Lecture 26 - Supervised Learning II

#notes#cs471

Recap §

Maximum Likelihood for Learning:

• $\theta =\frac{{\sum }_n{y}^{(n)}}{N}$

Naïve Bayes §

• Features are conditionally independent given target
• $P(Y,x1,x2,x3,x4)=P(Y){\Pi }_{k}P(x_k|Y)$

Laplace Smoothing §

• Pretend you saw every outcome X times more than you already did$P_{LAP}(x)=\frac{c(x)+1}{{\sum }_x[c(x)+1]}$

S-Fold cross validation §

• Every data point serves in training and validation dataset.
• Split data into s parts
• Use each part in turn as a validation dataset and others as training.
• Choose hyperparameter leading to best average performance
• Leave one out cross validation: Every data point is used as validation once