1. Review B.Ramamurthy 7/11/2014CSE651, B. Ramamurthy1

2.  Generative vs. discriminative classifiers  Probabilistic models  Generative: based on (affirmative) prior knowledge (like Naïve Bayes prior)  Discriminative: based on discriminating knowledge: “this and not this” happening: logistic regression  We covered Bayes for test 2, we will cover logistic regression for final exam Classifiers 7/11/2014CSE651, B. Ramamurthy2

3.  From probability to odds ratio:  Lets say the probability of success of an event is 0.8; failure is 1 – 0.8 = 0.2  Then the odds of success are : 0.8/0.2 = 4  That is odds of success is 4: 1  Transformation from probability to odds is a monotonic function. As the probability increases so does the odds.  Examples: 0.9  odds 9: 1  log of odds  2.19  0.999  odds is 999: 1  log of odds  6.9  0.9999  9999:1  log of odds  9.2  0.5  odds 1:1  log of odds  0  0.2  odds 0.25:1  1:4  log of odds  --1.38  Log of odds is also monotonic function: it is negative till p =0.5  Very often this is what is used in many cases where plausibility of an event needs to proven.  LR is discriminative since is uses positive + negative occurrence probs in its calculation. Odds ratio and logistic regression 7/11/2014CSE651, B. Ramamurthy3

4.  Simple logistic regression problem: 10 points  Simples Processing problem: 10 points  For Android: no coding but questions about the components : 30 points  For JS: coding: simple examples: html, css, and js: 30 points  Cloud computing: Google app engine: 10 points; aws amazon cloud: 10 points; general concepts such as PaaS, Saas, Iaas are also included. Final exam 7/11/2014CSE651, B. Ramamurthy4