06 - Probabilistic models and logistic regression

Overview

• Description:: models and regression

We can use two models:

$P(C_i|x)=posterior$ discriminative: given one input say what’s the probability to classify it in one class (first estimate $P(x|C_i)$ and the compute $P(C_i|x)$ using Bayes

or we can use a $P(x|C_i)=classconddensity$ generative approach: estimate class condition density directly

P(C1) = Pi1 P(C2) = Pi2 … P(Ck-1) = Pi k-1 P(Ck) = 1 - sum of all of them

for 8 class: 8 params but you must reply k-1

Dataset: $D=\{(x_n,t_n{)}_{n=1}^n\}$

There is a parameter called pi that is the sum between positive and negative over number of things we have. pi = 1 + 1 / N