Slideshare uses cookies to improve functionality and performance, and to provide you with relevant advertising. If you continue browsing the site, you agree to the use of cookies on this website. See our User Agreement and Privacy Policy.

Slideshare uses cookies to improve functionality and performance, and to provide you with relevant advertising. If you continue browsing the site, you agree to the use of cookies on this website. See our Privacy Policy and User Agreement for details.

Successfully reported this slideshow.

No Downloads

Total views

21,053

On SlideShare

0

From Embeds

0

Number of Embeds

69

Shares

0

Downloads

768

Comments

10

Likes

14

No notes for slide

- 1. Introduction to Machine Learning Lecture 10 Bayesian decision theory – Naïve Bayes Albert Orriols i Puig aorriols@salle.url.edu i l @ ll ld Artificial Intelligence – Machine Learning Enginyeria i Arquitectura La Salle gy q Universitat Ramon Llull
- 2. Recap of Lecture 9 Outputs the most probable hypothesis h∈H, given the data D + knowledge about prior probabilities of hypotheses in H Terminology: P(h|D): probability that h holds given data D. Posterior probability of h; confidence that h holds given D. P(h): prior probability of h (background knowledge we have about that h is a correct hypothesis) P(D): prior probability that training data D will be observed P(D|h): probability of observing D given h holds P (D | h )P (h ) P (h | D ) = P (D ) Slide 2 Artificial Intelligence Machine Learning
- 3. Bayes’ Theorem Given H the space of possible hypothesis The Th most probable h b bl hypothesis i the one that maximizes P(h|D) h i is h h ii P(h|D): P (D | h )P (h ) hMAP ≡ arg max P (h | D ) = arg max = arg max P (D | h )P (h ) P (D ) h∈H Slide 3 Artificial Intelligence Machine Learning
- 4. Today’s Agenda Bayesian Decision Theory y y Nominal Variables Continuous Variables A Medical Example Naïve Bayes Slide 4 Artificial Intelligence Introduction to C++
- 5. Bayesian Decision Theory Statistical approach to pattern classification pp p Forget about rule-based and tree-based models We will express the problem in probabilistic terms Goal Classify a pattern x into one of the two classes w1 or w2 to minimize the probability of misclassification P(error) Prior P i probability b bilit P(x) = Fraction of times that x belongs to class wk Without more information, we have to classify a new example x’. What should we do? if P ( w1 ) > P ( w2 ) ⎧w1 The best option if we know class of x = ⎨ nothing else about the domain! ⎩w 2 otherwise Slide 5 Artificial Intelligence Machine Learning
- 6. Bayesian Decision Theory Now, we measure a feature of each example x , p Threshold θ How we should classify these data? As the classes overlap, x1 cannot perfectly discriminate y At the end, I want my algorithm to put a threshold that defines the class boundaryy Slide 6 Artificial Intelligence Machine Learning
- 7. Bayesian Decision Theory Let’s dd L t’ add a second feature df t How we should classify these data? An oblique line will be a good discriminant So the problem turns out to be: How can we build or simulate this oblique line? Slide 7 Artificial Intelligence Machine Learning
- 8. Bayesian Decision Theory Assume that xi are nominal variables with possible values {xi1, xi2, …, xin} Let’s build a table of number of occurrences P(w1,xi1) = 1/8 x Xi1 Xi2 Xin Total P(w1) = 4/8 4 W1 1 3 0 P(xi1| w1) = 1/4 4 W2 0 2 2 Join probability P(wk,xij): Probability of a pattern having value xij for variable xi and belonging to class wk. That is, the value of each cell divided by the total number of examples. examples Priors P(wk): Marginal sums of each row Conditional P(xij| k) P b bilit th t a pattern h a value xij given th t it C diti l P( |w ): Probability that tt has l i that belongs to class wk. That is, each cell divided by the sum of each row. Slide 8 Artificial Intelligence Machine Learning
- 9. Bayesian Decision Theory Recall that recognizing that P(A,B)=P(B|A)P(A) = P(A|B)P(B) g g (,) (|)() (|)() P ( wk , xij ) = P ( xij | wk ) P ( wk ) P ( wk , xijj ) = P ( wk | xijj ) P ( xijj ) We have all these values Therefore P ( xij | wk ) P ( wk ) P ( wk | xij ) = P ( xij ) And the class: class of x =arg max k =1, 2 P ( wk | xij ) Slide 9 Artificial Intelligence Machine Learning
- 10. Bayesian Decision Theory From nominal to continuous attributes From probability mass functions to probability density functions ( (PDFs) s) b P ( x ∈ [a, b]) =∫ p ( x)dx where ∫ p(x)dx =1 a X As well, we have class-conditional PDFs p(x, wk) If we have d random variables x = ( 1, …, xd) e a e a do a ab es (x , r rr P( x ∈ R ) =∫ p ( x )dx R Slide 10 Artificial Intelligence Machine Learning
- 11. Naïve Bayes But step down… I still need to learn the probabilities from data p p described by Nominal attributes Continuous attributes That is, is Given a new instance with attributes (a1,a2,…,an), the Bayesian approach classifies it to the most probable value vMAP v MAP = arg max P (v j | a1, a2 ,..., an ) v j ∈V Using Bayes’ theorem: P (a1, a2 ,..., an | v j )P (v j ) v MAP = arg max = arg max P (a1, a2 ,..., an | v j )P (v j ) P (a1, a2 ,..., an ) v j ∈V v j ∈V How to compute P(vj) and P(a1,a2,…,an|vj) ? a a Slide 11 Artificial Intelligence Machine Learning
- 12. Naïve Bayes How to compute P(vj)? p () P(vj): counting the frequency with which each target value vj occurs in the training data. How to compute P(a1,a2,…,an|vj) ? P(a1,a2,…,an|vj) : we should have a very large dataset. The number of these terms=number of possible instances times the number of possible target values (infeasible). (i f ibl ) Simplifying assumption: the attribute values are conditionally independent given the target value. I.e., the probability of observing (a1,a2,…,an) is the product of the probabilities for the individual attributes. Slide 12 Artificial Intelligence Machine Learning
- 13. Naïve Bayes Prediction of Naïve Bayes classifier: v NB = arg max P (v j )∏ P (ai |v j ) v j ∈V i The learning algorithm: gg Training: Estimate the probabilities P(vj) and P(ai|vj) based on their frequencies over the training data Output after training: The l Th learned hypothesis consists of the set of estimates dh th i i t f th t f ti t Test: Use formula above to classify new instances Observations: Number of distinct P(ai|vj) terms =number of distinct attribute values times the number number of distinct target values The algorithm does not p g perform an explicit search through the space of p g p possible hypothesis (the space of possible hypotheses is the space of possible values that can be assigned to the various probabilities). Slide 13 Artificial Intelligence Machine Learning
- 14. Example Given the training examples: g p Day Outlook Temperature Humidity Wind PlayTennis D1 Sunny Hot High Weak No D2 Sunny S Hot Ht High Hi h Strong St No N D3 Overcast Hot High Weak Yes D4 Rain Mild High Weak Yes D5 Rain Cool Normal Weak Yes D6 Rain Cool Normal Strong No D7 Overcast Cool Normal Strong Yes D8 Sunny Mild High Weak No D9 Sunny Cool Normal Weak Yes D10 Rain Mild Normal Weak Yes D11 Sunny Mild Normal Strong Yes D12 Overcast Mild High Strong Yes D13 Overcast Hot Normal Weak Yes D14 Rain Mild High Strong No Classify the new instance: (Outlook=sunny, Temp=cool, Humidity=high, Wind=strong) Slide 14 Artificial Intelligence Machine Learning
- 15. Example Naive Bayes training Sunny|Yes 2/9 Sunny|No 3/5 Outlook|Yes Overcast|Yes 4/9 Outlook|No Overcast|No 0 Rain|Yes 3/9 Rain|No 2/5 Hot|Yes 2/9 Hot 2/5 Temperature|yYes Mild|Yes 4/9 Temperature|No Mild 2/5 Cool|Yes | 3/9 Cool 1/5 High 3/9 High 4/5 Humidity|Yes Humidity|No Normal 6/9 Normal 1/5 Weak 6/9 Weak 2/5 Wind|Yes Wind|Yes Strong 3/9 Strong 3/5 P(Yes)=9/14 P(No)=5/14 () Test: Classify (Outlook=sunny, Temp cool, Humidity high, Wind strong) (Outlook sunny, Temp=cool, Humidity=high, Wind=strong) max { 9/14·2/9·3/9·3/9·3/9, 5/14·3/5·1/5·4/5·3/5} = {.0053, .0206} = 0.0206 Do not play tennis! Slide 15 Artificial Intelligence Machine Learning
- 16. Estimation of Probabilities The explained process to estimate probabilities could lead to poor estimate if the number of observations is small E.g.: P( Outlook=overcast| No) = 0.008, but we only have 5 examples Use the following estimate nc + mp n+m where p is the prior estimate of the probability we wish to determine m : constant, equivalent sample size, which determines the weightg assigned to the observed data Assuming uniform distribution, p=1/k, being k the number of values of th attribute. f the tt ib t E.g., P(Outlook=overcast | No): nc + mp 0 + 1/ 3·2 = n+m 5+2 Slide 16 Artificial Intelligence Machine Learning
- 17. Next Class Neural Networks and Support Vector Machines Slide 17 Artificial Intelligence Introduction to C++
- 18. Introduction to Machine Learning Lecture 10 Bayesian decision theory – Naïve Bayes Albert Orriols i Puig aorriols@salle.url.edu i l @ ll ld Artificial Intelligence – Machine Learning Enginyeria i Arquitectura La Salle gy q Universitat Ramon Llull

No public clipboards found for this slide

Login to see the comments