1. Source for Information Gain Formula Artificial Intelligence: A Modern Approach by Stuart Russell and Peter Norvig Chapter 18: Learning from Observations

2. Similarity in CBR (Cont’d) Sources: –Chapter 4 –www.iiia.csic.es/People/enric/AICom.html –www.ai-cbr.org

3. Other Similarity Metrics Suppose that we have cases represented as attribute-value pairs (e.g., the restaurant domain) Suppose initially that the values are binary We want to define similarity between two cases of the form: X = (X 1, …, X n ) where X i = 0 or 1 Y = (Y 1, …,Y n ) where Y i = 0 or 1

4. Preliminaries Let:  A =  (i=1,n) X i Y i  B =  (i=1,n) X i (1-Y i )  C =  (i=1,n) (1-X i )Y i  D =  (i=1,n) (1-X i ) (1-Y i )  Then, A + B + C + D = (number of attributes for which X i =1 and Y i = 1) (number of attributes for which X i =1 and Y i = 0) (number of attributes for which X i =0 and Y i = 1) (number of attributes for which X i =0 and Y i = 0) n A+D = B+C= “matching attributes” “mismatching attributes”

5. Hamming Distance H(X,Y) = n –  (i=1,n) X i Y i –  (i=1,n) (1-X i )(1-Y i ) Properties:  Range of H:  H counts the mismatch between the attribute values  H is a distance metric:  H((1-X 1, …, 1-X n ), (1-Y 1, …,1-Y n )) = [0,n] H(X,X) = 0 H(X,Y) = H(Y,X) H((X 1, …, X n ), (Y 1, …,Y n ))

6. Simple-Matching-Coefficient (SMC)  H(X,Y) = n – (A + D) = B + C Another distance-similarity compatible function is f(x) = 1 – x/max (where max is the maximum value for x)  We can define the SMC similarity, sim H : sim H (X,Y) = 1 – ((n – (A+D))/n) = (A+D)/n = 1- ((B+C)/n) Homework (I): Show that f(x) is order inverting: if x f(y) Proportion of the difference # of mismatches

7. Simple-Matching-Coefficient (SMC) (II) If we use on sim H (X,Y) = (A+D)/n =1- ((B+C)/n) = factor(A, B, C, D)  Monotonic:  If A  A’ then:  If B  B’ then:  If C  C’ then:  If D  D’ then: factor(A,B,C,D)  factor(A’,B,C,D) factor(A,B’,C,D)  factor(A,B,C,D) factor(A,B,C’,D)  factor(A,B,C,D) factor(A,B,C,D)  factor(A,B,C,D’)  Symmetric: sim H (X,Y) = sim H (Y,X)

8. Variations of the SMC The hamming similarity assign equal value to matches (both 0 or both 1) There are situations in which you want to count different when both match with 1 as when both match with 0  Thus, sim((1-X 1, …, 1-X n ), (1-Y 1, …,1-Y n )) = sim((X 1, …, X n ), (Y 1, …,Y n )) may not hold  Example: Two symptoms of patients are similar if they both have fever (X i = 1 and Y i = 1) but not similar if neither have fever (X i = 0 and Y i = 0)  Specific attributes may be more important than other attributes Example: manufacturing domain: some parts of the workpiece are more important than others

9. Variations of SMC (III) We introduce a weight, , with 0 <  < 1: simH(X,Y) = (A+D)/n = (A+D)/(A+B+C+D) sim  (X,Y) = (  (A+D))/ (  (A+D) + (1 -  )(B+C))  For which  is sim  (X,Y) = sim H (X,Y)?  = 0.5  sim  (X,Y) preserves the monotonic and symmetric conditions Homework(II): Show that sim  (X,Y) is monotonic

10. The similarity depends only from A, B, C and D (3) What is the role of  ? What happens if  > 0.5? If  < 0.5? sim  (X,Y) = (  (A+D))/ (  (A+D) + (1 -  )(B+C)) 1 0 0 n  = 0.5  > 0.5  < 0.5 If  > 0.5 we give more weights to the matching attributes than to the miss- matching If  < 0.5 we give more weights to the miss- matching attributes than to the matching

11. Discarding 0-match Thus, sim((1-X 1, …, 1-X n ), (1-Y 1, …,1-Y n )) = sim((X 1, …, X n ), (Y 1, …,Y n )) may not hold Only when the attribute occurs (i.e., X i = 1 and Y i = 1 ) will contribute to the similarity  Possible definition of the similarity: sim = A / (A+ B+C)

12. Specific Attributes may be More Important Than Other Attributes Significance of the attributes varies Weighted Hamming distance: H W (X,Y) = 1 –  (i=1,n)  i X i Y i –  (i=1,n)  i (1-X i )(1-Y i )  There is a weight vector: (  1, …,  n ) such that  (i=1,n)  i = 1 Example: “Process planning: some features are more important than others”

13. Homework (Part III): Attributes May Have multiple Values X = (X 1, …, X n ) where X i  T i Y = (Y 1, …,Y n ) where Y i  T i Each T i is finite Define a formula for the Hamming distance in this context

14. Non Monotonic Similarity The monotony condition in similarity, formally, says that: sim(A,B)  sim(A’,B) always holds for any A and A’ such that A  A’ Informally the monotony condition can be expressed as: For any X, Y, X’ attribute-value vectors, If we obtain X’ by modifying X on the value of one attribute such that X’ and Y have the same value on that attribute then: sim(X,Y) sim(X’,Y) 

15. Non Monotonic Similarity (2) sim H (X,Y) =  (i=1,n) eq(X i,Y i ) / n  Is the hamming distance monotonic? Yes  Consider the XOR function:  (0,0) and (1,1) are on the same class (+)  (0,1) and (1,0) are on the same class (-)  Thus d((1,1),(1,0)) > d((1,1),(0,0))  Is this monotonic? No

16. Non Monotonic Similarity (3) You may think: “well that was mathematics, how about real world?” Suppose that we have two interconnected batteries B and B’ and 3 lamps X, Y and Z that have the following properties:  If X is on, B and B’ work  If Y is on, B or B’ work  If Z is on, B works 1 0 1 1 Ok Fail 2 0 1 0 Fail Ok 3 0 0 0 Fail Fail Situation X Y Z B B’ Thus: sim(1,3) > sim(1,2) Non monotonic!

17. Tversky Contrast Model Defines a non monotonic distance Comparison of a situation S with a prototype P (i.e, a case) S and P are sets of features The following sets:  A = S  P  B = P – S  C = S – P A S P C B

18. Tversky Contrast Model (2) Tversky-distance: Where f:  [0,  ) f, , , and  are fixed and defined by the user Example:  If f(A) = # elements in A   =  =  = 1  T counts the number of elements in common minus the differences  The Tversky-distance is not symmetric T(P,S) =  f(A) -  f(B) -  f(C)

19. Local versus Global Similarity Metrics In many situations we have similarity metrics between attributes of the same type (called local similarity metrics). Example: For a complex engine, we may have a similarity for the temperature of the engine In such situations a reasonable approach to define a global similarity sim  (x,y) is to “aggregate” the local similarity metrics sim i (x i,y i ). A widely used practice sim  (x,y) to increate monotonically with each sim i (x i,y i ). What requirements should we give to sim  (x,y) in terms of the use of sim i (x i,y i )?

20. Local versus Global Similarity Metrics (Formal Definitions) A local similarity metric on an attribute T i is a similarity metric sim i : T i  T i  [0,1] A function  : [0,1] n  [0,1] is an aggregation function if:   (0,0,…,0) = 0   is monotonic non-decreasing on every argument Given a collection of n similarity metrics sim 1, …, sim n, for attributes taken values from T i, a global similarity metric, is a similarity metric sim:V  V  [0,1], V in T 1  …  T n, such that there is an aggregation  function with: sim(X,Y) = sim  (X,Y) =  (sim 1 (X 1,Y 1 ), …,sim n (X n,Y n ))  (X 1,X 2,…,X n ) = (X 1 +X 2 +…+X n )/n Example: