- Add a temp fix for the subset used in base and random strategies;

- Add new results for stage4.

code/bolsonaro/trainer.py

@@ -42,10 +42,14 @@ class Trainer(object):
     def base_score_metric_name(self):
         return self._base_score_metric_name
 
-    def init(self, model):
+    def init(self, model, subsets_used='train,dev'):
         if type(model) in [RandomForestRegressor, RandomForestClassifier]:
-            self._X_forest = self._dataset.X_train
-            self._y_forest = self._dataset.y_train
+            if subsets_used == 'train,dev':
+                self._X_forest = self._dataset.X_train
+                self._y_forest = self._dataset.y_train
+            else:
+                self._X_forest = np.concatenate([self._dataset.X_train, self._dataset.X_dev])
+                self._y_forest = np.concatenate([self._dataset.y_train, self._dataset.y_dev])    
             self._logger.debug('Fitting the forest on train subset')
         elif model.models_parameters.subsets_used == 'train,dev':
             self._X_forest = self._dataset.X_train

code/compute_results.py

@@ -328,7 +328,7 @@ if __name__ == "__main__":
             ylabel=experiments_score_metric,
             title='Loss values of {}\nusing different training subsets'.format(args.dataset_name))"""
     elif args.stage == 4:
-        if len(args.experiment_ids) != 5:
+        if len(args.experiment_ids) != 3:
             raise ValueError('In the case of stage 4, the number of specified experiment ids must be 3.')
 
         # Retreive the extracted forest sizes number used in order to have a base forest axis as long as necessary
@@ -351,7 +351,7 @@ if __name__ == "__main__":
             omp_with_params_experiment_score_metric = extract_scores_across_seeds_and_extracted_forest_sizes(
                 args.models_dir, args.results_dir, args.experiment_ids[2])
 
-        # base_with_params
+        """# base_with_params
         logger.info('Loading base_with_params experiment scores 2...')
         _, _, base_with_params_test_scores_2, \
             _ = \
@@ -361,7 +361,7 @@ if __name__ == "__main__":
         logger.info('Loading random_with_params experiment scores 2...')
         _, _, random_with_params_test_scores_2, \
             _, _ = \
-            extract_scores_across_seeds_and_extracted_forest_sizes(args.models_dir, args.results_dir, args.experiment_ids[4])
+            extract_scores_across_seeds_and_extracted_forest_sizes(args.models_dir, args.results_dir, args.experiment_ids[4])"""
 
         # Sanity check on the metrics retreived
         if not (base_with_params_experiment_score_metric == random_with_params_experiment_score_metric
@@ -374,10 +374,8 @@ if __name__ == "__main__":
 
         Plotter.plot_stage2_losses(
             file_path=output_path + os.sep + 'losses.png',
-            all_experiment_scores=[base_with_params_test_scores, base_with_params_test_scores_2, random_with_params_test_scores,
-                random_with_params_test_scores_2,
-                omp_with_params_test_scores],
-            all_labels=['base_train-dev', 'base', 'random_train-dev', 'random', 'omp'],
+            all_experiment_scores=[base_with_params_test_scores, random_with_params_test_scores, omp_with_params_test_scores],
+            all_labels=['base', 'random', 'omp'],
             x_value=with_params_extracted_forest_sizes,
             xlabel='Number of trees extracted',
             ylabel=experiments_score_metric,

code/train.py

@@ -73,7 +73,7 @@ def process_job(seed, parameters, experiment_id, hyperparameters):
 
             model = ModelFactory.build(dataset.task, model_parameters)
 
-            trainer.init(model)
+            trainer.init(model, subsets_used=parameters['subsets_used'])
             trainer.train(model)
             trainer.compute_results(model, sub_models_dir)
     else:
@@ -95,7 +95,7 @@ def process_job(seed, parameters, experiment_id, hyperparameters):
 
         model = ModelFactory.build(dataset.task, model_parameters)
 
-        trainer.init(model)
+        trainer.init(model, subsets_used=parameters['subsets_used'])
         trainer.train(model)
         trainer.compute_results(model, sub_models_dir)
     logger.info('Training done')