1. Machine Learning Reading: Chapter 18

2. 2 Machine Learning and AI  Improve task performance through observation, teaching  Acquire knowledge automatically for use in a task  Learning as a key component in intelligence

3. 3 What does it mean to learn?

4. 4 Different kinds of Learning  Rote learning  Learning from instruction  Learning by analogy  Learning from observation and discovery  Learning from examples

5. 5 Inductive Learning  Input: x, f(x)  Output: a function h that approximates f  A good hypothesis, h, is a generalization or learned rule

6. 6 How do systems learn?  Supervised  Unsupervised  Reinforcement

7. 7 Three Types of Learning  Rule induction  E.g., decision trees  Knowledge based  E.g., using a domain theory  Statistical  E.g., Naïve bayes, Nearest neighbor, support vector machines

8. 8 Applications  Language/speech  Machine translation  Summarization  Grammars  IR  Text categorization, relevance feedback  Medical  Assessment of illness severity  Vision  Face recognition, digit recognition, outdoor scene recognition  Security  Intrusion detection, network traffic, credit fraud  Social networks  Email traffic  To think about: applications to systems, computer engineering, software?

9. 9 Language Tasks  Text summarization  Task: given a document which sentences could serve as the summary  Training data: summary + document pairs  Output: rules which extract sentences given an unseen document  Grammar induction  Task: produce a tree representing syntactic structure given a sentence  Training data: set of sentences annotated with parse tree  Output: rules which generate a parse tree given an unseen sentence

10. 10 IR Task  Text categorization http://www.yahoo.comwww.yahoo.com Task: given a web page, is it news or not?  Binary classification (yes, no)  Classify as one of business&economy,news&media, computer Training data: documents labeled with category Output: a yes/no response for a new document; a category for a new document

11. 11 Medical  Task: Does a patient have heart disease (on a scale from 1 to 4)  Training data:  Age, sex,cholesterol, chest pain location, chest pain type, resting blood pressure, smoker?, fasting blood sugar, etc.  Characterization of heart disease (0,1-4)  Output:  Given a new patient, classification by disease

12. 12 General Approach  Formulate task  Prior model (parameters, structure)  Obtain data  What representation should be used? (attribute/value pairs)  Annotate data  Learn/refine model with data (training)  Use model for classification or prediction on unseen data (testing)  Measure accuracy

13. 13 Issues  Representation  How to map from a representation in the domain to a representation used for learning?  Training data  How can training data be acquired?  Amount of training data  How well does the algorithm do as we vary the amount of data?  Which attributes influence learning most?  Does the learning algorithm provide insight into the generalizations made?

14. 14 Classification Learning  Input: a set of attributes and values  Output: discrete valued function  Learning a continuous valued function is called regression  Binary or boolean classification: category is either true or false

15. 15 Learning Decision Trees  Each node tests the value of an input attribute  Branches from the node correspond to possible values of the attribute  Leaf nodes supply the values to be returned if that leaf is reached

16. 16 Example  http://www.ics.uci.edu/~mlearn/MLSummary.ht ml http://www.ics.uci.edu/~mlearn/MLSummary.ht ml  Iris Plant Database  Which of 3 classes is a given Iris plant?  Iris Setosa  Iris Versicolour  Iris Virginica  Attributes  Sepal length in cm  Sepal width in cm  Petal length in cm  Petal width in cm

17. 17 Summary Statistics: Min Max Mean SD ClassCorrelation sepal length: 4.3 7.9 5.84 0.83 0.7826 sepal width: 2.0 4.4 3.05 0.43 -0.4194 petal length: 1.0 6.9 3.76 1.76 0.9490 (high!) petal width: 0.1 2.5 1.20 0.76 0.9565 (high!)  Rules to learn  If sepal length > 6 and sepal width > 3.8 and petal length < 2.5 and petal width < 1.5 then class = Iris Setosa  If sepal length > 5 and sepal width > 3 and petal length >5.5 and petal width >2 then class = Iris Versicolour  If sepal length 3 and petal length  2.5 and ≤ 5.5 and petal width  1.5 and ≤ 2 then class = Iris Virginica

18. Data S-lengthS-widthP-lengthP-widthClass 16.836.32.3Versicolour 273.92.42.2Setosa 3232.61.7Verginica 433.42.51.1Verginica 55.53.66.82.4Versicolour 67.74.11.21.4Setosa 76.34.31.61.2Setosa 813.72.82.2Verginica 964.25.62.1Versicolour

19. Data S-lengthS-widthP-lengthClass 16.836.3Versicolour 273.92.4Setosa 3232.6Verginica 433.42.5Verginica 55.53.66.8Versicolour 67.74.11.2Setosa 76.34.31.6Setosa 813.72.8Verginica 964.25.6Versicolour

20. 20 Constructing the Decision Tree  Goal: Find the smallest decision tree consistent with the examples  Find the attribute that best splits examples  Form tree with root = best attribute  For each value v i (or range) of best attribute  Selects those examples with best=v i  Construct subtree i by recursively calling decision tree with subset of examples, all attributes except best  Add a branch to tree with label=v i and subtree=subtree i

21. 21 Construct example decision tree

22. 22 Tree and Rules Learned S-length <5.5  5.5 3,4,8P-length  5.6 ≤2.4 1,5,9 2,6,7 If S-length < 5.5., then Verginica If S-length  5.5 and P-length  5.6, then Versicolour If S-length  5.5 and P-length ≤ 2.4, then Setosa

23. 23 Comparison of Target and Learned If S-length < 5.5., then Verginica If S-length  5.5 and P-length  5.6, then Versicolour If S-length  5.5 and P-length ≤ 2.4, then Setosa If sepal length > 6 and sepal width > 3.8 and petal length < 2.5 and petal width < 1.5 then class = Iris Setosa If sepal length > 5 and sepal width > 3 and petal length >5.5 and petal width >2 then class = Iris Versicolour If sepal length 3 and petal length  2.5 and ≤ 5.5 and petal width  1.5 and ≤ 2 then class = Iris Virginica

24. 24 Text Classification  Is text i a finance new article? PositiveNegative

25. 25 20 attributes  Investors 2  Dow 2  Jones 2  Industrial 1  Average 3  Percent 5  Gain 6  Trading 8  Broader 5  stock 5  Indicators 6  Standard 2  Rolling 1  Nasdaq 3  Early 10  Rest 12  More 13  first 11  Same 12  The 30

26. 26 20 attributes  Men’s  Basketball  Championship  UConn Huskies  Georgia Tech  Women  Playing  Crown  Titles  Games  Rebounds  All-America  early  rolling  Celebrates  Rest  More  First  The  same

27. 27 Example stockrollingtheclass 10340other 26835finance 37725other 45714other 58220finance 69425finance 75620finance 80235other 901125finance 1001528other

28. 28 Issues  Representation  How to map from a representation in the domain to a representation used for learning?  Training data  How can training data be acquired?  Amount of training data  How well does the algorithm do as we vary the amount of data?  Which attributes influence learning most?  Does the learning algorithm provide insight into the generalizations made?

29. 29 Constructing the Decision Tree  Goal: Find the smallest decision tree consistent with the examples  Find the attribute that best splits examples  Form tree with root = best attribute  For each value v i (or range) of best attribute  Selects those examples with best=v i  Construct subtree i by recursively calling decision tree with subset of examples, all attributes except best  Add a branch to tree with label=v i and subtree=subtree i

30. 30 Choosing the Best Attribute: Binary Classification  Want a formal measure that returns a maximum value when attribute makes a perfect split and minimum when it makes no distinction  Information theory (Shannon and Weaver 49)  H(P(v 1 ),…P(v n ))=∑-P(v i )log 2 P(v i ) H(1/2,1/2)=-1/2log 2 1/2-1/2log 2 1/2=1 bit H(1/100,1/99)=-.01log 2.01-.99log 2.99=.08 bits n i=1

31. 31 Information based on attributes = Remainder (A) P=n=10, so H(1/2,1/2)= 1 bit

32. 32 Information Gain  Information gain (from attribute test) = difference between the original information requirement and new requirement  Gain(A)=H(p/p+n,n/p+n)- Remainder(A)

33. 33 Example stockrollingtheclass 10340other 26835finance 37725other 45714other 58220finance 69425finance 75620finance 80235other 901125finance 1001528other

34. stockrolling <55-10  10 5-10 <5 1,8,9,102,3,4,5,6,7 1,5,6,8 2,3,4,7 9,10 Gain(stock)=1-[4/10H(1/10,3/10)+6/10H(4/10,2/10)]= 1-[.4((-.1* -3.32)-(.3*-1.74))+.6((-.4*-1.32)-(.2*-2.32))]= 1-[.303+.5952]=.105 Gain(rolling)=1-[4/10H(1/2,1/2)+4/10H(1/2,1/2)+2/10H(1/2,1/2)]=0

35. 35 Other cases  What if class is discrete valued, not binary?  What if an attribute has many values (e.g., 1 per instance)?

36. 36 Training vs. Testing  A learning algorithm is good if it uses its learned hypothesis to make accurate predictions on unseen data  Collect a large set of examples (with classifications)  Divide into two disjoint sets: the training set and the test set  Apply the learning algorithm to the training set, generating hypothesis h  Measure the percentage of examples in the test set that are correctly classified by h  Repeat for different sizes of training sets and different randomly selected training sets of each size.

37. 37

38. 38 Division into 3 sets  Inadvertent peeking Parameters that must be learned (e.g., how to split values) Generate different hypotheses for different parameter values on training data Choose values that perform best on testing data Why do we need to do this for selecting best attributes?

39. 39 Overfitting  Learning algorithms may use irrelevant attributes to make decisions For news, day published and newspaper  Decision tree pruning Prune away attributes with low information gain Use statistical significance to test whether gain is meaningful

40. 40 K-fold Cross Validation  To reduce overfitting  Run k experiments Use a different 1/k of data for testing each time Average the results  5-fold, 10-fold, leave-one-out

41. 41 Ensemble Learning  Learn from a collection of hypotheses  Majority voting  Enlarges the hypothesis space

42. 42 Boosting  Uses a weighted training set Each example as an associated weight w j 0 Higher weighted examples have higher importance  Initially, w j =1 for all examples  Next round: increase weights of misclassified examples, decrease other weights  From the new weighted set, generate hypothesis h 2  Continue until M hypotheses generated  Final ensemble hypothesis = weighted-majority combination of all M hypotheses  Weight each hypothesis according to how well it did on training data

43. 43 AdaBoost  If input learning algorithm is a weak learning algorithm L always returns a hypothesis with weighted error on training slightly better than random  Returns hypothesis that classifies training data perfectly for large enough M  Boosts the accuracy of the original learning algorithm on training data

44. 44 Issues  Representation  How to map from a representation in the domain to a representation used for learning?  Training data  How can training data be acquired?  Amount of training data  How well does the algorithm do as we vary the amount of data?  Which attributes influence learning most?  Does the learning algorithm provide insight into the generalizations made?