Python中LASSO网格搜索时for循环print失效、变量未更新

Hey there! Let's dig into why your manual LASSO grid search is giving you zeroed-out R²/MSE arrays and missing print outputs—frustrating, especially since your single-alpha 10-fold CV works perfectly. Here are the most likely fixes, tailored to your situation:

1. Double-Check How You're Updating Your Metrics Arrays #

First off, if your R2_Lasso and MSE_Lasso stay all zeros, you probably forgot to assign the cross-validation results to the correct index in the array. For example:

• If you initialized R2_Lasso = np.zeros(len(alphas)), you need to explicitly set R2_Lasso[i] = cv_r2_score inside your loop (where cv_r2_score is the mean R² from your 10-fold CV for that alpha).
• It’s easy to accidentally skip this step—double-check that your loop is actually writing results to the array, not just calculating them and discarding them.

2. Reinitialize the LASSO Model Every Loop #

Scikit-learn models are stateful—if you create a single Lasso() instance outside your loop and just change its alpha parameter, you might be carrying over leftover state from previous fits. Instead, create a new model inside the loop for each alpha:

# ❌ Bad: Reusing the same model instance
lasso = Lasso()
for i, alpha in enumerate(alphas):
    lasso.set_params(alpha=alpha)
    # ... CV code ...

# ✅ Good: Fresh model every time
for i, alpha in enumerate(alphas):
    lasso = Lasso(alpha=alpha)  # New instance here
    # ... CV code ...

This ensures each alpha is tested on a clean model, which avoids weird carryover issues that could break your metrics.

3. Fix Missing Print Outputs with flush=True #

Sometimes, IDEs (especially Jupyter or Spyder) buffer print outputs until the loop finishes. To force prints to show up in real time, add flush=True to your print statements:

print(f"Testing alpha: {alpha}, iteration: {i}", flush=True)

This bypasses the buffer and makes your loop’s progress visible immediately.

4. Use Scikit-Learn’s Built-In GridSearchCV (The Better Approach) #

Manual loops are error-prone—Scikit-learn has a dedicated GridSearchCV tool that handles all the cross-validation, parameter iteration, and metric collection for you. It’s more reliable and cleaner. Here’s a quick example tailored to your use case:

from sklearn.linear_model import Lasso
from sklearn.model_selection import GridSearchCV
from sklearn.metrics import make_scorer, r2_score, mean_squared_error

# Replace with your actual training data
X_train, y_train = your_training_data, your_training_labels

# Define your alpha grid
alphas = [0.001, 0.01, 0.1, 1, 10, 100]

# Initialize the LASSO model and grid search
lasso = Lasso()
grid_search = GridSearchCV(
    estimator=lasso,
    param_grid={"alpha": alphas},
    cv=10,  # 10-fold CV
    scoring={
        "R2": make_scorer(r2_score),
        "MSE": make_scorer(mean_squared_error, greater_is_better=False)
    },
    refit="R2",  # Refit the best model on full training data using R2
    verbose=2  # Print progress updates
)

# Run the grid search
grid_search.fit(X_train, y_train)

# Extract and print results
print("\nGrid Search Results:")
for alpha, mean_r2, mean_mse in zip(
    grid_search.cv_results_["param_alpha"],
    grid_search.cv_results_["mean_test_R2"],
    -grid_search.cv_results_["mean_test_MSE"]  # Convert back to positive MSE
):
    print(f"Alpha: {alpha:>6} | Mean R²: {mean_r2:.4f} | Mean MSE: {mean_mse:.4f}")

# Get the best model and parameters
print(f"\nBest Alpha: {grid_search.best_params_['alpha']}")
print(f"Best Mean R²: {grid_search.best_score_:.4f}")

This will handle all the iteration, CV, and metric tracking automatically—no more worrying about array assignments or model state issues.

Give these fixes a try, and your grid search should start working as expected!

内容的提问来源于stack exchange，提问作者Raul Guarini Riva