Machine Learning 2D1431. Lecture 3: Concept Learning

Size: px

Start display at page:

Download "Machine Learning 2D1431. Lecture 3: Concept Learning"

Reynard McDonald
6 years ago
Views:

1 Machine Learning 2D1431 Lecture 3: Concept Learning

2 Question of the Day? What row of numbers comes next in this series?

3 Outline Learning from examples General-to specific ordering of hypotheses Version spaces and candidate elimination algorithm Inductive bias

4 Training Examples for Concept Enjoy Sport Concept: days on which my friend Aldo enjoys his favourite water sports Task: predict the value of Enjoy Sport for an arbitrary day based on the values of the other attributes Sky Sunny Sunny Rainy Sunny Temp Warm Warm Cold Warm Humid Wind Normal High Strong instance Strong High Strong High Strong Water Warm Warm Warm Cool Forecast Same Same Change Change Enjoy Sport Yes Yes No Yes

5 Representing Hypothesis Hypothesis h is a conjunction of constraints on attributes Each constraint can be: A specific value : e.g. Water=Warm A don t care value : e.g. Water=? No value allowed (null hypothesis): e.g. Water=Ø Example: hypothesis h Sky Temp Humid Wind Water Forecast < Sunny?? Strong? Same >

6 Prototypical Concept Learning Task Given: Instances X : Possible days decribed by the attributes Sky, Temp, Humidity, Wind, Water, Forecast Target function c: EnjoySport X {0,1} Hypotheses H: conjunction of literals e.g. < Sunny?? Strong? Same > Training examples D : positive and negative examples of the target function: <x 1,c(x 1 )>,, <x n,c(x n )> Determine: A hypothesis h in H such that h(x)=c(x) for all x in D.

7 Inductive Learning Hypothesis Any hypothesis found to approximate the target function well over the training examples, will also approximate the target function well over the unobserved examples.

8 Number of Instances, Concepts, Hypotheses Sky: Sunny, Cloudy, Rainy AirTemp: Warm, Cold Humidity: Normal, High Wind: Strong, Weak Water: Warm, Cold Forecast: Same, Change #distinct instances : 3*2*2*2*2*2 = 96 #distinct concepts : 2 96 #syntactically distinct hypotheses : 5*4*4*4*4*4=5120 #semantically distinct hypotheses : 1+4*3*3*3*3*3=973

9 General to Specific Order Consider two hypotheses: h 1 =< Sunny,?,?,Strong,?,?> h 2 =< Sunny,?,?,?,?,?> Set of instances covered by h 1 and h 2 : h 2 imposes fewer constraints than h 1 and therefore classifies more instances x as positive h(x)=1. Definition: Let h j and h k be boolean-valued functions defined over X. Then h j is more general than or equal to h k (written h j h k ) if and only if x X : [ (h k (x) = 1) (h j (x) = 1)] The relation imposes a partial order over the hypothesis space H that is utilized many concept learning methods.

10 Instance, Hypotheses and more general Instances Hypotheses specific x 1 h 3 h 1 x 2 h 2 h 1 h 2 h 3 h 2 general x 1 =< Sunny,Warm,High,Strong,Cool,Same> x 2 =< Sunny,Warm,High,Light,Warm,Same> h 1 =< Sunny,?,?,Strong,?,?> h 2 =< Sunny,?,?,?,?,?> h 3 =< Sunny,?,?,?,Cool,?>

11 Find-S Algorithm 1. Initialize h to the most specific hypothesis in H 2. For each positive training instance x For each attribute constraint a i in h If the constraint a i in h is satisfied by x then do nothing else replace a i in h by the next more general constraint that is satisfied by x 3. Output hypothesis h

12 Hypothesis Space Search by Find-S x 3 Instances Hypotheses h 0 specific x 1 x 2 h 1 h 2,3 x 4 h 4 general x h 0 =< Ø, Ø, Ø, Ø, Ø, Ø,> 1 =<Sunny,Warm,Normal,Strong,Warm,Same>+ h 1 =< Sunny,Warm,Normal, x 2 =<Sunny,Warm,High,Strong,Warm,Same>+ Strong,Warm,Same> x 3 =<Rainy,Cold,High,Strong,Warm,Change> - h 2,3 =< Sunny,Warm,?, Strong,Warm,Same> x 4 =<Sunny,Warm,High,Strong,Cool,Change> + h 4 =< Sunny,Warm,?, Strong,?,?>

13 Properties of Find-S Hypothesis space described by conjunctions of attributes Find-S will output the most specific hypothesis within H that is consistent with the positve training examples The output hypothesis will also be consistent with the negative examples, provided the target concept is contained in H.

14 Complaints about Find-S Can t tell if the learner has converged to the target concept, in the sense that it is unable to determine whether it has found the only hypothesis consistent with the training examples. Can t tell when training data is inconsistent, as it ignores negative training examples. Why prefer the most specific hypothesis? What if there are multiple maximally specific hypothesis?

15 Version Spaces A hypothesis h is consistent with a set of training examples D of target concept if and only if h(x)=c(x) for each training example <x,c(x)> in D. Consistent(h,D) := <x,c(x)> D h(x)=c(x) The version space, VS H,D, with respect to hypothesis space H, and training set D, is the subset of hypotheses from H consistent with all training examples: VS H,D = {h H Consistent(h,D) }

16 List-Then Eliminate Algorithm 1. VersionSpace a list containing every hypothesis in H 2. For each training example <x,c(x)> remove from VersionSpace any hypothesis that is inconsistent with the training example h(x) c(x) 3. Output the list of hypotheses in VersionSpace

17 Example Version Space S: {<Sunny,Warm,?,Strong,?,?>} <Sunny,?,?,Strong,?,?> <Sunny,Warm,?,?,?,?> <?,Warm,?,Strong,?,?> G: {<Sunny,?,?,?,?,?>, <?,Warm,?,?,?>, } x 1 = <Sunny Warm Normal Strong Warm Same> + x 2 = <Sunny Warm High Strong Warm Same> + x 3 = <Rainy Cold High Strong Warm Change> - x 4 = <Sunny Warm High Strong Cool Change> +

18 Representing Version Spaces The general boundary, G, of version space VS H,D is the set of maximally general members. The specific boundary, S, of version space VS H,D is the set of maximally specific members. Every member of the version space lies between these boundaries VS H,D = {h H ( s S) ( g G) (g h s) where x y means x is more general or equal than y

19 Candidate Elimination Algorithm G maximally general hypotheses in H S maximally specific hypotheses in H For each training example d=<x,c(x)> If d is a positive example Remove from G any hypothesis that is inconsistent with d For each hypothesis s in S that is not consistent with d remove s from S. Add to S all minimal generalizations h of s such that h consistent with d Some member of G is more general than h Remove from S any hypothesis that is more general than another hypothesis in S

20 Candidate Elimination Algorithm If d is a negative example Remove from S any hypothesis that is inconsistent with d For each hypothesis g in G that is not consistent with d remove g from G. Add to G all minimal specializations h of g such that h consistent with d Some member of S is more specific than h Remove from G any hypothesis that is less general than another hypothesis in G

21 Example Candidate Elimination S: {<,,,,, >} G: {<?,?,?,?,?,?>} x 1 = <Sunny Warm Normal Strong Warm Same> + S: G: {< Sunny Warm Normal Strong Warm Same >} {<?,?,?,?,?,?>} x 2 = <Sunny Warm High Strong Warm Same> + S: {< Sunny Warm? Strong Warm Same >} G: {<?,?,?,?,?,?>}

22 Example Candidate Elimination S: G: {< Sunny Warm? Strong Warm Same >} {<?,?,?,?,?,?>} x 3 = <Rainy Cold High Strong Warm Change> - S: {< Sunny Warm? Strong Warm Same >} G: {<Sunny,?,?,?,?,?>, <?,Warm,?,?,?>, <?,?,?,?,?,Same>} x 4 = <Sunny Warm High Strong Cool Change> + S: {< Sunny Warm? Strong?? >} G: {<Sunny,?,?,?,?,?>, <?,Warm,?,?,?> }

23 Example Candidate Elimination Instance space: integer points in the x,y plane with 0 x,y 10 hypothesis space : rectangles, that means hypotheses are of the form a x b, c y d, assume d Lab1: Assignment 1 c a b

24 Example Candidate Elimination examples = {ø} G={0,10,0,10} S={ø} G

25 Example Candidate Elimination examples = {(3,4),+} G={(0,10,0,10)} S={(3,3,4,4)} G S +

26 Classification of Unseen Data S: {<Sunny,Warm,?,Strong,?,?>} <Sunny,?,?,Strong,?,?> <Sunny,Warm,?,?,?,?> <?,Warm,?,Strong,?,?> G: {<Sunny,?,?,?,?,?>, <?,Warm,?,?,?>, } x 5 = <Sunny Warm Normal Strong Cool Change> x 6 = <Rainy Cold Normal Light Warm Same> x 7 = <Sunny Warm Normal Light Warm Same> x 8 = <Sunny Cold Normal Strong Warm Same> + 6/0-0/6? 3/3? 2/4

27 Inductive Leap + <Sunny Warm Normal Strong Cool Change> + <Sunny Warm Normal Light Warm Same> S : <Sunny Warm Normal???> new example <Sunny Warm Normal Strong Warm Same> How can we justify to classify the new example as positive? Since S is the specific boundary all other hypotheses in the version space are more general. So if the example satifies S it will also satisfy every other hypothesis in VS. inductive bias: We assume that the hypothesis space H contains the target concept c. In other words that c can be described by a conjunction of literals.

28 What Example to Query Next? S: {<Sunny,Warm,?,Strong,?,?>} <Sunny,?,?,Strong,?,?> <Sunny,Warm,?,?,?,?> <?,Warm,?,Strong,?,?> G: {<Sunny,?,?,?,?,?>, <?,Warm,?,?,?>, } What would be a good query for the learner to pose at that point? Choose an instance that is classified positive by some of the hypothesis and negative by the others. <Sunny, Warm, Normal, Light, Warm, Same> If the example is positive S can be generalized, if it is negative G can be specialized.

29 Biased Hypothesis Space Our hypothesis space is unable to represent a simple disjunctive target concept : (Sky=Sunny) v (Sky=Cloudy) x 1 = <Sunny Warm Normal Strong Cool Change> + S 1 : { <Sunny, Warm, Normal, Strong, Cool, Change> } x 2 = <Cloudy Warm Normal Strong Cool Change> + S 2 : { <?, Warm, Normal, Strong, Cool, Change> } x 3 = <Rainy Warm Normal Strong Cool Change> - S 3 : {} The third example x 3 contradicts the already overly general hypothesis space specific boundary S 2.

30 Unbiased Learner Idea: Choose H that expresses every teachable concept, that means H is the set of all possible subsets of X called the power set P(X) X =96, P(X) =2 96 ~ distinct concepts H = disjunctions, conjunctions, negations e.g. <Sunny Warm Normal???> v <????? Change> H surely contains the target concept.

31 Unbiased Learner What are S and G in this case? Assume positive examples (x 1, x 2, x 3 ) and negative examples (x 4, x 5 ) S : { (x 1 v x 2 v x 3 ) } G : { (x 4 v x 5 ) } The only examples that are classified are the training examples themselves. In other words in order to learn the target concept one would have to present every single instance in X as a training example. Each unobserved instance will be classified positive by precisely half the hypothesis in VS and negative by the other half.

32 Futility of Bias-Free Learning A learner that makes no prior assumptions regarding the identity of the target concept has no rational basis for classifying any unseen instances. No Free Lunch!

33 Inductive Bias Consider: Concept learning algorithm L Instances X, target concept c Training examples D c ={<x,c(x)>} Let L(x i,d c ) denote the classification assigned to instance x i by L after training on D c. Definition: The inductive bias of L is any minimal set of assertions B such that for any target concept c and corresponding training data D c ( x i X)[B D c x i ] -- L(x i, D c ) Where A -- B means that A logically entails B.

34 Inductive Systems and Equivalent Deductive Systems training examples new instance training examples new instance assertion H contains target concept candidate elimination algorithm using hypothesis space H equivalent deductive system theorem prover classification of new instance or don t know classification of new instance or don t know

35 Three Learners with Different Biases Rote learner: Store examples classify x if and only if it matches a previously observed example. No inductive bias Version space candidate elimination algorithm. Bias: The hypothesis space contains the target concept. Find-S Bias: The hypothesis space contains the target concept and all instances are negative instances unless the opposite is entailed by its other knowledge.

36 No Free Lunch Theorem If we can make no prior assumptions about the nature of the learning problem, there is no reason to prefer one learning algorithm over the other. If one algorithm seems to outperform another in a particular situation, it is a consequence of its fit to the particular problem, not the general superiority of the algorithm. For any two learning algorithms, the expected offtraining-set classification error averaged over all possible target functions is the same. No matter how clever we are at choosing a good learning algorithm and a bad algorithm, if all target functions are equally likely, then the good algorithm will not outperform the bad one.

37 training data D No Free Lunch Theorem for Binary Data x For the target function F(x) clearly algorithm 1 is superior to algorithm 2. For each of the 2 5 possible distinct target functions consistent with D, there is exactly one other target function whose output is inverted for each of the patterns outside the training set D, so that the contributions to ε 1 (E F,D) and ε 2 (E F,D) cancel. F h h ε 1 (E F,D)=0.4 ε 2 (E F,D)=0.6

38 Question of the Day? What row of numbers comes next in this series? Each row describes the series in the previous row, the first row has One One, the second Two Ones, the third row One Two and One One, the fourth row One One, One Two and Two Ones.so the answer is:

Question of the Day? Machine Learning 2D1431. Training Examples for Concept Enjoy Sport. Outline. Lecture 3: Concept Learning

Question of the Day? Machine Learning 2D43 Lecture 3: Concept Learning What row of numbers comes next in this series? 2 2 22 322 3222 Outline Training Examples for Concept Enjoy Sport Learning from examples