This repository contains the code for demonstrating double descent on an extremely simple binary classification dataset with 6 samples, each with 2 features. The model uses random tanh features with ridge regression.
Due to the simplicity of the dataset and model, this example could potentially be used to demonstrate double descent in a classroom setting, or to test out new ideas.
min_double_descent.py produces a figure showing the peaking phenomenon in test error, test loss, condition number of the random features, and the norm of the weights.