Custom Pipeline with Grid Search

custom_pipeline_2.py

from sklearn.metrics import recall_score, make_scorer

# Define the pipeline
pipeline = Pipeline(steps= [
                    ("ColumnTransformer", SelectColumnsTransformer(columns=features_to_use)),
                    ("TransformCategorical", Transform_Categorical()),
                    ("SMOTE", SMOTE()),
                    ("GradientBooster", GradientBoostingClassifier())
                    ])

# Create a param grid experimenting with different feature sets and parameters
param_grid = {
              "ColumnTransformer__columns": [['account length', 'international plan', 'voice mail plan'],
                                          ['account length', 'international plan', 'total charge'], 
                                           ['international plan', 'total minutes', 'total calls']],
              "SMOTE__sampling_strategy": [0.5, 1],
              "GradientBooster__loss": ['deviance', 'exponential'],
              "GradientBooster__n_estimators": [100, 150],
              }

# Run a grid search with the pipeline
# Use scoring parameter to maximize the recall score
gs_pipeline = GridSearchCV(pipeline, param_grid=param_grid, verbose=2, scoring=make_scorer(recall_score))
gs_pipeline.fit(X_train, y_train)

# Store the best model
best_model = gs_pipeline.best_estimator_

# Make predictions on our validation set and see recall score of best model
y_validation_preds = best_model.predict(X_validation)
recall_score(y_validation, y_validation_preds)