Identify optimal L1 penalty coefficient

You will now tune the C parameter for the L1 regularization to discover the one which reduces model complexity while still maintaining good model performance metrics. You will run a for loop through possible C values and build logistic regression instances on each, as well as calculate performance metrics.

A list C has been created with the possible values. The l1_metrics array has been built with 3 columns, with the first being the C values, and the next two being placeholders for non-zero coefficient counts and the recall score of the model. The scaled features and target variables have been loaded as train_X, train_Y for training, and test_X, test_Y for testing.

Both numpy and pandas are loaded as np and pd as well as the recall_score function from sklearn.

This exercise is part of the course

Machine Learning for Marketing in Python

View Course

Exercise instructions

Run a for loop over the range from 0 to the length of the list C.
For each C candidate, initialize and fit a Logistic Regression and predict churn on test data.
For each C candidate, store the non-zero coefficients and the recall score in the second and third columns of l1_metrics.
Create a pandas DataFrame out of l1_metrics with the appropriate column names.

Hands-on interactive exercise

Have a go at this exercise by completing this sample code.

# Run a for loop over the range of C list length
for index in ___(0, len(C)):
  # Initialize and fit Logistic Regression with the C candidate
  logreg = ___(penalty='l1', C=C[___], solver='liblinear')
  logreg.fit(___, train_Y)
  # Predict churn on the testing data
  pred_test_Y = logreg.___(test_X)
  # Create non-zero count and recall score columns
  l1_metrics[index,1] = np.___(logreg.coef_)
  l1_metrics[index,2] = recall_score(___, pred_test_Y)

# Name the columns and print the array as pandas DataFrame
col_names = ['C','Non-Zero Coeffs','Recall']
print(pd.DataFrame(l1_metrics, columns=___))

Edit and Run Code