mljams

personal (if partial) implementations of major machine learning algos

multiclass classifiers

• KNN - predict class by the nearest K observations,
• multivariate gaussian Bayes classifier - learned via maximum likelihood estimation
• softmax logistic regression classifiers - learned via gradient descent

adaptive boosting

train an ensemble of weak classifiers

provably drives training error to zero (training error over iterations in blue)

and gives pretty solid generalizability (test error over iterations in green)

clustering and matrix factorization

• K-means clustering
• probabilistic collaborative filtering