Adjusting the regularization strength

Your current Lasso model has an \(R^2\) score of 84.7%. When a model applies overly powerful regularization it can suffer from high bias, hurting its predictive power.

Let's improve the balance between predictive power and model simplicity by tweaking the alpha parameter.

Deze oefening maakt deel uit van de cursus

Dimensionality Reduction in Python

Cursus bekijken

Oefeninstructies

Find the highest value for alpha that gives an \(R^2\) value above 98% from the options: 1, 0.5, 0.1, and 0.01.

Praktische interactieve oefening

Probeer deze oefening eens door deze voorbeeldcode in te vullen.

# Find the highest alpha value with R-squared above 98%
la = Lasso(____, random_state=0)

# Fits the model and calculates performance stats
la.fit(X_train_std, y_train)
r_squared = la.score(X_test_std, y_test)
n_ignored_features = sum(la.coef_ == 0)

# Print peformance stats 
print(f"The model can predict {r_squared:.1%} of the variance in the test set.")
print(f"{n_ignored_features} out of {len(la.coef_)} features were ignored.")

Code bewerken en uitvoeren