Évaluer votre premier modèle

Training a model is only half the job. Next you ask the real question: how well does it actually predict? 🧐

Every estimator offers a score method that returns a single number summarizing performance. One call, one quick verdict.

You pass both features and the real answers, so score can compare its predictions against what truly happened.

For regressors, score returns R squared: the fraction of the target's variation your model manages to explain.

An R squared of 1.0 is a perfect fit and 0.0 is no better than guessing the mean. Higher is better, and it can even go negative.

Always score on data the model never saw. A great training score with a poor test score is the classic sign of overfitting.

For finer detail, the metrics module offers many measures. You compare true and predicted values directly with a chosen function.

Mean absolute error tells you the typical miss in the target's own units. It is wonderfully easy to explain to anyone.

For classifiers, the score method returns accuracy: the share of labels predicted correctly. Same method name, different meaning.

A single metric never tells the whole story. Pair it with plots and several measures before you trust a model.

A score only means something next to a baseline, like always predicting the mean. Beat that simple guess to prove real value.

What does score return for a regression model?

Use score for a fast R squared, reach for metrics like MAE for detail, and always test on unseen data versus a baseline. 🏁