Add Grid Search Functionality for Best Hyperparameter Tuning

[Reinaldo-Kn]

The GridSearchModel class encapsulates the process of hyperparameter tuning using GridSearchCV, providing a modular and flexible interface for model selection and training. It includes methods for fitting a model, predicting, saving/loading models, and additional utility functions.

Parameters

model: sklearn estimator, optional (default=RandomForestRegressor())
    The machine learning model to be tuned. This can be any model compatible with GridSearchCV.

param_grid: dict, optional (default={'n_estimators': [10, 100], 'max_depth': [None, 10 ], 'min_samples_split': [2, 4 ]})
    A dictionary containing the hyperparameters and their respective ranges for the grid search. The grid will search through all possible combinations of these hyperparameters.

scoring: str, optional (default='neg_mean_absolute_error')
    The scoring metric used for evaluating the models during grid search. This should be a valid scoring metric recognized by scikit-learn.

cv: int, optional (default=5)
    The number of cross-validation folds to be used during the grid search.

test_size: float, optional (default=0.01)
    The proportion of the dataset to use as the test set. The default is 1% of the data.

Methods

• fit(self, df, target_column)
  Fits the model using grid search with the specified dataset and target variable.

    Parameters:
        df (pandas.DataFrame): The input dataset containing both features and the target variable.
        target_column (str): The name of the column that represents the target variable (the variable to predict).
    Returns:
        best_estimator_ (sklearn estimator): The model fitted with the best hyperparameters found during the grid search.

• predict(self, X)

Makes predictions using the best model found by grid search.

    Parameters:
        X (pandas.DataFrame or numpy.ndarray): Input data to make predictions on.
    Returns:
        y_pred (numpy.ndarray): The predicted values.

• score(self, X_test, y_test)

Evaluates the performance of the best model on a test dataset.

    Parameters:
        X_test (pandas.DataFrame or numpy.ndarray): Features of the test set.
        y_test (pandas.Series or numpy.ndarray): True values of the test set.
    Returns:
        score (float): The performance score of the best model on the test set.

• get_best_params(self)

Returns the best hyperparameters found by the grid search.

    Returns:
        best_params (dict): Dictionary of the best hyperparameters.

• save_model(self, filename)

Saves the best model to a file.

    Parameters:
        filename (str): The path where the model should be saved.
    Returns:
        None

• load_model(self, filename)

Loads a previously saved model from a file.

    Parameters:
        filename (str): The path where the model is saved.
    Returns:
        None

• plot_feature_importance(self, feature_names)

Plots the feature importances from the best model. Only works for models that have the feature_importances_ attribute, such as RandomForest.

    Parameters:
        feature_names (list): List of feature names in the same order as they appear in the dataset.
    Returns:
        None

• cross_val_score_summary(self, X, Y)

Generates a summary of cross-validation scores for the best model.

    Parameters:
        X (pandas.DataFrame): The features for cross-validation.
        Y (pandas.Series): The target variable for cross-validation.
    Returns:
        summary (dict): A dictionary containing the mean and standard deviation of the cross-validation scores, along with individual fold scores.

Example Usage

from sklearn.ensemble import RandomForestRegressor
from _gridsearch import GridSearchModel

# Initialize the model with default or custom hyperparameters
grid_model = GridSearchModel(
    model=RandomForestRegressor(),
    param_grid={'n_estimators': [50, 100], 'max_depth': [None, 10]},
    scoring='neg_mean_squared_error',
    cv=5,
    test_size=0.2
)

# Fit the model to the dataset
best_model = grid_model.fit(df, 'target_column')

# Get the best hyperparameters
best_params = grid_model.get_best_params()
print("Best Parameters: ", best_params)

# Predict on new data
y_pred = grid_model.predict(X_new)

# Evaluate the model on the test set
test_score = grid_model.score(X_test, y_test)
print("Test Score: ", test_score)

# Save the model
grid_model.save_model('best_model.joblib')

# Plot feature importance
grid_model.plot_feature_importance(feature_names)

# Get cross-validation summary
cv_summary = grid_model.cross_val_score_summary(X, Y)
print("CV Mean Score: ", cv_summary['mean_score'])
print("CV Standard Deviation: ", cv_summary['std_dev'])

You can view the new functions in Colab