Machine Learning Algorithms - Second Edition

Learning curve

In many tasks, it's helpful to check how the number of samples impacts training performance. This can be achieved by plotting a learning curve, which is normally based on both training and validation scores (preferably CV score). Let's consider the Wine dataset and a simple logistic regression. scikit-learn provides the built-in learning_curve() function, which can automatically compute the scores for a different number of training samples:

import numpy as npfrom sklearn.datasets import load_winefrom sklearn.model_selection import learning_curvefrom sklearn.linear_model import LogisticRegressionfrom sklearn.utils import shufflewine = load_wine()X, Y = shuffle(wine['data'], wine['target'])tsize, training_score, test_score = ...