classical-mlsupervisedclassification
Logistic regression
Binary and multinomial classification via the sigmoid/softmax — the neural-network of one layer.
Уровни глубины
L0Intro~2ч
Knows sigmoid maps reals to (0,1); reads decision-boundary plots.
L1Basics~10ч
Derives cross-entropy via MLE; fits with gradient descent by hand on 2D data.
L2Working~15ч
Handles multiclass softmax, class imbalance, calibration; implements from scratch in NumPy.
L3Advanced~25ч
Analyses convergence of logistic loss; uses L-BFGS/IRLS; regularised variants.
L4Research~50ч
Max-margin implicit bias; generalised linear models on manifolds.