- In compute_results, add loadenv, load raw results, and update experiment_ids...

- In compute_results, add loadenv, load raw results, and update experiment_ids so that it's possible to specify a list of experiments ids. The default behavior is to load all exp ids;
- Fix normalization option in train.py. By default it normalizes D, and it doesn't when specify wo_normalization option;
- Fix logger.warn to logger.warning in train.py
- Replace the dumping of result in trainer.py by a dedicated class to save and load the trained model and training metadatas: model_raw_results.py;
- Rename too long func DatasetLoader.load_from_name to DatasetLoader.load;
- Add loading functions in dataset_parameters and model_parameters;
- Set console logging level to INFO to summarize the most important console logs;
- Add a load function in model_factory;
- In omp_forest_regressor, move private funcs to the bottom of the file.

TODO: compute the plot from the loaded raw results in compute_results file.

code/bolsonaro/data/dataset_loader.py

@@ -11,7 +11,7 @@ from sklearn.model_selection import train_test_split
 class DatasetLoader(object):
 
     @staticmethod
-    def load_from_name(dataset_parameters):
+    def load(dataset_parameters):
         name = dataset_parameters.name
         if name == 'boston':
             dataset_loading_func = load_boston

code/bolsonaro/data/dataset_parameters.py

@@ -48,3 +48,16 @@ class DatasetParameters(object):
             },
             output_file,
             indent=4)
+
+    @staticmethod
+    def load(directory_path, experiment_id):
+        with open(directory_path + os.sep + 'dataset_parameters_{}.json'.format(experiment_id), 'r') as input_file:
+            parameters = json.load(input_file)
+        return DatasetParameters(
+            name=parameters['name'],
+            test_size=parameters['test_size'],
+            dev_size=parameters['dev_size'],
+            random_state=parameters['random_state'],
+            normalize=parameters['normalize'],
+            train_on_subset=parameters['train_on_subset']
+        )

code/bolsonaro/error_handling/logger_factory.py

@@ -50,7 +50,7 @@ class LoggerFactory(object):
 
         # Create console handler
         ch = logging.StreamHandler()
-        ch.setLevel(logging.DEBUG)
+        ch.setLevel(logging.INFO)
 
         # Create formatter
         formatter = logging.Formatter('%(asctime)s - %(filename)s:%(lineno)s - %(name)s - %(levelname)s - %(message)s')

code/bolsonaro/models/model_factory.py

 from bolsonaro.models.omp_forest_classifier import OmpForestClassifier
 from bolsonaro.models.omp_forest_regressor import OmpForestRegressor
 from bolsonaro.data.task import Task
+from bolsonaro.models.model_parameters import ModelParameters
+
+import os
+import pickle
 
 
 class ModelFactory(object):
@@ -14,3 +18,11 @@ class ModelFactory(object):
         else:
             raise ValueError("Unsupported task '{}'".format(task))
         return model_func(model_parameters)
+
+    @staticmethod
+    def load(task, directory_path, experiment_id, model_raw_results):
+        model_parameters = ModelParameters.load(directory_path, experiment_id)
+        model = ModelFactory.build(task, model_parameters)
+        model.set_forest(model_raw_results.forest)
+        model.set_weights(model_raw_results.weights)
+        return model

code/bolsonaro/models/model_parameters.py

@@ -36,3 +36,14 @@ class ModelParameters(object):
             },
             output_file,
             indent=4)
+
+    @staticmethod
+    def load(directory_path, experiment_id):
+        with open(directory_path + os.sep + 'model_parameters_{}.json'.format(experiment_id), 'r') as input_file:
+            parameters = json.load(input_file)
+        return ModelParameters(
+            forest_size=parameters['forest_size'],
+            extracted_forest_size=parameters['extracted_forest_size'],
+            seed=parameters['seed'],
+            normalize=parameters['normalize']
+        )

code/bolsonaro/models/model_raw_results.py

+import pickle
+import os
+import datetime
+
+
+class ModelRawResults(object):
+
+    def __init__(self, forest, weights, training_time,
+        datetime, train_score, dev_score, test_score,
+        score_metric, train_score_regressor, dev_score_regressor,
+        test_score_regressor):
+
+        self._forest = forest
+        self._weights = weights
+        self._training_time = training_time
+        self._datetime = datetime
+        self._train_score = train_score
+        self._dev_score = dev_score
+        self._test_score = test_score
+        self._score_metric = score_metric
+        self._train_score_regressor = train_score_regressor
+        self._dev_score_regressor = dev_score_regressor
+        self._test_score_regressor = test_score_regressor
+    
+    @property
+    def forest(self):
+        return self._forest
+
+    @property
+    def weights(self):
+        return self._weights
+
+    @property
+    def training_time(self):
+        return self._training_time
+
+    @property
+    def datetime(self):
+        return self._datetime
+
+    @property
+    def train_score(self):
+        return self._train_score
+
+    @property
+    def dev_score(self):
+        return self._dev_score
+
+    @property
+    def test_score(self):
+        return self._test_score
+
+    @property
+    def score_metric(self):
+        return self._score_metric
+
+    @property
+    def train_score_regressor(self):
+        return self._train_score_regressor
+
+    @property
+    def dev_score_regressor(self):
+        return self._dev_score_regressor
+
+    @property
+    def test_score_regressor(self):
+        return self._test_score_regressor
+
+    @staticmethod
+    def save(models_dir, model, end_time, begin_time, dataset, logger):
+        output_file_path = models_dir + os.sep + 'model_raw_results.pickle'
+        logger.debug('Saving trained model and raw results to {}'.format(output_file_path))
+        with open(output_file_path, 'wb') as output_file:
+            pickle.dump({
+                'forest': model.forest,
+                'weights': model.weights,
+                'training_time': end_time - begin_time,
+                'datetime': datetime.datetime.now(),
+                'train_score': model.score(dataset.X_train, dataset.y_train),
+                'dev_score': model.score(dataset.X_dev, dataset.y_dev),
+                'test_score': model.score(dataset.X_test, dataset.y_test),
+                'score_metric': model.default_score_metric,
+                'train_score_regressor': model.score_regressor(dataset.X_train, dataset.y_train),
+                'dev_score_regressor': model.score_regressor(dataset.X_dev, dataset.y_dev),
+                'test_score_regressor': model.score_regressor(dataset.X_test, dataset.y_test)
+            }, output_file)
+
+    @staticmethod
+    def load(models_dir):
+        model_file_path = models_dir + os.sep + 'model_raw_results.pickle'
+        with open(model_file_path, 'rb') as input_file:
+            model_data = pickle.load(input_file)
+        return ModelRawResults(
+            forest=model_data['forest'],
+            weights=model_data['weights'],
+            training_time=model_data['training_time'],
+            datetime=model_data['datetime'],
+            train_score=model_data['train_score'],
+            dev_score=model_data['dev_score'],
+            test_score=model_data['test_score'],
+            score_metric=model_data['score_metric'],
+            train_score_regressor=model_data['train_score_regressor'],
+            dev_score_regressor=model_data['dev_score_regressor'],
+            test_score_regressor=model_data['test_score_regressor']
+        )

code/bolsonaro/models/omp_forest_regressor.py

@@ -9,72 +9,41 @@ import numpy as np
 
 class OmpForestRegressor(BaseEstimator):
 
+    default_score_metric = 'mse'
+
     def __init__(self, models_parameters):
         self._regressor = RandomForestRegressor(n_estimators=models_parameters.forest_size,
             random_state=models_parameters.seed)
         self._models_parameters = models_parameters
         self._logger = LoggerFactory.create(LOG_PATH, __name__)
 
-    def fit(self, X_train, y_train):
-        self._forest = self._train_forest(X_train, y_train)
-
-        self._weights = self._extract_subforest(X_train, y_train)
-
-        return self
-
     @property
     def forest(self):
         return self._forest
 
+    def set_forest(self, forest):
+        self._forest = forest
+        self._regressor.estimators_ = forest
+
     @property
     def weights(self):
         return self._weights
 
+    def set_weights(self, weights):
+        self._weights = weights
+
     @property
     def models_parameters(self):
         return self._models_parameters
 
+    def fit(self, X, y):
+        self._forest = self._train_forest(X, y)
+        self._weights = self._extract_subforest(X, y)
+        return self
+
     def score_regressor(self, X, y):
         return self._regressor.score(X, y)
 
-    def _train_forest(self, X_train, y_train):
-        self._regressor.fit(X_train, y_train)
-        forest = self._regressor.estimators_
-        return forest
-    
-    def _extract_subforest(self, X_train, y_train):
-        """
-        Given an already estimated regressor: apply OMP to get the weight of each tree.
-
-        The X_train data is used for interrogation of every tree in the forest. The y_train data
-        is used for finding the weights in OMP.
-
-        :param X_train: (n_sample, n_features) array
-        :param y_train: (n_sample,) array
-        :return:
-        """
-        self._logger.debug("Forest make prediction on X_train")
-        D = self._forest_prediction(X_train)
-
-        if self._models_parameters.normalize:
-            # question: maybe consider other kinds of normalization
-            self._logger.debug("Compute norm of predicted vectors on X_train")
-            self._forest_norms = np.linalg.norm(D, axis=0)
-            D /= self._forest_norms
-
-        omp = OrthogonalMatchingPursuit(
-            n_nonzero_coefs=self._models_parameters.extracted_forest_size,
-            fit_intercept=False, normalize=False)
-        self._logger.debug("Apply orthogonal maching pursuit on forest for {} extracted trees."
-                           .format(self._models_parameters.extracted_forest_size))
-        omp.fit(D, y_train)
-        weights = omp.coef_
-        # question: why not to use directly the omp estimator instead of bypassing it using the coefs?
-        return weights
-
-    def _forest_prediction(self, X):
-        return np.array([tree.predict(X) for tree in self._forest]).T
-
     def predict(self, X):
         """
         Apply the OMPForestRegressor to X.
@@ -91,8 +60,7 @@ class OmpForestRegressor(BaseEstimator):
 
         return predictions
 
-
-    def score(self, X, y, metric="mse"):
+    def score(self, X, y, metric=default_score_metric):
         """
         Evaluate OMPForestRegressor on (`X`, `y`) using `metric`
 
@@ -103,9 +71,47 @@ class OmpForestRegressor(BaseEstimator):
         """
         predictions = self.predict(X)
 
-        if metric == "mse":
+        if metric == 'mse':
             evaluation = np.mean(np.square(predictions - y))
         else:
-            raise ValueError("Metric value {} is not known.")
+            raise ValueError("Unsupported metric '{}'.".format(metric))
 
         return evaluation
+
+    def _train_forest(self, X, y):
+        self._regressor.fit(X, y)
+        forest = self._regressor.estimators_
+        return forest
+    
+    def _extract_subforest(self, X, y):
+        """
+        Given an already estimated regressor: apply OMP to get the weight of each tree.
+
+        The X data is used for interrogation of every tree in the forest. The y data
+        is used for finding the weights in OMP.
+
+        :param X: (n_sample, n_features) array
+        :param y: (n_sample,) array
+        :return:
+        """
+        self._logger.debug("Forest make prediction on X")
+        D = self._forest_prediction(X)
+
+        if self._models_parameters.normalize:
+            # question: maybe consider other kinds of normalization
+            self._logger.debug("Compute norm of predicted vectors on X")
+            self._forest_norms = np.linalg.norm(D, axis=0)
+            D /= self._forest_norms
+
+        omp = OrthogonalMatchingPursuit(
+            n_nonzero_coefs=self._models_parameters.extracted_forest_size,
+            fit_intercept=False, normalize=False)
+        self._logger.debug("Apply orthogonal maching pursuit on forest for {} extracted trees."
+                           .format(self._models_parameters.extracted_forest_size))
+        omp.fit(D, y)
+        weights = omp.coef_
+        # question: why not to use directly the omp estimator instead of bypassing it using the coefs?
+        return weights
+
+    def _forest_prediction(self, X):
+        return np.array([tree.predict(X) for tree in self._forest]).T

code/bolsonaro/trainer.py

+from bolsonaro.models.model_raw_results import ModelRawResults
 from bolsonaro.error_handling.logger_factory import LoggerFactory
 from . import LOG_PATH
 
-import pickle
-import os
 import time
-import datetime
 
 
 class Trainer(object):
@@ -14,26 +12,17 @@ class Trainer(object):
         self._logger = LoggerFactory.create(LOG_PATH, __name__)
 
     def train(self, model, models_dir):
-        self._logger.info('Training model using train set...')
+        self._logger.debug('Training model using train set...')
         begin_time = time.time()
-        if self._dataset.dataset_parameters.train_on_subset == 'train':
+        train_on_subset = self._dataset.dataset_parameters.train_on_subset
+        if train_on_subset == 'train':
             X, y = self._dataset.X_train, self._dataset.y_train
-        elif self._dataset.dataset_parameters.train_on_subset == 'dev':
+        elif train_on_subset == 'dev':
             X, y = self._dataset.X_dev, self._dataset.y_dev
         else:
-            raise ValueError("Unsupported train_on_subset value '{}'".format(self._dataset.dataset_parameters.train_on_subset))
+            raise ValueError("Unsupported train_on_subset value '{}'".format(train_on_subset))
+        self._logger.debug('Fitting on {} subset'.format(train_on_subset))
         model.fit(X, y)
         end_time = time.time()
 
-        self._dump_raw_results(models_dir, model, end_time, begin_time)
-
-    def _dump_raw_results(self, models_dir, model, end_time, begin_time):
-        output_file_path = models_dir + os.sep + 'model.pickle'
-        self._logger.info('Saving trained model to {}'.format(output_file_path))
-        with open(output_file_path, 'wb') as output_file:
-            pickle.dump({
-                'forest': model.forest,
-                'weights': model.weights,
-                'training_time': end_time - begin_time,
-                'datetime': datetime.datetime.now()
-            }, output_file)
+        ModelRawResults.save(models_dir, model, end_time, begin_time, self._dataset, self._logger)

code/compute_results.py

+from bolsonaro.data.dataset_parameters import DatasetParameters
+from bolsonaro.data.dataset_loader import DatasetLoader
+from bolsonaro.models.model_raw_results import ModelRawResults
+from bolsonaro.models.model_factory import ModelFactory
+
 import argparse
 import pathlib
+from dotenv import find_dotenv, load_dotenv
+import os
 
 
 if __name__ == "__main__":
-    default_results_dir = 'results'
-    default_models_dir = 'models'
-    default_experiment_id = -1
+    # get environment variables in .env
+    load_dotenv(find_dotenv('.env.example'))
+
+    default_results_dir = os.environ["project_dir"] + os.sep + 'results'
+    default_models_dir = os.environ["project_dir"] + os.sep + 'models'
+    default_experiment_ids = None
 
     parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
     parser.add_argument('--results_dir', nargs='?', type=str, default=default_results_dir, help='The output directory of the results.')
     parser.add_argument('--models_dir', nargs='?', type=str, default=default_models_dir, help='The output directory of the trained models.')
-    parser.add_argument('--experiment_id', nargs='?', type=int, default=default_experiment_id, help='Compute the results of a single experiment id')
+    parser.add_argument('--experiment_ids', nargs='+', type=int, default=default_experiment_ids, help='Compute the results of the specified experiment id(s)')
     args = parser.parse_args()
 
     pathlib.Path(args.results_dir).mkdir(parents=True, exist_ok=True)
 
-    if args.experiment_id == -1:
-        pass
-    else:
-        pass
+    experiments_ids = [str(experiment_id) for experiment_id in args.experiment_ids] \
+        if args.experiment_ids is not None \
+        else os.listdir(args.models_dir)
+
+    if experiments_ids is None or len(experiments_ids) == 0:
+        raise ValueError("No experiment id was found or specified.")
+
+    for experiment_id in experiments_ids:
+        experiment_id_path = args.models_dir + os.sep + experiment_id
+        experiment_seed_root_path = experiment_id_path + os.sep + 'seeds'
+        for seed in os.listdir(experiment_seed_root_path):
+            experiment_seed_path = experiment_seed_root_path + os.sep + seed
+            dataset_parameters = DatasetParameters.load(experiment_seed_path, experiment_id)
+            dataset = DatasetLoader.load(dataset_parameters)
+            extracted_forest_size_root_path = experiment_seed_path + os.sep + 'extracted_forest_size'
+            for extracted_forest_size in os.listdir(extracted_forest_size_root_path):
+                extracted_forest_size_path = extracted_forest_size_root_path + os.sep + extracted_forest_size
+                model_raw_results = ModelRawResults.load(extracted_forest_size_path)
+                model = ModelFactory.load(dataset.task, extracted_forest_size_path, experiment_id, model_raw_results)
+                
+                

code/train.py

@@ -21,6 +21,7 @@ if __name__ == "__main__":
 
     default_dataset_name = 'boston'
     default_normalize = True
+    default_wo_normalization = False
     default_forest_size = 100
     default_extracted_forest_size = 10
     # the models will be stored in a directory structure like: models/{experiment_id}/seeds/{seed_nb}/extracted_forest_size/{nb_extracted_trees}
@@ -34,7 +35,7 @@ if __name__ == "__main__":
 
     parser = argparse.ArgumentParser(formatter_class=argparse.ArgumentDefaultsHelpFormatter)
     parser.add_argument('--dataset_name', nargs='?', type=str, default=default_dataset_name, help='Specify the dataset. Regression: boston, diabetes, linnerud, california_housing. Classification: iris, digits, wine, breast_cancer, olivetti_faces, 20newsgroups, 20newsgroups_vectorized, lfw_people, lfw_pairs, covtype, rcv1, kddcup99.')
-    parser.add_argument('--normalize', action='store_true', default=default_normalize, help='Normalize the data by doing the L2 division of the pred vectors.')
+    parser.add_argument('--wo_normalization', action='store_true', default=default_wo_normalization, help='Withouyt normalize the data by doing the L2 division of the pred vectors.')
     parser.add_argument('--forest_size', nargs='?', type=int, default=default_forest_size, help='The number of trees of the random forest.')
     parser.add_argument('--extracted_forest_size', nargs='+', type=int, default=default_extracted_forest_size, help='The number of trees selected by OMP.')
     parser.add_argument('--models_dir', nargs='?', type=str, default=default_models_dir, help='The output directory of the trained models.')
@@ -47,22 +48,27 @@ if __name__ == "__main__":
 
     pathlib.Path(args.models_dir).mkdir(parents=True, exist_ok=True)
 
-    logger = LoggerFactory.create(LOG_PATH, __name__)
+    logger = LoggerFactory.create(LOG_PATH, os.path.basename(__file__))
 
     args.extracted_forest_size = args.extracted_forest_size \
         if type(args.extracted_forest_size) == list \
         else [args.extracted_forest_size]
 
     if args.seeds != None and args.random_seed_number > 1:
-        logger.warn('seeds and random_seed_number parameters are both specified. Seeds will be used.')    
+        logger.warning('seeds and random_seed_number parameters are both specified. Seeds will be used.')    
+
     seeds = args.seeds if args.seeds is not None \
         else [random.randint(begin_random_seed_range, end_random_seed_range) \
         for i in range(args.random_seed_number)]
 
+    normalize = default_normalize and args.wo_normalization is False
+    logger.debug('normalize={}'.format(normalize))
+
     experiment_id = resolve_experiment_id(args.models_dir)
     experiment_id_str = str(experiment_id)
 
     for seed in seeds:
+        logger.debug('Seed={}'.format(seed))
         seed_str = str(seed)
         models_dir = args.models_dir + os.sep + experiment_id_str + os.sep + 'seeds' + \
             os.sep + seed_str
@@ -77,16 +83,17 @@ if __name__ == "__main__":
             test_size=args.test_size,
             dev_size=args.dev_size,
             random_state=seed,
-            normalize=args.normalize,
+            normalize=normalize,
             train_on_subset=args.train_on_subset
         )
         dataset_parameters.save(models_dir, experiment_id_str)
 
-        dataset = DatasetLoader.load_from_name(dataset_parameters)
+        dataset = DatasetLoader.load(dataset_parameters)
 
         trainer = Trainer(dataset)
 
         for extracted_forest_size in args.extracted_forest_size:
+            logger.debug('extracted_forest_size={}'.format(extracted_forest_size))
             sub_models_dir = models_dir + os.sep + 'extracted_forest_size' + os.sep + str(extracted_forest_size)
             try:
                 os.makedirs(sub_models_dir)
@@ -98,7 +105,7 @@ if __name__ == "__main__":
                 forest_size=args.forest_size,
                 extracted_forest_size=extracted_forest_size,
                 seed=seed,
-                normalize=args.normalize
+                normalize=normalize
             )
             model_parameters.save(sub_models_dir, experiment_id)