Lucamiras/drivendata_earthquakedamage

The main.ipynb notebook is performing a grid search to find the best hyperparameters for a Random Forest classifier. Here's a step-by-step explanation:

1. Define parameters grid for Random Forest: A dictionary param_grid is defined with possible values for the hyperparameters min_samples_split, min_samples_leaf, and n_estimators. These hyperparameters control the behavior of the Random Forest classifier.

2. Initialize Random Forest classifier: A RandomForestClassifier object rf_classifier is created with n_jobs=20, which means that the classifier will use 20 CPU cores for parallel computation.

3. Initialize Grid Search with 5-fold cross-validation: A GridSearchCV object grid_search is created with the Random Forest classifier, the parameters grid, and a 3-fold cross-validation. The scoring metric is f1_micro, which is the F1 score computed globally by counting the total true positives, false negatives, and false positives. The verbose=3 parameter means that the grid search will output detailed information about its progress.

4. Fit the grid search to the data: The grid_search object is fitted to the data X and the target y. This will perform the grid search, which involves training a Random Forest classifier for each combination of hyperparameters in the grid and evaluating its performance using 3-fold cross-validation.

5. Print the best parameters found: The best combination of hyperparameters found by the grid search is printed.

6. Print the best mean cross-validated score found: The mean cross-validated score of the best estimator found by the grid search is printed. This is the average F1 score over the 3 folds of the cross-validation for the best estimator.

7. Write the best parameters and score to a log file: The best parameters and score are written to a log file logs/logs.txt. This is useful for keeping a record of the results of the grid search.