50
Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Embed Size (px)

Citation preview

Page 1: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Machine Learning in PracticeLecture 3

Carolyn Penstein Rosé

Language Technologies Institute/ Human-Computer Interaction

Institute

Page 2: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Plan for Today Announcements

Assignment 2Quiz 1

Weka helpful hints Topic of the day: Input and Output More on cross-validation ARFF format

Page 3: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Weka Helpful Hints

Page 4: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Increase Heap Size

Page 5: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Weka Helpful Hint: Documentation!!

Click on Morebutton!

Page 6: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Output Predictions Option

Page 7: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Output Predictions Option

Important note: Because of the way Weka randomizes the data forcross-validation, the only circumstance under which you can matchthe instance numbers to positions in your data is if you have separate train and test sets so the order will be preserved!

Page 8: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

View Classifier Errors

Page 9: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Input and Output

Page 10: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Representations

Concept: the rule you want to learn

Instance: one data point from your training or testing data (row in table)

Attribute: one of the features that an instance is composed of (column in table)

Page 11: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Numeric versus Nominal Attributes What kind of reasoning does your

representation enable? Numeric attributes allow instances to be

ordered Numeric attributes allow you to measure

distance between instances Sometimes numeric attributes make too fine

grained of a distinction

.2 .25 .28 .31 .35 .45 .47 .52 .6 .63

Page 12: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Numeric versus Nominal Attributes

.2 .25 .28 .31 .35 .45 .47 .52 .6 .63

Numeric attributes can be discretized into nominal values Then you lose ordering and distance Another option is applying a function that maps a range

of values into a single numeric attribute

Nominal attributes can be mapped into numbers i.e., decide that blue=1 and green=2 But are inferences made based on this valid?

Page 13: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Numeric versus Nominal Attributes

.2 .25 .28 .31 .35 .45 .47 .52 .6 .63

.2 .3 .5 .6

Numeric attributes can be discretized into nominal values Then you lose ordering and distance Another option is applying a function that maps a range

of values into a single numeric attribute

Nominal attributes can be mapped into numbers i.e., decide that blue=1 and green=2 But are inferences made based on this valid?

Page 14: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Example!

Problem: Learn a rule that predicts how much time a person spends doing math problems each day

Attributes: You know gender, age, socio-economic status of parents, chosen field if any

How would you represent age, and why? What would you expect the target rule to look like?

Page 15: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Styles of Learning Classification – learn rules from labeled

instances that allow you to assign new instances to a class

Association – look for relationships between features, not just rules that predict a class from an instance (more general)

Clustering – look for instances that are similar (involves comparisons of multiple features)

Numeric Prediction (regression models)

Page 16: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Food Web

http://www.cas.psu.edu/DOCS/WEBCOURSE/WETLAND/WET1/identify.html

Page 17: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Food Web

http://www.cas.psu.edu/DOCS/WEBCOURSE/WETLAND/WET1/identify.html

What else would be affected if wheatwere to disappear?

Page 18: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Food Web

http://www.cas.psu.edu/DOCS/WEBCOURSE/WETLAND/WET1/identify.html

How would you represent this data?

Page 19: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Food Web

http://www.cas.psu.edu/DOCS/WEBCOURSE/WETLAND/WET1/identify.html

What would the learned rule look like?

Page 20: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Food Web

http://www.cas.psu.edu/DOCS/WEBCOURSE/WETLAND/WET1/identify.html

What would the learned rule look like?

Page 21: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Food Web

http://www.cas.psu.edu/DOCS/WEBCOURSE/WETLAND/WET1/identify.html

Page 22: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Food Web What if you wanted a more general rule: i.e., Affects(Entity1, Entity2)

http://www.cas.psu.edu/DOCS/WEBCOURSE/WETLAND/WET1/identify.html

Page 23: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Food Web What if you wanted a more general rule: i.e., Affects(Entity1, Entity2)

http://www.cas.psu.edu/DOCS/WEBCOURSE/WETLAND/WET1/identify.html

Page 24: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Food Web What if you wanted a more general rule: i.e., Affects(Entity1, Entity2)

http://www.cas.psu.edu/DOCS/WEBCOURSE/WETLAND/WET1/identify.html

122 rows altogether!Now let’s look at the learned rule….

Page 25: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Food Web What if you wanted a more general rule: i.e., Affects(Entity1, Entity2)

http://www.cas.psu.edu/DOCS/WEBCOURSE/WETLAND/WET1/identify.html

122 rows altogether!Now let’s look at the learned rule….

Page 26: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Food Web What if you wanted a more general rule: i.e., Affects(Entity1, Entity2)

http://www.cas.psu.edu/DOCS/WEBCOURSE/WETLAND/WET1/identify.html

122 rows altogether!Now let’s look at the learned rule…. Does it have to be this complicated?

Page 27: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Food Web

http://www.cas.psu.edu/DOCS/WEBCOURSE/WETLAND/WET1/identify.html

What would your representation for Affects(Entity1, Entity2) look like?

Page 28: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Food Web

http://www.cas.psu.edu/DOCS/WEBCOURSE/WETLAND/WET1/identify.html

What would your representation for Affects(Entity1, Entity2) look like?

Page 29: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Food Web

http://www.cas.psu.edu/DOCS/WEBCOURSE/WETLAND/WET1/identify.html

What would your representation for Affects(Entity1, Entity2) look like?

Page 30: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

More on Cross-Validation

Page 31: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Cross Validation Exercise

What is the same?What is different?

1 2

3 45

What surprises you?

Page 32: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Compare Folds with Tree Trained on Whole Set1 2

3 45

Page 33: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Train Versus TestPerformance on Training Data Performance on Testing Data

Page 34: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Which Model Do You Think Will Perform Best on Test Set?1 2

3 45

Page 35: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Fold 1

Page 36: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Fold 2

Page 37: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Fold 3

Page 38: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Fold 4

Page 39: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Fold 5

Page 40: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Total Performance

What do you notice?

Page 41: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Total Performance

Average Kappa = .5

Page 42: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Starting to think about Error Analyses

Step 1: Look at the confusion matrix Where are most of the errors occurring? What are possible explanations for systematic

errors you see? Are the instances in the confusable classes too similar

to each other? If so, how can we distinguish them? Are we paying attention to the wrong features? Are we missing features that would allow us to see

commonalities within classes that we are missing?

Page 43: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

What went wrong on Fold 3?1 2

3 45

Page 44: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

What went wrong on Fold 3?

Training Set Performance Testing Set Performance

Hypotheses?

Page 45: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

What went wrong on Fold 3?

Training Set Performance Testing Set Performance

Hypotheses?

Page 46: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

What’s the difference?

Page 47: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Hypothesis: Problem with first cut

Page 48: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

Some Examples

Page 49: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

What do you conclude?

Page 50: Machine Learning in Practice Lecture 3 Carolyn Penstein Rosé Language Technologies Institute/ Human-Computer Interaction Institute

What do you conclude?

Problem with Fold 3 was probably just a sampling fluke.Distribution of classes different between train and test.