K-fold cross-validation demo for kernel classifiers

Build a 2D binary classification dataset, choose a kernelized classifier and its hyperparameters, and then inspect each fold’s training/validation split. In every fold view, training points are circles and validation points are squares.

1. Create a 2D binary classification dataset

Red class (label -1) Blue class (label +1)

Click anywhere in the square to add points. You can erase the nearest point in erase mode. Coordinates run from -1 to 1 on both axes.

2. Choose the cross-validation and model settings

Number of folds k

Stratified split

Loss

Kernel

Regularization λ

Learning rate

Epochs

Boundary resolution

Background shading in each fold panel shows the learned classifier on that training split, and the black curve shows the classification boundary. Classification error is measured by the sign of the learned score.

3. The k validation split panels

Red training point Blue training point Red validation point Blue validation point