Course contentsShow
← Machine Learning and Deep Learning
Lesson 182 of 3,5382. Machine Learning Mindset and Core ConceptsFree lesson

Model Evaluation with Accuracy and Score Methods

Using .score() for quick evaluation and understanding default metrics for classification vs regression tasks.

Model Evaluation with Accuracy and Score Methods

What you'll learn: How to quickly evaluate your trained model's performance using scikit-learn's built-in .score() method and understand what metric it uses automatically.

The Quick Way to Check Model Performance

After fitting a model in scikit-learn, you need to know: how well does it work? The .score() method gives you an instant performance measurement with a single line of code.

Think of .score() as your model's report card—it takes your test data (features and true labels) and returns a number telling you how well your model performed. But here's the key: what that number means depends on whether you're doing classification or regression.

Classification vs Regression: Different Default Metrics

When you call .score() on a classifier (like the models you've been fitting), it returns accuracy—the percentage of predictions that were correct. If you predict 85 out of 100 labels correctly, you get a score of 0.85.

# After fitting a classifier
accuracy = model.score(X_test, y_test)
# Returns: fraction of correct predictions (0.0 to 1.0)

When you call .score() on a regressor (predicting continuous numbers), it returns the R² score (coefficient of determination)—a measure of how well your predictions match the actual values, where 1.0 is perfect and values can even be negative for very poor models.

# After fitting a regressor
r2_score = model.score(X_test, y_test)
# Returns: R² score (1.0 is perfect, can be negative)

Why This Matters

The .score() method is your first-line evaluation tool—quick, simple, and built-in. It's perfect for rapid iteration and comparing models during development.

Key Takeaway: Use .score() for quick model evaluation; it automatically returns accuracy for classifiers and R² for regressors, giving you an instant performance measurement without importing additional metrics.