,HipHop,50

64089,alfasuarawuolorunninbe,1,2,gospel,50

64120,raskimono,1,2,raggae,50

78601,Soule Baba,2,2,Naija,50

Fields are separated by commas (,).

On the Type node, Direction of TONEID is set to Primary Key, and Direction for other fields is set to

Input.

Model Input Example (Incremental Calculation Mode)

The following shows the initial full input data:

TONEID,SINGER,SINGERSEX,TONELANGUAGE,TONEINFO,PRICE,SIGN

1,Nubia,2,2,HipHop,50,r

2,Usher,2,2,HipHop,50,r

3,Benie Man,2,2,HipHop,50,r

In the data, SIGN is the flag field. It can be left empty for full data import. The default value of SIGN is r

by default.

Feature similarity modeling process

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 62 of 124

Similarity parameter settings

The following shows the similarity analysis result in full calculation mode:

1,2,0.800

1,3,0.800

2,3,0.800

2,1,0.800

3,2,0.800

3,1,0.800

The following shows the incremental data:

TONEID,SINGER,SINGERSEX,TONELANGUAGE,TONEINFO,PRICE,SIGN

5,Benie Man,2,2,HipHop,50,n //Record creation

3,Benie Man,2,2,HipHop,50,d //Record deletion

2,Usher,2,2,HipHop,10,u //Record update

Each imported incremental data file overwrites the previous one, and the system automatically saves the

previous modeling data.

The following shows the similarity analysis result in incremental calculation mode:

1,5,0.800

1,2,0.600

2,5,0.600

2,1,0.600

5,1,0.800

5,2,0.600

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 63 of 124

Parameter Description

Parameters

Parameter Description

Attribute Parameters

Multi-Value Indicates whether a field can contain multiple values. For example, the SINGER

field of a song can have multiple values.

Weight Field weight. A field with a larger weight has greater impact on the similarity

results.

Multi-Value

Separator

Separator used to separate values of a field if the field has multiple values. For

example, Tom;Anna, which indicates that the values of SINGER are separated by

semicolons (;).

Similarity Parameters (Hadoop-based)

Min. Cluster

Members

Minimum number of data subsets on a cluster. The default value is 5, indicating

that a cluster contains a minimum of 5 subsets.

A larger value indicates less calculation workload and lower accuracy. A smaller

value indicates more calculation workload and higher accuracy.

Max. Neighbors Maximum number of records to be selected from a cluster to compare with the

existing records for calculating similarity after records are clustered using the

MinHash clustering algorithm.

Number of Hash

Functions

Number of Hash functions required for the similarity calculation.

The value of the parameter ranges from 1 to 100.

The default value is 20.

Number of Hash

Function Values

Number of items to be compared between two objects. If the items are the same, the

two objects are similar. The greater the parameter value is, the lower the similarity

probability is.

The value of the parameter ranges from 1 to 100.

The default value is 2.

NOTE Number of Hash Functions cannot be greater than Number of Hash Function Values.

Similarity Parameters (Spark-based)

Similarity

Threshold

Threshold lower than which similarities are not displayed in the analysis result.

If there is a large amount of input data and the calculated similarities are low, set

this parameter to a smaller value to increase the number of records in the generated

analysis result.

Max similarity

items

Maximum number of similarity records in the analysis result that contains

similarities in ascending order.

Computing

resource number

It is recommended that this parameter be set to spark.executor.instances

multiplied by spark.executor.cores (the two parameters are defined in the

${HOME}/conf/smartminer.spark.properties/smartminer.spark.properties file

on the SmartMiner server). Ensure that the value of this parameter is less than or

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 64 of 124

Parameter Description

equal to the number of virtual cores in the Hadoop (which is specified by the

VCoresTotal parameter).

You can check the value of VCores Total in the Yarn service on the Hadoop

Manager GUI.

Incremental Mode Indicates whether to use the incremental calculation mode. If this parameter is set

to No, the full calculation mode is used by default and the input data does not need

to contain a flag field.

Identifier Field Specifies a flag field. This parameter is valid only when Incremental Mode is set

to Yes. The SmartMiner automatically reads the field whose Role is set to Set on

the Type node.

Delete Identifier Specifies the identifiers. This parameter is valid only when Incremental Mode is

set to Yes. The SmartMiner automatically reads the values of the field whose Role

is set to Set on the Type node. Update Identifier

New Identifier

Reset Identifier

FullConnected Node

Function

The FullConnected node is used to find fully connected submaps for home networks.

Restriction The node must follow a source (TextImport, ImportFeatureLibrary or DatabaseImport) node and a

Type node.

The data mining process in which the FullConnected node is configured must contain a minimum of

two input fields.

Model Input Example

USER LUSER

A B,C,D,E,F,H

B A,D,E,G,H

C A,E,F,G,H

D A,B,F,G,H

E A,B,C,F,H

F A,C,D,E

G B,C,D,H

H A,B,C,D,E,G1

Fields are separated by commas (,).

On the Type node, Direction of all fields is set to Input.

On the FullConnected node:

− Vertex field: USER

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 65 of 124

− Neighboring field: LUSER

− Neighbor separator: ,

Parameter Description

Parameters

Parameter Description

Convert to Undirected Figure Whether to use the bidirectional data transmission

mode. A bidirectional relationship example is as

follows: A and B have called each other.

Vertex Field Vertex field to be analyzed. In a fully connected

map, every two vertex fields are connected. For

example, if field A, B, C, and D are all vertex

fields, the data format is as follows:

● A|B;C;D

● B|B;C;D

● C|B;C;D

● D|B;C;D

NOTE

In the preceding format, the field before | is a vertex field, fields after | are neighboring fields, and ; is a neighboring field separator.

Neighbor Field Neighboring field of a vertex field. A vertex may

have multiple neighboring fields. Therefore,

Neighboring Field may contain multiple values.

Max. Full Connections Maximum number of vertex fields. In a fully

connected map, every two records are connected.

The parameter specifies the number of vertex

fields in a fully connected map.

TextClassify Node

Function

The TextClassify node segments text and forecasts its classification.

Restriction The node must follow a ImportFolder node and a Type node.

Generally, the TextClassify node follows the FolderImport node. If the TextClassify node follows

another node, it must contain the CATALOG, SUBCATALOG, FILENAME, and CONTENT input

fields.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 66 of 124

Model Input Example

The process of using the TextClassify node is: import a directory -> use a Type node to process data ->

classify text.

You can upload a directory to the Data directory of the corresponding project, for example,

/homedcp01/autoTest/Projects/TextClassify_mod/Data/classify/mine.app.TextClassify.functiona.027.txt,

, and then select the corresponding directory on the ImportFolder node.

Parameter Description

Parameters

Parameter Description

Model File Model file name. Model files are stored in the

Project name/Model directory.

Use Partition Specifies whether to use only the training data set

to build models if the Partition node is configured.

● Yes: Use only data in the training data set.

● No: Use data from all data sets.

Text Type Text format.

NOTE Text of the Wed type only supports web pages using the WAP protocol

SPA Node

Function

The SPA node expands influence and identifies users based on the SNS network.

You can use the SPA node to forecast results by classification, for example, customer loss probability and

whether a customer will accept an offer. For example, if the system wants to forecast customer loss

probability, it defines some lost customers on the SNE network, and finds the influence the lost customers

have on other customers based on their call frequency and duration. Then the system calculates the

customer loss probability based on the obtained data and iteratively expands the calculated probability

through the influence spreading expression until the probability seldom changes.

Restriction The node must follow a source (ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

The SPA node must contain a minimum of three input fields that function as analysis fields.

Model Input Example

A,B,C

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 67 of 124

a,y,b|1

b,n,f|1;e|5

c,n,e|1

d,y,

e,n,

f,y,e|1;d|1;g|1

g,n,d|1

Fields are separated by commas (,).

On the Type node, Direction of all fields is set to Input.

On the SAP node:

− Vertex field: A

− Type field: B

− Neighbor field: C

− Separator between neighboring fields: ;

− Separator between neighboring weights: |

Parameter Description

Parameters

Parameter Description

Input/Output Parameters

Vertex Field Vertex field to be analyzed. In a fully connected

map, every two vertex fields are connected. For

example, if field A, B, C, and D are all vertex

fields, the data format is as follows:

● A|B;C;D

● B|B;C;D

● C|B;C;D

● D|B;C;D

NOTE

In the preceding format, the field before | is a vertex field, fields after | are neighboring fields, and ; is a neighboring field separator.

Neighbor Field Neighboring field of a vertex field. A vertex may

have multiple neighboring fields. Therefore,

Neighbor Field may contain multiple values.

Neighbor Separator Separator between neighboring fields.

Neighbor Contain Weight Specifies whether neighboring fields contain

weight. The options are as follows:

● Yes

Read neighboring fields in the

V1|weight1|V2|weight2|V3|weight3 format.

● No

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 68 of 124

Parameter Description

Read neighboring fields in the V1|V2|V3

format. The weight is set to 1.

Neighbor Weight Separator Neighboring field weight separator. The value

cannot be the same as the field separator or the

value of NNeighbor Separator.

This parameter is valid only when Neighbor

Contain Weight is set to Yes.

Type Field User type field, for example, a field that indicates

whether a user is online or offline.

Predict Type Forecast type. For example, if the options of Type

Field are A and B, and the value of Predict Type

is B, the SmartMiner will forecast the probability

of the event that A is changed into B.

Use Partition Specifies whether to use only the training data set

to build models if the Partition node is configured.

● Yes: Use only data in the training data set.

● No: Use data from all data sets.

Generate Forecast Record Specifies whether to generate records of the

forecast type. For example, if the options of Type

Field are A and B, and the value B indicates the

forecast type, this parameter specifies whether to

generate records whose Type Field is B.

Algorithm Parameters

Spreading Factor Percentage of spread influence to the original

influence. A smaller Spreading Factor value

indicates greater influence on the vertexes around

the influence source, and a larger Spreading

Factor value indicates a larger influence scope.

End Threshold Spread end threshold. If the accepted influence is

lower than the threshold, the spreading operation

ends. A smaller End Threshold value indicates a

larger influence scope.

Classification Threshold Forecast result classification threshold. If the final

influence is higher than the threshold, the system

forecasts records as the forecast type. Therefore, a

smaller Classification Threshold value indicates

that more records are forecasted as the forecast

type.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 69 of 124

OverlapNeighbour Node

Function

The OverlapNeighbour node finds node pairs that have overlapping neighboring points.

Restriction The node must follow a source (ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

The data mining process in which the OverlapNeighbour node is configured must contain a minimum

of three input fields.

Model Input Example 1 (The Neighbor Field Contains No Coefficient)

User Type Neighbor

A TYPE1 B|C|D|E|F

B TYPE1 A|D|E|G

Fields are separated by commas (,). The values of the neighbor field are separated by vertical bars (|).

On the Type node, Direction of all fields is set to Input. The roles and value ranges of all fields do not

need to be configured.

On the OverlapNeighbor node:

− Vertex field: User

− Type field: Type

− Neighbor field: Neighbor

− Neighbor field separator: |

− Neighbor field coefficient: No

Model Input Example 2 (The Neighbor Field Contains Coefficient)

User Type Neighbor

A TYPE1 B,5|C,10|D,15|E,20|F,5

B TYPE1 A,5|D,10|E,15|G,20

Fields are separated by commas (,). The values of the neighbor field are separated by vertical bars (|). The

neighboring field and its coefficient are separated by comma (,).

On the Type node, Direction of all fields is set to Input. The roles and value ranges of all fields do not

need to be configured.

On the OverlapNeighbor node:

− Vertex field: User

− Type field: Type

− Neighbor field: Neighbor

− Neighbor field separator: |

− Neighbor field coefficient: Yes

− Coefficient separator: ,

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 70 of 124

Parameter Description

Parameters

Parameter Description

Convert to Undirected Graph Whether to use the bidirectional data transmission

mode. A bidirectional relationship example is as

follows: A and B have called each other. This parameter

is used to set the data transmission mode to

bidirectional.

Vertex Field Vertex field to be analyzed. In a fully connected map,

every two vertex fields are connected. For example, if

field A, B, C, and D are all vertex fields, the data format

is as follows:

● A|B;C;D

● B|B;C;D

● C|B;C;D

● D|B;C;D

NOTE

In the preceding format, the field before | is a vertex field, fields after | are neighboring fields, and ; is a neighboring field separator.

Neighbor Field Neighboring field of a vertex field. A vertex may have

multiple neighboring fields. Therefore, Neighboring

Field may contain multiple values.

Vertex Type Specifies whether vertex fields are of a same type. For

example, if two vertex fields share a neighboring field

in a batch of data, the vertex fields are of a same type.

● Different Type

Scenario: Assume that a mobile phone user whose

number is A subscribed to a new package and

changed the mobile number to B in October, and

consumption data of number B is generated in

November. The SmartMiner analyzes the data of

number A in October and the data of number B in

November and finds that contacts of number A and

B overlap. As a result, the system can draw the

conclusion that number A and number B belong to a

same user. In this case, a mobile number is a vertex

field and the vertex fields are of different types.

● Same Type

Scenario: The SmartMiner analyzes a batch of data

and finds that the contacts of mobile number E and F

overlap. As a result, the system can draw the

conclusion that the user of number E may be an

acquaintance to the user of number F, and the

service side can recommend F as a friend to E. In

this case, a mobile number is a vertex field and the

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 71 of 124

Parameter Description

vertex fields are of a same type.

Min. Overlapping Neighbors Minimum number of overlapping neighbors. When the

number of overlapping neighbors between two nodes

reaches the value of Min. Overlapping Neighbors, the

fields are similar nodes.

SparseLinear Node

Function

The SparseLinear node supports a large number of features, precisely analyzes multi-dimensional data, and

builds models.

Restriction The node must follow a source (ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

A data mining process containing the SparseLinear node must meet the following conditions:

− The process contains a minimum of two input fields and only one output field.

− The input fields cannot be of the time, date, or timestamp type. If an input field is of the character

string type, the corresponding Role cannot be set to Range.

− For output fields, Role must be set to Sign.

Model Input Example

ID,UserID,Age,ARPU,Tags,Sex,Is,ItemID,Type,Score

1,user123,28,20.3,China|Huawei1|Huawei2|Huawei3|Huawei4,male,Y,item1234,MI|Honor,1

2,user456,30,30.3,China|Huawei,female,Y,item7890,MI|Honor,1

3,user123,28,20.3,China|ZTE,male,N,item1235,MI|Apple,0

4,user789,27,20.3,China|Huawei,male,Y,item8888,MI|Honor,1

5,user100,15,22.3,China|Huawei,female,Y,item1234,MI|Honor,0

6,user101,16,21.3,China|Huawei,female,Y,item1234,MI|Honor,0

7,user102,18,31.3,China|Ericsson,female,N,item1234,MI|Samsung,0

Fields are separated by commas (,).

On the Type node, Direction of ID is set to Primary Key, Direction of Score is set to Output, and

Direction of other fields is set to Input.

Parameter Description

Parameters

Parameter Description

Parameters

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 72 of 124

Parameter Description

Model File Model file name. Model files are stored in the Project

name/Model directory.

By default, MOD files are exported from the

${SmartMiner_HOME}/smartminer/Projects/test/Model

directory.

NOTE In the preceding directory, test indicates the name of the project where the process is located.

Use Partition Specifies whether to use only the training data set to build models

if the Partition node is configured.

● Yes: Use only data in the training data set.

● No: Use data from all data sets.

Split Set into Multiple Columns Specifies whether to convert a field of the set type into multiple

fields.

Support Multiple Values Specifies whether the source data can be multi-value data.

Base Category Reference value of the model forecast field, which is one of the

flag values of modeling node output fields.

Advanced Parameters

Advanced Parameters Algorithms for SparseLinear modeling. Currently, only the

LBFGS algorithm is supported.

RandomForest Node

Function

The RandomForest node supports a large number of features and builds multiple decision tree models to

abstract classification rules through random sampling, which avoids overfitting caused by the use of a

single decision tree.

Restriction The node must follow a source (ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

A data mining process containing the RandomForest node must meet the following conditions:

− The process contains a minimum of one input field and only one output field.

− If the input fields are of the character type, Role must be set to Sign, Set, or Range. If the input

fields are of the number type, Role does not need to be set.

− For output fields, Role must be set to Sign or Set.

Model Input Example

id,sign1,set1,rangei1,rangef2,string1,rangei3,rangef4,string2,sign2,set2,sign3,set3

0,0,b,-474,7600.86,DJPVBH,-148,3299.60,JPVBHN,no,1,0,b

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 73 of 124

1,1,b,114,4083.32,PVBHNT,219,8399.50,QWCIOU,no,2,1,b

2,1,c,282,2765.76,SYEKQW,-37,1875.44,EKQWCI,yes,1,0,b

3,0,a,75,7768.26,GMSYEK,10,1717.19,SYEKQW,no,2,0,c

4,1,c,443,9209.32,GMSYEK,100,6838.23,LRXDJP,no,3,1,c

Fields are separated by commas (,).

On the Type node, Direction of ID is set to Primary Key, and Direction for other fields is set to Input.

Parameter Description

Parameters

Parameter Description

Parameters

Model File Model file name. Model files are stored in the Project

name/Model directory.

By default, MOD files are exported from the

${SmartMiner_HOME}/smartminer/Projects/test/Model

directory.

NOTE In the preceding directory, test indicates the name of the project where the process is located.

Number of Trees Number of decision tree submodels required during modeling.

Maximum Number of Bin Maximum number of bins for a feature during extraction.

Default value: 32

Maximum Number of Feature Maximum number of features required for building each decision

tree submodel.

Default value: 5

NOTE During sample extraction, sampling without replacement is used.

Maximum Number of Tree Depth Maximum layer of nodes in a decision tree.

Default value: 10

Use Partition Specifies whether to use only the training data set to build models

if the Partition node is configured.

● Yes: Use only data in the training data set.

● No: Use data from both the training and test data sets.

Min. Leaf Nodes NOTE It is recommended that you use the value obtained by dividing the number of records in the training data set by 2 to the power of L, in which L indicates the number of input fields in the training data set.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 74 of 124

PageRank Node

Function

The PageRank algorithm measures node importance. For example, it measures the importance of website

pages and ranks them by importance.

Restriction The node must follow a source (ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

A data mining process containing the PageRank node must meet the following conditions:

− The process must contain a minimum of three input fields.

− The Role field does not need to be configured.

− The data type of a start point and an end point can be integer, real, or string character; a weight can

be of the integer or real type.

Model Input Example

Start,End,Weight

A,B,10

B,A,10

A,C,20

C,A,20

A,D,30

D,A,30

Fields are separated by commas (,).

On the Type node, Direction of all fields is set to Input.

Parameter Description

Parameters

Parameter Description

Start Point Start point of an edge.

End Point End point of an edge.

Weight Weight of an edge.

Iteration Times Maximum number of iteration times for the

PageRank modeling algorithm. The model training

ends when the number of iteration times reaches

the value of Iteration Times.

Spreading Propagation Factor Percentage of spread influence to the original

influence. A smaller Spread Factor value

indicates greater influence on the vertexes around

the influence source, and a larger Spread Factor

value indicates a larger influence scope.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 75 of 124

Parameter Description

Iteration End Threshold Minimum change value of a PageRank value. The

iteration ends when the change value of a

PageRank value is less than the value specified by

this parameter

LDA Node

Function

Latent Dirichlet Allocation (LDA) is a way of automatically discovering themes in a large number of files

and predicting the generation of a theme model. LDA can also find categories that users prefer and

recommend by category.

Restriction The node must follow a source (ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

A data mining process containing the LDA node must meet the following conditions:

− Ensuring that word fields of the data type fields of the metadata on the source node are of the

character, floating point, or integer type and value fields are of the integer or floating point type.

− The process must contain a minimum of three input fields.

− The Role field does not need to be configured.

Model Input Example

document,word,score

1,1,3

1,2,3

1,3,3

1,4,3

1,5,3

1,6,3

Fields are separated by commas (,).

On the Type node, Direction of all fields is set to Input.

Parameter Description

Parameters

Parameter Description

Document Document used to build an LDA model. Example:

papers (during paper clustering)

Word Word used to build an LDA model. Example: a

word in a paper

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 76 of 124

Parameter Description

Score Score used to build an LDA model. Example: the

number of times that a word appears in a paper

Iteration Times Maximum number of iteration times for the LDA

modeling algorithm. The model training ends

when the number of iteration times reaches the

value of Iteration Times.

Number of Themes Number of clusters that the LDA model classifies

documents into.

Recommend Specifies whether to use an LDA model for user

clustering and content recommendation.

Application Nodes

Application nodes correspond to modeling nodes. You are advised to place an Export node after an

Application node to check the application result.

1.1.1.1 NaiveBayesApply Node

Function

The NaiveBayesApply node uses models generated on the NaiveBayes node to forecast sample

classification based on test sample data.

Restriction The node must follow a source(ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

The NaiveBayes model file exists.

Parameter Description

Parameters

Parameter Description

Positive Value Customer care type. For example, in a customer

model, the loss of a customer is marked as a

positive value and the retaining of a customer is

marked as a negative value.

Select a value of the sign role from the drop-down

list box.

NOTE

This parameter is available only when the output field is of the sign role.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 77 of 124

Parameter Description

Positive Value Rate Threshold Used for changing the default threshold in the

algorithm. A type is regarded as positive only

when the rate of the type exceeds the threshold.

Enter a real number ranging from 0 to 1.

NOTE This parameter is available only when the output field is of the sign role.

Use Partition Specifies whether to use only the test data set to

build models if the Partition node is configured.

● Yes: Use only data in the test data set.

● No: Use data from all data sets.

DecisionTreeApply Node

Function

The DecisionTreeApply node uses models generated on the DecisionTree node to forecast sample

classification based on test sample data.

Restriction The node must follow a source(ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

The DecisionTree model file exists.

Parameter Description

Parameters

Parameter Description

Positive Value Customer care type. For example, in a customer

model, the loss of a customer is marked as a

positive value and the retaining of a customer is

marked as a negative value.

Select a value of the sign role from the drop-down

list box.

NOTE

This parameter is available only when the output field is of the sign role.

Positive Value Rate Threshold Used for changing the default threshold in the

algorithm. A type is regarded as positive only

when the rate of the type exceeds the threshold.

Enter a real number ranging from 0 to 1.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 78 of 124

Parameter Description

NOTE This parameter is available only when the output field is of the sign role.

Use Partition Specifies whether to use only the test data set to

build models if the Partition node is configured.

● Yes: Use only data in the test data set.

● No: Use data from all data sets.

LogisticsApply Node

Function

he LogisticsApply node uses models generated on the Logistics node to forecast sample classification

based on test sample data.

Restriction The node must follow a source(ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node, and can not follow a Binning node.

The Logistics model file exists.

Parameter Description

Parameters

Parameter Description

Positive Value Customer care type. For example, in a customer

model, the loss of a customer is marked as a

positive value and the retaining of a customer is

marked as a negative value.

Select a value of the sign role from the drop-down

list box.

NOTE This parameter is available only when the output field is of the sign role.

Positive Value Rate Threshold Used for changing the default threshold in the

algorithm. A type is regarded as positive only

when the rate of the type exceeds the threshold.

Enter a real number ranging from 0 to 1.

NOTE

This parameter is available only when the output field is of the sign role.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 79 of 124

Parameter Description

Use Partition Specifies whether to use only the test data set to

build models if the Partition node is configured.

● Yes: Use only data in the test data set.

● No: Use data from all data sets.

LinearApply Node

Function

The LinearApply node uses models generated on the Linear node to forecast sample classification based on

test sample data.

Restriction The node must follow a source(ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node, and can not follow a Binning node.

The Linear model file exists.

Parameter Description

Parameters

Parameter Description

Use Partition Specifies whether to use only the test data set to

build models if the Partition node is configured.

● Yes: Use only data in the test data set.

● No: Use data from all data sets.

GBDTApply Node

Function

The GBDTApply node uses models generated on the GBDT node to forecast values of specified output

fields and sample classification based on test sample data.

Restriction The node must follow a source(ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

The corresponding GBDT model file exists.

The GBDTApply node follows a Type, Binning, Partition, Filler, Filter, Sampling, Select or GBDT

node, and is followed by a NumericalEvaluation, ClassifyEvaluation, ExportText or ExportDataBase

node.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 80 of 124

Parameter Description

Parameters

Parameter Description

Positive Value Customer care type. For example, in a customer

model, the loss of a customer is marked as a

positive value and the retaining of a customer is

marked as a negative value.

Select a value of the sign role from the drop-down

list box.

NOTE

This parameter is available only when the output field is of the sign role.

Positive Value Rate Threshold Used for changing the default threshold in the

algorithm. A type is regarded as positive only

when the rate of the type exceeds the threshold.

Enter a real number ranging from 0 to 1.

NOTE This parameter is available only when the output field is of the sign role.

Use Partition Specifies whether to use only the test data set if the

Partition node is configured.

● Yes: Use only data in the test data set.

● No: Use data from all data sets.

KmeansApply Node

Function

The KmeansApply node uses models generated on the Kmeans node to forecast sample classification based

on test sample data.

Restriction The node must follow a source(ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

The Kmeans model file exists.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 81 of 124

Parameter Description

Parameters

Parameter Description

Use Partition Specifies whether to use only the test data set to build models if the

Partition node is configured.

● Yes: Use only data in the test data set.

● No: Use data from all data sets.

Generate Distance Field Specifies whether to generate a distance field.

● Yes: Generate a distance field that records the distance between a

record and its cluster centers.

● No: Not generate a distance field.

EMApply Node

Function

The EMApply node uses model files generated on the EM node to forecast sample classification based on

test sample data.

Restriction The node must follow a source(ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

The corresponding EM model file exists.

The EMApply node follows an EM or Type node, and is followed by a ExportText, ExportDatabase or

ClusterEvaluation node.

Parameter Description

Parameters

Parameter Description

Use Partition Specifies whether to use only the test data set if the Partition node is configured.

● Yes: Use only data in the test data set.

● No: Use data from all data sets.

Generate Rate

Field

Specifies whether to generate a probability field.

● Yes: Generate a rate field, which specifies the probability that a record belongs

to a forecast cluster.

● No: Not to generate a rate field.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 82 of 124

AprioriApply Node

Function

The AprioriApply node uses models generated on the Apriori node to forecast sample classification based

on test sample data.

Restriction The node must follow a source(ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

The Apriori model file exists.

Parameter Description

Parameters

Parameter Description

Data Format Data format. The format is inherited from the

Apriori model. The Sparse Matrix and

Key-Value pairs formats are available.

Use Partition Specifies whether to use only the test data set to

build models if the Partition node is configured.

● Yes: Use only data in the test data set.

● No: Use data from all data sets.

TimeSeriesApply Node

Function

The TimeSeriesApply node uses models generated on the TimeSeries node to forecast results in the next

period based on test sample data.

Restriction The node must follow a source(ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

The TimeSeries model file exists.

Parameter Description

Parameters

Parameter Description

Time Points to Predict Number of time points in the next period to forecast results.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 83 of 124

PCAApply Node

Function

The PCAApply node uses models generated on the PCA node to extract principal components from test

sample data.

Restriction The node must follow a source(ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

The PCA model file exists.

Parameter Description

Parameters

Parameter Description

Use Partition Specifies whether to use only the test data set to

build models if the Partition node is configured.

● Yes: Use only data in the test data set.

● No: Use data from all data sets.

TextClassifyApply Node

Function

The TextClassifyApply node uses models generated on the TextClassify node to forecast sample

classification based on test sample data.

Restriction The node must follow a FolderImport node and a Type node.

The TextClassify model file exists.

Parameter Description

Parameters

Parameter Description

Text Type Data file text type.

NOTE Text of the Wed type only supports web pages using the WAP protocol.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 84 of 124

Parameter Description

Use Partition Specifies whether to use only the test data set to

build models if the Partition node is configured.

● Yes: Use only data in the test data set.

● No: Use data from both the training and test

data sets.

SparseLinearApply Node

Function

The SparseLinearApply node uses models generated on the SparseLinear node to forecast sample

classification or recommend based on test sample data.

Restriction The node must follow a source (ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

The corresponding SparseLinear model file exists.

Parameter Description

Parameters

Parameter Description

Model File Name Select a model file to be applied.

Use Partition Specifies whether to use only the test data set if the

Partition node is configured.

● Yes: Use only data in the test data set.

● No: Use data from all data sets.

Generate Recommendation List Specifies whether to generate a recommendation

list.

Use Full Recommendation Specifies whether to use full recommendation.

User Field User attribute. For example, in the event that

Henry goes to the supermarket to buy coke, Henry

is the value of this parameter.

Item Field Item attribute. For example, in the event that

Henry goes to the supermarket to buy coke, coke is

the value of this parameter.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 85 of 124

Parameter Description

Recommendation Coefficient Threshold Recommendation result threshold. If the

recommendation result of an offer in a

recommendation list is greater than the value, the

offer will be reserved. Otherwise, the offer will be

discarded.

Max. Recommendations Maximum number of items that a recommendation

list can contain.

RandomForestApply Node

Function

The RandomForestApply Node uses the model file generated by the RandomForest Node to predict the

sample classification.

Restriction The node must follow a source (ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

The corresponding RandomForest model file exists.

Parameter Description

Parameters

Parameter Description

Model File Select a model file to be applied.

Maximum Number of Feature Maximum number of features required for building

each decision tree submodel.

Default value: 5

Use Partition Specifies whether to use only the test data set if the

Partition node is configured.

● Yes: Use only data in the test data set.

● No: Use data from all data sets.

Generate Primary Key Only Specifies whether to generate the primary key only.

Graph Nodes

A Graph node collects data feature values and evaluates the model application result.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 86 of 124

1.1.1.1 ClassifyEvaluation Node

Function

The ClassifyEvaluation node checks the forecast accuracy of NaiveBayes, DecisionTree, Logistics and

GBDT models by analyzing data generated during the model application.

Restriction The value set of the evaluation field on the ClassifyEvaluation node must be the same as that of the

output field in the modeling process. If null values exist, the system discards the values.

For non-third-party data evaluation, the ClassifyEvaluation node follows a NaiveBayesApply,

DecisionTreeApply, LogisticsApply, and GBDTApply node.

For third-party data evaluation, the ClassifyEvaluation node follows a Type node.

Parameter Description

Parameters

Parameter Description

Evaluation File Name Exported evaluation file name. Evaluation files are stored in

the Project name/Evaluation directory.

Evaluation Field Name of the field to evaluate. Evaluation Field is an actual

field.

Prediction Field Algorithm forecast result.

NOTE

This parameter

r is not displayed for non-third-party data evaluation.

Visualization

Click the ClassifyEvaluation file. The system displays the classify evaluation information, as shown in

Figure.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 87 of 124

ClassifyEvaluation file information

NOTE The AUC parameter will not be displayed if the ClassifyEvaluation node follows a Type node.

ClusterEvaluation Node

Function

The ClusterEvaluation node checks the clustering accuracy of EM and Kmeans models by analyzing data

generated during the model application.

Restriction The ClusterEvaluation node can only follow a ClusterApply node. Currently, EMApply and

KmeansApply nodes are supported.

Generate Primary Key Only cannot be selected for the ClusterApply node.

Parameter Description

Parameters

Parameter Description

Evaluation File Name Name of an exported evaluation file. Evaluation files are

stored in the Project name/Evaluation directory.

Visualization

Click the ClusterEvaluation file.The system displays the cluster evaluation information, as shown in Figure.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 88 of 124

ClusterEvaluation file information

NumericalEvaluation Node

Function

The NumericalEvaluation node checks the forecast accuracy of Linear and GBDT models by analyzing

data generated during the model application. The NumericalEvaluation node supports third-party data

evaluation.

Restriction For non-third-party data evaluation, the NumericalEvaluation node follows a LinearApply node.

For third-party data evaluation, the NumericalEvaluation node follows a Type node. The Type node

must be configured with a minimum of two input fields of the Range type.

Parameter Description

Parameters

Parameter Description

Evaluation File Name of the evaluation file that is generated.

Evaluation files are stored in the Project

name/Evaluation directory.

Evaluation Field Field to be evaluated. It is an actual field.

Prediction Field Forecast result of the algorithms.

NOTE This parameter is not displayed for non-third-party data evaluation.

Model Verification Counter Mode for verifying the application result.

The options are as follows:

● Error Rate

● Pearson

● Anova

● Kolmogorov-Smirnov

Residual Differences between the actual value and

forecasted value, including differences in the basic

and advanced statistics items.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 89 of 124

Parameter Description

● Basic

Valid records, percentage of valid records,

maximum value, minimum value, and average

value.

● Advanced

Standard deviation, deviation, skewness,

skewness standard deviation, kurtosis, and

kurtosis standard deviation.

NOTE If you select Advanced, Basic will be selected automatically. If you deselect Basic, Advanced will be deselected automatically.

Bin Count Defines the segments on the X axis of the residual

distribution chart used during numerical evaluation

result visualization.

The value ranges from 30 to 200.

Visualization

Click the NumricalEvaluation file. The system displays the numerical evaluation information, as shown in

0. In the chart, the columns indicate the distribution of model residuals. The default number of bins is 50

and the value range that is binned is from the standardized maximum residual to the standardized minimum

residual, which is the value range on the X axis. The Y axis indicates the number of residual records falling

into a specified bin. The blue curve indicates the theoretical distribution of residuals.

NumericalEvaluation file information

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 90 of 124

RecommenderEvaluation Node

Function

The RecommenderEvaluation node analyzes model application data generated by the CF, SNSRS,

PersonalTag nodes, DiscriminationTree Node and SparseLinear Node to evaluate the rating counter,

classification counter, coverage, accuracy, variety, and novelty.

Restriction For non-third-party data evaluation, the RecommenderEvaluation node follows a CF, SNSRS,

PersonalTag nodes, DiscriminationTree Node and SparseLinear Node.

For third-party data evaluation, the RecommenderEvaluation node follows a Type node.

Parameter Description

Parameters

Parameter Description

Evaluation File Exported evaluation file name. Evaluation files are

stored in the Project name/Evaluation directory.

Rating Indicator Rating indicator to be calculated. Select rating

indicators by clicking the check boxes. The options

are as follows:

● MAE: A smaller value indicates more accurate

scores.

● MSE: A smaller value indicates more accurate

scores.

● RMS error: A smaller value indicates more

accurate scores.

● Distance-Based mean evaluation counter: A

smaller value indicates more accurate scores.

Classification Counter Classification indicator to be calculated. Select

classification indicators by clicking the check

boxes. The options are as follows:

● Accuracy: A larger value indicates better

recommendation effects.

● Callback Rate: A larger value indicates better

recommendation effects.

● Average Recommended Length: It displays

the average value of all the users' actual

recommended lengths.

Coverage Coverage to be calculated. Select coverage by

clicking the check boxes. The options are as

follows:

● Product Coverage: A larger value indicates

higher product coverage.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 91 of 124

Parameter Description

● User Coverage: A larger value indicates higher

user coverage.

NOTE This parameter is not displayed for third-party data evaluation.

Variety Variety to be calculated. Select variety by clicking

the check boxes. The options are as follows:

Mean Hamming distance: A larger value

indicates higher variety.

Novelty Novelty to be calculated. Select novelty by

clicking the check boxes. The options are as

follows:

Mean degree: A smaller value indicates higher

novelty.

Recommendation Algorithm Type The parameter value automatically inherits the

setting from the upper-level node and cannot be

changed.

NOTE This parameter is not displayed for third-party data evaluation.

User Field ● For non-third-party data evaluation, the

parameter value automatically inherits the

setting from the upper-level node and cannot be

changed.

● For third-party data evaluation, select a value

from the drop-down list box.

Item Field ● For non-third-party data evaluation, the

parameter value automatically inherits the

setting from the upper-level node and cannot be

changed.

● For third-party data evaluation, select a value

from the drop-down list box.

Rating Field ● For non-third-party data evaluation, if the

prepositional node does not have rating fields,

this parameter is not displayed, otherwise, this

parameter value automatically inherits the

setting from the upper-level node and cannot be

changed.

● For third-party data evaluation, select a value

from the drop-down list box.

Use Partition ● For non-third-party data evaluation, the

parameter value automatically inherits the

setting from the upper-level node and cannot be

changed.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 92 of 124

Parameter Description

● This parameter is not displayed for third-party

data evaluation.

Visualization

You can find the evaluation files generated by the RecommenderEvaluation node in the Project

name/Evaluation directory, as shown in Figure.

Personalized recommendation evaluation files

DataAudit Node

Function

The DataAudit calculates statistics such as roles, ranges, and minimum/maximum values for fields and

generates evaluation files in the Analytic directory.

Restriction

the DataAudit node can be connected to source(ImportText, ImportFeatureLibrary or ImportDatabase) node

or data preprocessing node.

Parameter Description

Parameters

Parameter Description

DataAudit File Exported evaluation file name. Evaluation files are

stored in the Project name/Analytic directory.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 93 of 124

Parameter Description

Overlay Field Overlay field to be calculated. The DataAudit node

calculates the correlation between the analysis and

overlay fields, correlation verification statistics,

and the degree of freedom of correlation

verification statistics.

Statistics Grade Statistics grade.

The options are as follows:

● Base

Basic statistics, such as roles, ranges, and

minimum/maximum values of fields.

● Advance

Advanced statistics, such as correlation,

covariance, standard deviation, and deviation

between the analysis and overlay fields.

Visualization Scenario

Assume that the DataAudit node needs to calculate the basic and advanced statistics for the ID, AGE,

SEX, REGIONand INCOME fields in a batch of user data.

Visualization Result

The DataAudit node will generate data audit files in the Project name/Analytic directory, as shown in

0 and 0.

− The Base Statistics tab page is displayed as follows:

Base Statistics

− The Advance Statistics tab page is displayed as follows:

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 94 of 124

Advance Statistics

Statistics Node

Function

The Statistics calculates statistics such as counts, means, minimum/maximum values, and sums for fields of

the value type and generates statistics files in the Analytic directory.

Restriction the Statistics node can be connected to source(ImportText, ImportFeatureLibrary or ImportDatabase)

node or data preprocessing node.

The system calculates the correlation counters between Statistics Field and Correlated Field only

when Correlated Field is set and the Variance, Satandard Deviation or SEM statistical item is

selected.

The values of Statistics Field and Correlated Field must be of the numeral type.

Parameter Description

Parameters

Parameter Description

Statistics File Exported evaluation file name. Evaluation files are

stored in the Preject name/Analytic directory.

Correlated Field Correlated field between which and the statistics

field the correlation is to be calculated.

Visualization Scenario

Assume that the Statistics node needs to calculate the statistics of the smoking field and the

correlation between the smoking and isill fields in a batch of user data.

Visualization Result=

The Statistics node will generate statistics files in the Project name/Analytic directory as shown in

Figure.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 95 of 124

Figure Statistics

Mean Node

Function

The Mean node compares the means between group fields and test fields, or between other correlated field

pairs to check whether remarkable differentials exist.

Restriction The node must follow a source(ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

The node can follow a data preprocessing node.

Parameter Description

Parameters

Parameter Description

Means Comparison File Exported evaluation file name. Evaluation files are stored in

the Project name/Analytic directory.

Compare Type Mean comparison mode.

The options are as follows:

● Between groups

Grouping Field: Group data based on the group field.

Test Fields: Calculate mean-related statistics by group.

● Between field pairs

Calculate statistics for the means for the fields in a field

pair and the mean difference between the fields in the pair.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 96 of 124

Visualization Scenario 1

Assume that the Mean node needs to calculate the statistics for the means of the RANGE1 fields

grouped by the SET1 field in a batch of user data. The Mean node will generate mean comparison

files in the Project name/Analytic directory, as shown in Figure.

Figure Between groups

Scenario 2

Assume that the Mean node needs to calculate the statistics for the means of the RANGE1 and

OUTPUT fields. The Mean node will generate mean comparison files in the Project name/Analytic

directory, as shown in Figure.

Figure Between field pairs

Distribution Node

Function

The distribution node collects statistics on field value distribution.

Restriction The node must follow a source (ImportText, ImportFeatureLibrary or ImportDatabase) node and a

Type node.

The node can follow a data preprocessing node except the FeatureSelection node.

Parameter Description

Parameters

Parameter Description

Display Style Data distribution display mode. The options are as

follows:

● Graph

Displays data distribution in graphs.

● Data File

Displays the number of each field in exported data

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 97 of 124

Parameter Description

files.

NOTE If you want to use this mode, configure the TextFileExport node before the Distribution node.

Distribution Graph Name Name of the data distribution graph file.

NOTE When the Graph display mode is selected, data distribution

files are stored in the Analytic directory.

When the Data File display mode is selected, exported files are stored in the Data directory.

Analysis Field Field to be analyzed. For example, you can click

Settings, and set Analysis Field to Age.

Exchange Field Field to be evaluated. For example, if you want to analyze

the gender distribution by age segment and view the

distribution information in a graph, you can set Exchange

Field to Gender.

Details Analysis field details. The parameter is valid only when

Display Style is set to Graph.

Analysis field details include:

● Numbers per Segment

If the analysis field is of the Range role, you need to

set the parameter. The default value is 25. For

example, if you set Analysis Field to Age, Age ranges

from 1 to 100, and Numbers per Segment is 25,

gender distribution is displayed by age segment such

as 1 to 4, 5 to 8, and 9 to 12.

● Title

Distribution graph title.

● X-axis

Title of the X axis in the distribution graph. The title is

the same as the value of Analysis Field.

● Y-axis

Title of the Y axis in the distribution graph. The

default value is COUNT.

Delete Deletes an analysis field.

Visualization Scenario

Assume that you want to view the gender distribution by age segment in a batch of user data

containing the Age and Sex fields.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 98 of 124

You need to set Display Style to Graph. Set Analysis Field to Age. The age field must be of the Set

role. Set Exchange Field to Sex.

Visualization Result

The distribution graph will be generated in the Project name/Analytic directory, as shown in Figure.

Distribution graph

In the preceding figure, the X axis indicates the age, the Y axis indicates the user count, the bar in

yellow indicates the female user count, and the bar in blue indicates the male user count. You can view

the gender distribution clearly from the graph.

Correlate Node

Function

On the Correlate node, it is recommended that you use output fields to be modeled as forecast fields and

use other fields as analysis fields. After the process defined on the Correlate node is executed, the fields

that correlate strongly with the output fields can be identified. These identified fields can be used as input

fields in the modeling process to build more accurate models.

The Correlate node generates analysis files in the Analytic directory. You can check the correlation analysis

file list by clicking Analytic. In the directory, you can compare analysis results between samples.

(Not required by AUC) The system converts fields of the Sign role to the Set role for processing.

Algorithms applicable to fields of the Set role also apply to fields of the Sign role.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 99 of 124

Restriction The Correlate node can be connected to a source node (ImportText, ImportFeatureLibrary or

ImportDatabase) or data preprocessing node.

The Data Type of the field cannot be Time, Date, or Timestamp.

Parameter Description

Parameters

Parameter Description

Correlate File Exported analysis file name. Analysis files are

stored in the Project name/Analytic directory.

Prediction Field Forecast field. Select an option from the

drop-down list box.

The parameter value cannot be the same as the

value of Analysis Field.

Correlation Significance Counter You can use the Chi-square test and Anova test to

analyze the correlation between fields. 0 describes

the calculation of the counters.

Correlation Strength Counter Error decrease rate, including five counters. The

parameter indicates the forecast error rate that can

be decreased when the value of Prediction Field is

forecasted based on a given Analysis Field value.

0 describes the calculation of the four counters.

NOTE When the system calculates the Pearson correlation

coefficient between two fields, it also generates the distance between them, which also indicates the correlation between the two fields. You can select a counter to calculate the preceding information based on the field value type.

The difference between significance analysis and strength analysis is that the former calculates qualitative correlation between fields and the later calculates quantitative correlation between fields.

Correlation counter calculation description

Distance Analysis Role Prediction Field Role Field Type

Chi-squared Set Set Integer, real, and string

Anova Set Range Integer, real, and string

Range Set

Tau Set and Range Set Integer, real, and string

Eta Set Range Integer, real, and string

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 100 of 124

Distance Analysis Role Prediction Field Role Field Type

Pearson Range Range Integer and real

Spearman Set Set Integer and real

Range Range

Auc Sign, Set and Range Set (of two values) and Sign Integer, real, and string

Mae Range Range Integer and real

Mse Range Range Integer and real

Rmse Range Range Integer and real

Visualization Scenario

Assume there is a batch of data shown in 0. The node will analyze the correlation between the first

three fields in 0 and the last RANGE field.

Random data

SET NUM SIZE RANGE

1 1 a 120

2 2 b 119

3 3 c 118

4 4 a 117

1 5 b 116

2 6 b 115

... ... ... ...

4 120 c 1

Visualization Result

You can find the analysis files generated by the Correlate node in the Project name/Analytic directory,

as shown in 0.

Correlation analysis files

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 101 of 124

Export Nodes

An Export node can export processed data into a file or a table in the database.

1.1.1.1 ExportText Node

Function

The ExportText node exports data or models generated by other nodes.

Restriction The node must follow a ImportText, ImportFeatureLibrary or ImportDatabase Node and a Type node.

Related data has been imported.

Parameter Description

Parameters

Parameter Description

File System Type File system to store exported text files.

The options are as follows:

● Local

● HDFS

● FTP

NOTE You need to enable the FTP service before selecting a data file. For details, see Configuring the FTP Service.

File to Export Exported file name. Expressions are supported, for example,

sm_user_retain_#date(yyyyMMddHHmmss)#.csv.

Exported files are stored in the Project name/Data directory. By

default, CSV files are exported from the

${SmartMiner_HOME}/smartminer/Projects/test/Data directory.

NOTE

In the preceding directory, test indicates the name of the project where the process is located.

ExportDatabase Node

Function

The ExportDatabase node writes data to the database.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 102 of 124

Restriction The ExportDatabase node follows a PCAFactor node and a PCAFactor model exists.

The ExportDatabase cannot follow an SNSRS, FullConnected, OverlapNeighbor node, or

DiscriminationTree node.

The ExportDatabase cannot follow a TextClassifyApply node.

Parameter Description

Parameters

Parameter Description

Database Database name. Select a currently available database from the

drop-down list box. Oracle and DB2 databases are supported.

Schema Table mode in the database. The default value of the table mode is

the name of the created schema. Select a value from the drop-down

list box.

For example, if database user U1 creates tables T1 and T2 and user

U2 creates table T3 in the database, the options of the table mode in

the database are U1 and U2. When you select a mode, only the tables

of the selected mode are displayed.

Table Database table name. Select a value from the drop-down list box.

Load Type Mode for loading data.

The options are as follows:

● Insert

Data is inserted into the database.

● Update

Existing data in the database is updated. You can search for a

record in the database by keyword. If the record exists, the system

updates the record. If the record does not exist, no operation will

be performed.

● Replace

All records in a table are replaced.

● Update or insert

You can search for a record in the database by keyword. If the

record exists, the system updates the record. If the record does not

exist, the record is inserted into the database.

Target Field Name of a field in the database.

Data Type Type of a field in the database.

Source Field Field that matches a field in the database.

Key Field Specifies whether a field is a key field. Records of the same key field

will be updated, and records of different key fields will not be

updated.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 103 of 124

NLP

NLP is short for Natural Language Processing. The NLP node provides various text information mining

functions.

1.1.1.1 Segment

Function

Chinese segmentation refers to the process of dividing written Chinese text into meaningful words based on

specific rules, that is, converting the original unstructured text into structured information that computers

can process.

Algorithms on the Segment node are based on the Ansj framework, which is the Java version of the

ICTCLAS (Institute of Computing Technology, Chinese Lexical Analysis System). The SmartMiner has

implemented the parallel computing capability for segmentation, improving the segmentation speed and

accuracy.

Restriction The node must follow an ImportText node and a Type node.

The Type node must have at least two input fields: Primary Key and Input.

The segmentation result is exported using the ExportText node.

Model Input Example

The following input text is movie comments from a video website:

In the input text, you can use id and review_content as input fields. Set Direction of id to Primary Key,

and set Direction of review_content to Input.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 104 of 124

Parameter Description

Parameters

Parameter Description

Part-of-Speech Tagging Indicates whether to enable the part-of-speech tagging

function. If this function is enabled, the system

identifies the part of speech of each word and displays it

in the segmentation result. For details about the part of

speech definition in Ansj, see Part of Speech Codes.

If this function is disabled, the system will not identify

the part of speech for words. In this case, the function of

reserving parts of speech is disabled.

Reserve Part of Speech Specifies words to be reserved in the segmentation

result. This parameter is valid only when

Part-of-Speech Tagging is set to Yes. The options of

this parameter are as follows:

● All: reserves words of each part of speech.

● Define: reserves words whose part of speech is

configured in Part of Speech.

Part of Speech Defines parts of speech of words to be reserved in the

segmentation result. This parameter is valid only when

Reserve Part of Speech is set to Define.

For example, if Adjective and Noun are selected, the

segmentation result contains only adjectives and nouns.

Transform Complex Font Indicates whether to perform segmentation for

traditional Chinese. If the input text contains traditional

Chinese, set this parameter to Yes.

Remove Stop Words Indicates whether to remove stop words. Stop words

indicate words that cannot reflect topics, for example,

conjunctions such as then and therefore. Stop words

interference keyword extraction and the system filters

out them by default.

If this parameter is set to Yes, the system filters out stop

words, including words indicating object features or

categories, conjunctions, exclamations, onomatopoetic

words, prepositions, auxiliary word, adjectives

indicating status, pronouns, and punctuation.

Output Result Example

Model Input Example shows the input text, and 0 describes the process configuration.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 105 of 124

Segmentation process

0 shows the configuration of the Segment node.

Segment node configuration

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 106 of 124

The following shows the segmentation result:

User-defined Dictionary

You can replace user-defined dictionaries in directories on the SmartMiner server. 0 describes the

directories.

NOTE Back up the original file before replacing a user-defined dictionary file.

User-defined dictionary directories

Dictionary File Directory Format Description

default.dic ${HOME}/data/nlp/

ansj/custom

Word Part of Speech Frequency

For example:

Longguan station n 2

Longyu nz 2

Hengda nt 20

Defines service-related

feature words that users

concern with. If the

segmentation result

does not contain the

words, they can be

added to the

user-defined dictionary.

newWordFilter.dic ${HOME}/data/nlp/

ansj/stopword

Stop word 1

Stop word 2

...

Defines stop words to

be filtered out.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 107 of 124

Part of Speech Codes

Part of speech codes

Code Part of Speech Code Part of Speech Code Part of Speech Code Part of

Speech

a Adjective i Idiom nz Other dedicated

noun

uj Auxiliary

word

ad Adverb j Acronyms and

abbreviations

o Onomatopoeia ul Conjunction

ag Adjective

morpheme

k Subsequent element p Preposition uv Conjunction

an Adjective that

functions as a

noun

l Common words q Quantifier uz

b Word indicating

the object feature

or category

m Numeral r Pronoun v Verb

bg Morpheme of the

word indicating

the object feature

or category

mg Numeral morpheme rg Pronoun

morpheme

vd Adverb that

functions as

an adverbial

modifier

c Conjunction n Noun s Space-related

word

vg Verb

morpheme

d Adverb ng Noun morpheme t Time-related

word

vn Verb that

functions as

a noun

dg Adverb

morpheme

nr Name tg Time-related

word morpheme

w Punctuation

e Exclamation ns Area name u Auxiliary word y Interjection

f Position-related

word

nt Organization-related

word

ud Auxiliary word yg Interjection

morpheme

h Preceding

element

nx Letter-based dedicated

nouns

ug

TrajectoryAnalysis Node

The TrajectoryAnalysis node analyzes customers' trajectory. Trajectory analysis includes stay point analysis,

permanent location analysis, similar trajectory analysis, and real-time trajectory analysis.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 108 of 124

1.1.1.1 StayPointAnalysis Node

Function

In a trajectory, some points denote locations where people have stayed for a while, such as the shopping

malls, tourist attractions, or gas stations. These kinds of points are called stay points. There are two types of

stay points. One is a single point location, such as Stay Point 1 in Figure. This situation is very rare,

because a user's positioning device usually generates different readings even in the same location. The other,

such as Stay Point 2, is more generally observed in trajectories, representing the places where people move

around. The clustering algorithm of the StayPointAnalysis node can calculate the longitude and latitude of

the stay points.

Stay point examples

Restriction The node must follow an ImportText node and a Type node.

The input fields of the node must include User ID, BTS ID, Timestamp, Longitude, and Latitude.

The stay point analysis result is exported using the ExportText node.

Model Input Example

The input text is a user' trajectory.

User_id,BTS_id,Timestamp,Longitude,Latitude

user1,BS1,1431000000,120.0000,30.0000

user1,BS2,1431000100,120.0030,30.0040

user1,BS3,1431002100,120.0030,30.0000

user1,BS4,1431002110,120.0230,30.0000

user1,BS5,1431002210,120.0030,30.0010

user1,BS6,1431002212,120.0040,30.0030

user1,BS7,1431003212,120.0100,30.0100

user1,BS8,1431003312,120.1000,30.1000

user1,BS9,1431003412,120.0100,30.0100

user1,BS10,1431006000,120.0100,30.0100

user1,BS12,1431007000,120.0140,30.0130

user2,BS9,1431003412,120.0100,30.0100

user2,BS10,1431004000,120.0100,30.0100

user2,BS11,1431004100,120.0140,30.0130

user3,BS9,1431003412,120.0100,30.0100

user3,BS10,1431006000,120.0100,30.0100

user3,BS11,1431007000,120.0140,30.0130

user4,BS12,1431001412,120.0101,30.0101

user4,BS12,1431003412,120.0101,30.0101

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 109 of 124

user4,BS13,1431003482,120.0400,30.0400

user4,BS13,1431003562,120.0400,30.0400

user4,BS14,1431003632,120.0400,30.0400

In the preceding information, Direction of the User_id, BTS_id, Timestamp, Longitude, and Latitude

fields on the Type node must be set to Input.

Parameter Description

Parameters

Parameter Description

User ID Maps the field name of an input field. For example, this parameter can

be set to User_id or Field0 for User ID (the headers of the input file

are not read).

The five fields record users' trajectories. Each record in the input file

indicates that a BTS (specified by BTS_ID) obtains a user's (specified

by User_ID) longitude and latitude (specified by Longitude and

Latitude respectively) at a specific time (specified by Timestamp).

BTS ID

Timestamp

Longitude

Latitude

Tolerance Distance (m) Tolerance distance. If the distance between two stay points is shorter

than the value of this parameter, the system considers that the two stay

points are the same stay point.

Min Time Span (h) Minimum time span for stay point detection. If the duration that a user

stays at a location is longer than or equal the value of this parameter,

the system considers the location as the stay point.

Speed Threshold (m/s) Threshold of users' moving speed between two stay points. If a user's

moving speed is faster than the value of this parameter, the system

considers that the two stay points are abnormal points. Abnormal stay

points are removed from the output data.

Abnormal points contain error data or data that is not applicable to

analysis scenarios. For example, to obtain the stay points when users

walk or drive a car, set this parameter to 50 m/s (180 km/s).

Max BTS Distance (m) Maximum distance between two adjacent BTSs.

● If BTSs are used to collect information for calculating users'

location information, the value of this parameter must be greater

than the value of Tolerance Distance (m).

● If BTSs are not used to collect information for calculating users'

location information, this parameter can be set to the value of

Tolerance Distance (m)

Output Result Example

Model Input Example shows the input text, and Figure describes the process configuration.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 110 of 124

Stay point analysis process

Figure shows the configuration of the StayPointAnalysis node.

StayPointAnalysis node configuration

The following shows the output of the StayPointAnalysis node:

User_ID,Staying Start Time,Staying End Time,Center Longitude,Center Latitude,Virtual Center

Longitude,Virtual Center Latitude,BTS_ID

user3,1431003412,1431007000,120.01,30.01,120.01133333333333,30.011,BS10

user1,1431000000,1431002212,120.003,30.001,120.0025,30.001,BS5

user1,1431003412,1431007000,120.01,30.01,120.01133333333333,30.011,BS10

user4,1431001412,1431003632,120.0101,30.0101,120.0101,30.0101,BS12

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 111 of 124

In the preceding information:

The Virtual Center Longitude and Virtual Center Latitude fields indicate the longitude and latitude

of the center of the range where users stay respectively. The Center Longitude and Center Latitude

fields indicate the longitude and latitude of the BTS nearest to the virtual center respectively.

Each record in the output indicates that a user (specified by User_ID) stays at a virtual center

(specified by Virtual Center Latitude and Virtual Center Longitude) for a period (specified by

Staying Start Time and Staying End Time).

Points That You Might Be Interested In

Cascading Models

Function

If you want to build a model after it is applied, you do not have to import source data again. Instead, use an

Application node (TextClassifyApply node excluded) as the source node in the new modeling process. The

Application node can be followed by a Type, DataAudit, Statistics, GraphVisualize, Correlate, or Filter

node.

Process Instance

Process instance

Several fields will be created after the Application node is executed, and these fields will be displayed in

the nodes cascaded to the Application node, such as the Type and DataAudit nodes. In this example, the

FORCAST_CATEGORY, PROB_yes, and PROB_no fields are created after the NaiveBayesApply node

is executed. Of the fields, yes and no are two options of an output field.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 112 of 124

Real-Time Query

You can query process information in real time in the system.

Procedure

Choose Developer > Analysis Miner > Smartminer > Real-Time Query.

Configure the search criteria to query process information.

Figure shows the page for querying process information in real time.

Real-time Info page

Parameters on the page are described as follows:

Parameters on the Real-time Info page

Parameter Description

ProjectPath Path or name of the project to be queried. Fuzzy

query is supported.

Process Name of the process to be queried. Fuzzy query is

supported.

Start Time Start time of the start time range to be queried.

To End time of the start time range to be queried.

Stop Indicates whether to stop the running process that

is queried.

Refresh Interval Interval for refreshing the query result. The default

value is 2, in seconds. The options are 2, 5, 10, and

20.

Process Status Running status of the process to be queried. The

value Executing... indicates that the process is

running properly.

End Time Processes queried in real time are running

processes. As a result, this parameter is

unavailable.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 113 of 124

History Query

You can query historical process information in the system.

Procedure

Choose Developer > Analysis Miner > Smartminer > History Query.

Configure the search criteria to query process information.

Figure shows the page for querying historical process information.

Historical Info page

Parameters on the page are described as follows:

Parameters on the Historical Info page

Parameter Description

ProjectPath Path or name of the project to be queried. Fuzzy

query is supported.

Process Name of the process to be queried. Fuzzy query is

supported.

Process Status Status of the process to be queried. The options are

as follows:

● Executing...

● Completed

● Stopped

● Failed

● Stopping...

Start Time Start time of the process start time range to be

queried.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 114 of 124

Parameter Description

To End time of the process start time range to be

queried.

End Time Start time of the process termination time range to

be queried.

To End time of the process termination time range to

be queried.

Process Status Running status of the process to be queried.

Start Time Start time of the process to be queried.

End Time End time of the process to be queried.

Configuring the FTP Service

To select files from the FTP server, you need to enable the FTP service.

Procedure

Modify the ${HOME}/conf/smartminer.properties file of the SmartMiner user.

A configuration example is as follows:

smart_ftp_ip=10.41.28.33

smart_ftp_port=22

smart_ftp_username=username

smart_ftp_input_dir=/home/ftphome

smart_ftp_output_dir=/home/username

smart_ftp_keytype=0

smart_ftp_password=ytpnga9zY0GwAy/G0mP6FA==

smart_ftp_keypath=

smart_ftp_passphrase=

Parameter Description and Setting

smart_ftp_ip IP address of the FTP server.

smart_ftp_port SSH port number on the FTP server. The 22 port is used by

default. That is, the SFTP protocol is used to connect to the

FTP service.

smart_ftp_username User name for logging in to the FTP server.

smart_ftp_input_dir Directory on the FTP server where the file to be read is

located. The directory must start with a slash (/).

smart_ftp_output_dir Directory on the FTP server to which a file is to be

exported. The directory must start with a slash (/).

smart_ftp_keytype Key type.

● 0: password

● 1: private key

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 115 of 124

Parameter Description and Setting

smart_ftp_password If the password mode is used, use the script in the

${HOME}/tools/interactive_encrypt.sh directory of the

SmartMiner user to encrypt the password.

smart_ftp_keypath If the private key mode is used, set the full path for the

private key file.

smart_ftp_passphrase If the private key mode is used, you need to encrypt the

private key file. The encryption method is the same as that

of the password mode.

Run the restart-ide.sh commands to restart the SmartMiner for the settings to take effect.

Importing and Exporting a Project

You can import or export a project package in the SmartMiner console.

Context

Project importing and exporting share one temporary file storage directory. You can customize the

temporary file storage directory by configuring smart_temp_dir in

${HOME}/conf/smartminer.properties.

Importing a Project

Right-click Project in the navigation tree on the left and choose Import.

The dialog box shown in Figure is displayed.

Importing a project

Click to select a package to import.

NOTE The size of the package cannot exceed 100 MB. If the name of the package to import already exists in the destination directory, change it.

Click Import.

Exporting a Project

1. Right-click Project in the navigation tree on the left and choose Export.

The dialog box shown in Figure is displayed.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 116 of 124

Exporting a project

Select the data file to be exported, Click Export.

The system generates a project package. The name format of the project package is project

name_timestamp.zip, and the format of the timestamp is YYYYMMDDHHMMSSMMM. After the

project package is compressed, the temporary file will be deleted automatically, and a dialog box is

displayed for the users to export the project package.

NOTE The size of the package cannot exceed 200 MB.

Feature Management

1.1.1.1 Managing Features Files

You can configure files used for creating features on the SmartMiner GUI.

Prerequisites

A feature file has been uploaded to the HDFS.

Procedure

Choose Developer > Analysis Miner > Smartminer > Data Files.

Right-click a directory in the navigation tree on the left and choose Create File from the shortcut menu.

The dialog box for adding a feature file is displayed.

Set the parameters.

0 describes the parameters.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 117 of 124

Parameters

Parameter Description

File ID The description of the file.

File Type The default value is HDFS.

File Path Directory where a feature file is stored.

Replace the last part of the path that indicates the

month with an asterisk (*), for example,

/DataStoreage4Feature/Customer/u2f2/*

File Name Name of the feature file to be added. You can click

Obtain File Fields to obtain fields in the feature

file and then select a primary key. Sample data

extraction indicates the process of an

ImportFeatureLibary node combines forecast fields

and input fields of the same primary key.

Corresponding Primary Key Primary key of the feature file to be added.

Object Type Object type associated with the feature file to be

added.

NOTE The file separator must be consistent with the actual file separator.

Files referenced by features cannot be deleted.

Managing Features

You can create, import, and export features on the SmartMiner GUI.

Prerequisites

A feature file has been defined.

Context The following two parameters need to be added to the

${SmartMiner_HOME}/conf/smartminer.properties file:

smart_feature_maxstoremonth=12

smart_feature_data_rootpath=hdfs://133.34.223.46:8920

The preceding configuration indicates that the maximum data storage duration is 12 months and data

is stored in the HDFS.

Feature data is stored by month with the same data view. A large data file can be divided in to multiple

small files by user and feature. Currently, a large file can be divided into multiple small files by user

for features. All small files use the same primary key, for example, user ID.

A small file contains only data, and the file name is in userfeature_part*_part* format, in which the

first part* indicates the user division index and the second part* indicates the feature division index.

For example, a large file that contains features of 1,000,000 users (including the age and gender). The

large file can be divided into the following files:

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 118 of 124

− userfeature-part1-part1: stores the age information of 500,000 users; userfeature-part2-part1:

stores the age information of another 500,000 users. The common file name is

userfeature-part?-part1.

− userfeature-part1-part2: stores the gender information of 500,000 users;

userfeature-part2-part2: stores the gender information of another 500,000 users. The common

file name is userfeature-part?-part2.

Field name information is saved in a header file that contains only one line. The header file name is in

userfeature_part*.head format. Small files for a feature shares one header file.

Creating a Feature

Choose Developer > Analysis Miner > Smartminer > Feature Manager.

Click Create Feature.

The dialog box for creating a feature is displayed.

Set the parameters.

0 describes the parameters.

Parameters

Parameter Description

Feature ID File ID. You can click Verify Uniqueness to check whether a feature

ID is unique.

Statistics Period The default value is Month. Currently, the parameter can only be set to

Month.

Associated File File referenced by the feature to be created.

File Fields Fields used for creating features in a feature file.

Value Type Data type of the file field.

Value Range Value range of the data type of the primary key.

● If Value Type is set to Range, Lower Limit and Upper Limit

need to be set.

● If Value Type is set to Sign, Flag Value 1 and Flag Value 2 need

to be set.

● If Value Type is set to Set, Value Range does not need to be set.

Click OK.

The newly created feature is displayed in the feature list.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 119 of 124

NOTE A field of a file can be used to create only one feature.

Only enabled features can be used to abstract sample data.

Features referenced by an ImportFeatureLibrary node cannot be deleted.

Importing Features

1. Right-click the root directory in the navigation tree on the left and choose Import Feature from the

shortcut menu.

Click and select the feature package to be uploaded.

Click Import.

NOTE The feature package to be uploaded cannot exceed 100 MB.

Exporting Features

1. Right-click the directory from which features need to be exported in the navigation tree on the left and

choose Export Feature from the shortcut menu.

Click Export.

NOTE Information including file definition (file and field definitions included), category directory definition, and feature definition is exported to a package.

Other Common Operations

Click to delete a feature.

Click

to edit a feature.

Click to view details of a feature.

Click to enable a feature.

Click to disable a feature.

Model Management

1.1.1.1 Creating a Theme

A theme is a specific service topic for example, the deregistration analysis theme. A theme can contain

multiple data mining models created based on the same target.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 120 of 124

Procedure

Choose Developer > Analysis Miner > Smartminer > Model Manager.

Right-click the root node of the navigation tree on the left and choose Create Theme from the shortcut menu.

The dialog box for creating a theme is displayed.

Set the parameters.

0 describes the parameters.

Parameters

Parameter Description

Theme ID Theme ID. You can click Verify Uniqueness to

check the uniqueness of a theme ID.

Theme Name Theme name.

Theme Category Theme category. Currently, the following theme

categories are supported: Value forecast,

classification, and clustering.

Evaluation Interval Interval between executions of the theme

evaluation scheduled task.

● When Scheduling Type is None, scheduled

task is not need to set.

● When Scheduling Type is Update

Interval,Second is need to set.

● When Scheduling Type is Every Day, Hour

and Minute are need to set.

● When Scheduling Type is Every Month,

Date, Hour and Minute are need to set.

Evaluation Timeout Interval (minutes) Timeout interval of the theme evaluation

scheduled task.

Application Interval Interval between executions of the theme

application scheduled task.

● When Scheduling Type is None, scheduled

task is not need to set.

● When Scheduling Type is Update

Interval,Second is need to set.

● When Scheduling Type is Every Day, Hour

and Minute are need to set.

● When Scheduling Type is Every Month,

Date, Hour and Minute are need to set.

Application Timeout Interval (minutes) Timeout interval of the theme application

scheduled task.

Evaluation Counter Evaluation counter of a theme. You can click the

entry box and select theme evaluation counters in

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 121 of 124

Parameter Description

the dialog box that is displayed.

Threshold Evaluation counter threshold.

Click OK.

The newly created theme is displayed in the navigation tree on the left.

Follow-up Procedure

After a theme is created, you can perform the following operations:

1. Click the theme to view theme details. The theme details are displayed on the right of the page.

Right-click the theme and choose Import Model from the shortcut menu to import a model. For details about how

to import a model, see Importing a Model.

Delete the theme.

a Suspend all models under the theme.

b Delete all models under the theme.

c Suspend the theme.

d Delete the theme.

Importing a Model

After a theme is created, you can import a model under the theme. During model import, you need to

specify a model file and a process file and an evaluation file that are associated with the model.

Procedure

Right-click a theme and choose Import Model from the shortcut menu.

Enter basic information, including model ID and name.

Click OK.

Select a model file, process file, and an evaluation file.

Click Next.

Configure a data source.

For details about how to configure a data source, see ImportText Node.

Click Next.

Configure fields.

For details about how to configure fields, see Creating an Auto Theme Process.

Click Complete.

NOTE

You can import multiple models for a theme. After the models are configured, the SmartMiner automatically and periodically evaluates the models based on the scheduled task configured in the associated theme, and selects and executes the optimal model.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 122 of 124

FAQs

Predictive Analytics Module Fault Diagnosis Methods

Symptom

Common failure symptoms are as follows:

Failed to jump to the SmartMiner GUI from the Portal.

Failed to execute processes.

Operations performed on the GUI cannot be submitted.

Possible Causes

Common fault causes

Symptom Possible Cause

Failed to access the SmartMiner GUI. ● The SmartMiner cannot be started because the

startup port is occupied.

● The SmartMiner service has not been registered

to the SLB.

Failed to execute processes. ● The flow configuration is incorrect:

– The data field types are different from the

actual types.

– The data input and output types do not

meet the modeling requirements.

● The communication between systems is

abnormal:

– The IP address used for communication

between the SmartMiner and Hadoop is

not in the SmartMiner whitelist.

– At the site, the IP address segment of the

management plane is separated from that

of the services plane. The IP address used

for communication between the

SmartMiner and Hadoop is an IP address

of the management plane and cannot

communicate with the services plane.

– The communication between the

DataNode in the Hadoop and the Hadoop

cluster is faulty.

● The database is abnormal. The metadata

database is locked and the SmartMiner cannot

read data from and write data to the database.

● The system resource is insufficient. The

memory and kernel resources of the Hadoop

cluster is not sufficient for the execution of

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 123 of 124

Symptom Possible Cause

processes.

Operations performed on the GUI cannot be

submitted.

The browser is not compatible.

Failure to Access the SmartMiner GUI

1. Log in to the SLB and check whether the SmartMiner service has been correctly registered.

Choose Monitor > Servers. Normally, a SmartMiner node is online.

If no node is online, view the tomcat startup and stopping logs of the SmartMiner to check whether the

SmartMiner is normally started.

Check the startup and stop information in the Tomcat startup and stopping log to check whether the SmartMiner is

properly started.

a Log in to the SmartMiner server as the SmartMiner installation user.

b View the tomcat startup and stopping logs ${HOME}/tomcat/logs/catalina.out to check the port

number is occupied.

If the logs indicate that the SmartMiner is not properly started and the port number is occupied,

port already exits.

Change the port number in the ${HOME}/conf/universe.config.properties file to an idle port

number.

http_port=9380

https_port=9343

Restart the SmartMiner for the configuration to take effect.

% stop_all.sh

% start_all.sh

If the password of the Oracle database is changed recently, check whether the Oracle database password in the

SmartMiner has been changed accordingly.

For details about how to change the password in the SmartMiner configuration file, see

Maintenance > Password Change > Changing Oracle Database User Passwords in the product

documentation.

SmartMiner Process Execution Failure

1. Check whether the data field types are different from the actual ones.

Click the faulty node in a process and modify the node configuration according to the error code and

rectification suggestion.

Check whether the data input and output types do not meet the modeling requirements.

Verify the configuration of the Type node. For details about the model input data type requirements,

see chapter "Operation" in the product documentation.

If no, modify the input data and configuration of the Type node based on the requirements.

View the process execution information in the ${HOME}/logs/debug/smartminer_debug.log file:

− Check whether the Hadoop node whose IP address is in the error logs can properly communicate

with the Hadoop cluster.

Log in to the FusionInsight Manager and check the node connection status.

− Check whether the Hadoop IP address in the error logs belongs to the management plane.

SmartMiner INTERNAL

2017-09-07 Huawei confidential. No spreading without permission. Page 124 of 124

If yes, change the JobHistory address in the smartminer.hadoop.properties to the IP address on

the service plane and restart the SmartMiner service for the modification to take effect.

− Check whether the Hadoop IP address in the error logs has been added to the /etc/hosts on the

SmartMiner server. All IP addresses that need to communicate with the SmartMiner must be added

to the file.

Log in to the HOM JobHistory page and check whether any jobs failed to be executed. Diagnose the fault based

on the error information and rectify the fault.

The URL of the JobHistory monitoring page is http://IP address:Port number/jobhistory.

In the URL:

− IP address: Set it to the IP address of the JobHistory Server node.

− Port number: Set it to the port number configured during the installation.

Set the memory and kernel count required for node execution to smaller values in the SmartMiner configuration

file and execute the process again.

Modify the ${HOME}/conf/smartminer.hadoop.properties file based on the running Hadoop node,

and modify the ${HOME}/conf/smartminer.spark.properties file based on the running Spark node.

Failure to Submit Operations on the GUI

Configure browser compatibility. For details, see Operation > Unified Analytics Runtime Platform >

System Management > Getting Started > Basic Operations > Configuring the Browser in the product

documentation.

If the fault persists, collect the SmartMiner fault information and contact Huawei technical support. For

details about how to collect fault information, see Fault Information Collection.