Big changes: Create intermediate classes OMPForest and SingleOmpForest for...

Big changes: Create intermediate classes OMPForest and SingleOmpForest for code factoring: share code between OmpForestRegressor and OmpForestBinaryClassifer. Remove set_wweights and set_forest which are not relevant anymore. load function from model_factory isn't trustfull now: raises an error. TODO: multiclass classifier

code/bolsonaro/data/dataset_loader.py

@@ -8,6 +8,13 @@ from sklearn.datasets import fetch_olivetti_faces, fetch_20newsgroups, \
 from sklearn.model_selection import train_test_split
 from sklearn import preprocessing
 
+def change_binary_func_load(base_load_function):
+    def func_load(return_X_y):
+        X, y = base_load_function(return_X_y=return_X_y)
+        assert len(set(y).difference({0, 1})) == 0, "Classes for binary classifier should be {-1, +1}"
+        y[y==0] = -1
+        return X, y
+    return func_load
 
 class DatasetLoader(object):
 
@@ -33,7 +40,7 @@ class DatasetLoader(object):
             dataset_loading_func = load_wine
             task = Task.CLASSIFICATION
         elif name == 'breast_cancer':
-            dataset_loading_func = load_breast_cancer
+            dataset_loading_func = change_binary_func_load(load_breast_cancer)
             task = Task.CLASSIFICATION
         elif name == 'olivetti_faces':
             dataset_loading_func = fetch_olivetti_faces

code/bolsonaro/models/model_factory.py

-from bolsonaro.models.omp_forest_classifier import OmpForestClassifier
+from bolsonaro.models.omp_forest_classifier import OmpForestBinaryClassifier
 from bolsonaro.models.omp_forest_regressor import OmpForestRegressor
 from bolsonaro.data.task import Task
 from bolsonaro.models.model_parameters import ModelParameters
@@ -12,7 +12,7 @@ class ModelFactory(object):
     @staticmethod
     def build(task, model_parameters):
         if task == Task.CLASSIFICATION:
-            model_func = OmpForestClassifier
+            model_func = OmpForestBinaryClassifier
         elif task == Task.REGRESSION:
             model_func = OmpForestRegressor
         else:
@@ -21,8 +21,10 @@ class ModelFactory(object):
 
     @staticmethod
     def load(task, directory_path, experiment_id, model_raw_results):
+        raise NotImplementedError
         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)
+        # todo faire ce qu'il faut ici pour rétablir correctement le modèle
+        # model.set_forest(model_raw_results.forest)
+        # model.set_weights(model_raw_results.weights)
         return model

code/bolsonaro/models/model_raw_results.py

@@ -6,13 +6,12 @@ import datetime
 
 class ModelRawResults(object):
 
-    def __init__(self, forest, weights, training_time,
+    def __init__(self, model_object, 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._model_object = model_object
         self._training_time = training_time
         self._datetime = datetime
         self._train_score = train_score
@@ -24,12 +23,8 @@ class ModelRawResults(object):
         self._test_score_regressor = test_score_regressor
     
     @property
-    def forest(self):
-        return self._forest
-
-    @property
-    def weights(self):
-        return self._weights
+    def model_object(self):
+        return self.model_object
 
     @property
     def training_time(self):

code/bolsonaro/models/omp_forest.py

+from abc import abstractmethod, ABCMeta
+
+import numpy as np
+from sklearn.linear_model import OrthogonalMatchingPursuit
+
+from bolsonaro import LOG_PATH
+from bolsonaro.error_handling.logger_factory import LoggerFactory
+from sklearn.base import BaseEstimator
+
+
+class OmpForest(BaseEstimator, metaclass=ABCMeta):
+    def __init__(self, models_parameters, base_forest_estimator):
+        self._base_forest_estimator = base_forest_estimator
+        self._models_parameters = models_parameters
+        self._logger = LoggerFactory.create(LOG_PATH, __name__)
+
+    @property
+    def models_parameters(self):
+        return self._models_parameters
+
+    def score_base_estimator(self, X, y):
+        return self._base_forest_estimator.score(X, y)
+
+
+    def _base_estimator_predictions(self, X):
+        return np.array([tree.predict(X) for tree in self._base_forest_estimator.estimators_]).T
+
+    @property
+    def forest(self):
+        return self._base_forest_estimator.estimators_
+
+    # sklearn baseestimator api methods
+    @abstractmethod
+    def fit(self, X_forest, y_forest, X_omp, y_omp):
+        pass
+
+    @abstractmethod
+    def predict(self, X):
+        pass
+
+    @abstractmethod
+    def score(self, X, y):
+        pass
+
+class SingleOmpForest(OmpForest):
+    def __init__(self, models_parameters, base_forest_estimator):
+        # fit_intercept shouldn't be set to False as the data isn't necessarily centered here
+        # normalization is handled outsite OMP
+        self._omp = OrthogonalMatchingPursuit(
+            n_nonzero_coefs=models_parameters.extracted_forest_size,
+            fit_intercept=True, normalize=False)
+
+        super().__init__(models_parameters, base_forest_estimator)
+
+    def fit(self, X_forest, y_forest, X_omp, y_omp):
+        self._base_forest_estimator.fit(X_forest, y_forest)
+        self._extract_subforest(X_omp, y_omp)  # type: OrthogonalMatchingPursuit
+        return self
+
+    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._base_estimator_predictions(X)
+
+        if self._models_parameters.normalize_D:
+            # question: maybe consider other kinds of normalization.. centering?
+            self._logger.debug("Compute norm of predicted vectors on X")
+            self._forest_norms = np.linalg.norm(D, axis=0)
+            D /= self._forest_norms
+
+        self._logger.debug("Apply orthogonal maching pursuit on forest for {} extracted trees."
+                           .format(self._models_parameters.extracted_forest_size))
+
+        return self._omp.fit(D, y)
+
+    def predict(self, X):
+        """
+        Apply the SingleOmpForest to X.
+
+        Make all the base tree predictions then apply the OMP weights for pruning.
+
+        :param X:
+        :return:
+        """
+        forest_predictions = self._base_estimator_predictions(X)
+
+        if self._models_parameters.normalize_D:
+            forest_predictions /= self._forest_norms
+
+        if self._models_parameters.normalize_weights:
+            # we can normalize weights (by their sum) so that they sum to 1
+            # and they can be interpreted as impact percentages for interpretability.
+            # this necessits to remove the (-) in weights, e.g. move it to the predictions (use unsigned_coef)
+
+            # question: je comprend pas le truc avec nonszero?
+            # predictions = self._omp.predict(forest_predictions) * (1 / (np.sum(self._omp.coef_) / len(np.nonzero(self._omp.coef_))))
+            coef_signs = np.sign(self._omp.coef_)[np.newaxis, :] # add axis to make sure it will be broadcasted line-wise (there might be a confusion when forest_prediction is square)
+            unsigned_coef = (coef_signs * self._omp.coef_).squeeze()
+            intercept = self._omp.intercept_
+
+            adjusted_forest_predictions = forest_predictions * coef_signs
+            predictions = adjusted_forest_predictions.dot(unsigned_coef) + intercept
+
+        else:
+            predictions = self._omp.predict(forest_predictions)
+
+        return predictions
\ No newline at end of file

code/bolsonaro/models/omp_forest_classifier.py

+from collections import namedtuple
+
 from sklearn.base import BaseEstimator
 from sklearn.ensemble import RandomForestClassifier
+from sklearn.linear_model import OrthogonalMatchingPursuit
+
+from bolsonaro import LOG_PATH
+from bolsonaro.error_handling.logger_factory import LoggerFactory
+from bolsonaro.models.omp_forest import OmpForest, SingleOmpForest
+import numpy as np
+
+class OmpForestBinaryClassifier(SingleOmpForest):
+
+    DEFAULT_SCORE_METRIC = 'indicator'
+
+    def __init__(self, models_parameters):
+        estimator = RandomForestClassifier(n_estimators=models_parameters.forest_size,
+                                           random_state=models_parameters.seed, n_jobs=-1)
+        super().__init__(models_parameters, estimator)
+
+    def _check_classes(self, y):
+        assert len(set(y).difference({-1, 1})) == 0, "Classes for binary classifier should be {-1, +1}"
+
+    def fit(self, X_forest, y_forest, X_omp, y_omp):
+        self._check_classes(y_forest)
+        self._check_classes(y_omp)
+
+        return super().fit(X_forest, y_forest, X_omp, y_omp)
+
+
+    def score(self, X, y, metric=DEFAULT_SCORE_METRIC):
+        """
+        Evaluate OMPForestClassifer on (`X`, `y`) using `metric`
+
+        :param X:
+        :param y:
+        :param metric: might be "indicator"
+        :return:
+        """
+        predictions = self.predict(X)
+
+        if metric == 'indicator':
+            evaluation = np.abs(np.mean(np.abs(np.sign(predictions) - y) - 1))
+        else:
+            raise ValueError("Unsupported metric '{}'.".format(metric))
+
+        return evaluation
+
+
+class OmpForestMulticlassClassifier(BaseEstimator):
+    def __init__(self, models_parameters):
+        self._models_parameters = models_parameters
+        self._base_forest_estimators = RandomForestClassifier(n_estimators=models_parameters.forest_size,
+                                           random_state=models_parameters.seed, n_jobs=-1)
+        self._logger = LoggerFactory.create(LOG_PATH, __name__)
 
 
-class OmpForestClassifier(BaseEstimator):
 
-    def __init__(self):
-        raise ValueError('Classification tasks are not supported for now')
 
-    def fit(self, X, y):
-        pass
+if __name__ == "__main__":
+    forest = RandomForestClassifier(n_estimators=10)
+    X = np.random.rand(10, 5)
+    y = np.random.choice([-1, +1], 10)
+    forest.fit(X, y)
+    print(forest.predict(np.random.rand(10, 5)))
\ No newline at end of file

code/bolsonaro/models/omp_forest_regressor.py

-from bolsonaro import LOG_PATH
-from bolsonaro.error_handling.logger_factory import LoggerFactory
+
 
 from sklearn.ensemble import RandomForestRegressor
-from sklearn.linear_model import OrthogonalMatchingPursuit
-from sklearn.base import BaseEstimator
 import numpy as np
 
+from bolsonaro.models.omp_forest import SingleOmpForest
+
 
-class OmpForestRegressor(BaseEstimator):
+class OmpForestRegressor(SingleOmpForest):
 
     DEFAULT_SCORE_METRIC = 'mse'
 
     def __init__(self, models_parameters):
-        self._regressor = RandomForestRegressor(n_estimators=models_parameters.forest_size,
-            random_state=models_parameters.seed, n_jobs=-1)
-        self._models_parameters = models_parameters
-        self._logger = LoggerFactory.create(LOG_PATH, __name__)
-
-    @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_forest, y_forest, X_omp, y_omp):
-        self._forest = self._train_forest(X_forest, y_forest)
-        self._omp = self._extract_subforest(X_omp, y_omp)
-        self._weights = self._omp.coef_
-        return self
-
-    def score_regressor(self, X, y):
-        return self._regressor.score(X, y)
-
-    def predict(self, X):
-        """
-        Apply the OMPForestRegressor to X.
-
-        :param X:
-        :return:
-        """
-        forest_predictions = self._forest_prediction(X)
+        estimator = RandomForestRegressor(n_estimators=models_parameters.forest_size,
+                                          random_state=models_parameters.seed, n_jobs=-1)
 
-        if self._models_parameters.normalize_D:
-            forest_predictions /= self._forest_norms
 
-        predictions = self._omp.predict(forest_predictions) * (1 / (np.sum(self._omp.coef_) / len(np.nonzero(self._omp.coef_)))) \
-            if self._models_parameters.normalize_weights \
-            else self._omp.predict(forest_predictions)
+        super().__init__(models_parameters, estimator)
 
-        return predictions
 
     def score(self, X, y, metric=DEFAULT_SCORE_METRIC):
         """
@@ -79,39 +34,4 @@ class OmpForestRegressor(BaseEstimator):
         else:
             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_D:
-            # 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))
-        return omp.fit(D, y)
-
-    def _forest_prediction(self, X):
-        return np.array([tree.predict(X) for tree in self._forest]).T
+        return evaluation
\ No newline at end of file

code/bolsonaro/trainer.py

@@ -8,12 +8,26 @@ import numpy as np
 
 
 class Trainer(object):
+    """
+    Class capable of fitting any model object to some prepared data then evaluate and save results through the `train` method.
+    """
 
     def __init__(self, dataset):
+        """
+
+        :param dataset: Object with X_train, y_train, X_dev, y_dev, X_test and Y_test attributes
+        """
         self._dataset = dataset
         self._logger = LoggerFactory.create(LOG_PATH, __name__)
 
     def train(self, model, models_dir):
+        """
+
+        :param model: Object with
+        :param models_dir: Where the results will be saved
+        :return:
+        """
+        # todo cette fonction ne fait pas que "train", elle choisit le jeu de données, train et evalue le modèle -> nom à changer
         self._logger.debug('Training model using train set...')
         begin_time = time.time()
 
@@ -45,16 +59,18 @@ class Trainer(object):
         )
         end_time = time.time()
 
-        ModelRawResults(
-            forest=model.forest,
-            weights=model.weights,
+        results = ModelRawResults(
+            model_object=model,
             training_time=end_time - begin_time,
             datetime=datetime.datetime.now(),
             train_score=model.score(self._dataset.X_train, self._dataset.y_train),
             dev_score=model.score(self._dataset.X_dev, self._dataset.y_dev),
             test_score=model.score(self._dataset.X_test, self._dataset.y_test),
             score_metric=model.DEFAULT_SCORE_METRIC, # TODO: resolve the used metric in a proper way
-            train_score_regressor=model.score_regressor(self._dataset.X_train, self._dataset.y_train),
-            dev_score_regressor=model.score_regressor(self._dataset.X_dev, self._dataset.y_dev),
-            test_score_regressor=model.score_regressor(self._dataset.X_test, self._dataset.y_test)
-        ).save(models_dir)
+            train_score_regressor=model.score_base_estimator(self._dataset.X_train, self._dataset.y_train),
+            dev_score_regressor=model.score_base_estimator(self._dataset.X_dev, self._dataset.y_dev),
+            test_score_regressor=model.score_base_estimator(self._dataset.X_test, self._dataset.y_test)
+        )
+        results.save(models_dir)
+        self._logger.info("Base performance on test: {}".format(results.test_score_regressor))
+        self._logger.info("Performance on test: {}".format(results.test_score))

code/train.py

@@ -18,9 +18,20 @@ import json
 
 
 def process_job(seed, parameters, experiment_id):
+    """
+    Experiment function.
+
+    Will be used as base function for worker in multithreaded application.
+
+    :param seed:
+    :param parameters:
+    :param experiment_id:
+    :return:
+    """
     logger = LoggerFactory.create(LOG_PATH, 'training_seed{}_ti{}'.format(
         seed, threading.get_ident()))
     logger.info('seed={}'.format(seed))
+
     seed_str = str(seed)
     experiment_id_str = str(experiment_id)
     models_dir = parameters['models_dir'] + os.sep + experiment_id_str + os.sep + 'seeds' + \
@@ -35,12 +46,12 @@ def process_job(seed, parameters, experiment_id):
         dataset_normalizer=parameters['dataset_normalizer']
     )
     dataset_parameters.save(models_dir, experiment_id_str)
-
     dataset = DatasetLoader.load(dataset_parameters)
 
     trainer = Trainer(dataset)
 
     for extracted_forest_size in parameters['extracted_forest_size']:
+        # question if training is too long, one may also split experiments for different forest sizes into different workers
         logger.info('extracted_forest_size={}'.format(extracted_forest_size))
         sub_models_dir = models_dir + os.sep + 'extracted_forest_size' + os.sep + str(extracted_forest_size)
         pathlib.Path(sub_models_dir).mkdir(parents=True, exist_ok=True)
@@ -62,7 +73,7 @@ def process_job(seed, parameters, experiment_id):
 
 if __name__ == "__main__":
     # get environment variables in .env
-    load_dotenv(find_dotenv('.env.example'))
+    load_dotenv(find_dotenv('.env'))
 
     DEFAULT_EXPERIMENT_CONFIGURATION_PATH = 'experiments'
     DEFAULT_DATASET_NAME = 'boston'
@@ -109,6 +120,7 @@ if __name__ == "__main__":
 
     logger = LoggerFactory.create(LOG_PATH, os.path.basename(__file__))
 
+    # The number of tree to extract from forest (K)
     parameters['extracted_forest_size'] = parameters['extracted_forest_size'] \
         if type(parameters['extracted_forest_size']) == list \
         else [parameters['extracted_forest_size']]
@@ -116,6 +128,7 @@ if __name__ == "__main__":
     if parameters['seeds'] != None and parameters['random_seed_number'] > 1:
         logger.warning('seeds and random_seed_number parameters are both specified. Seeds will be used.')    
 
+    # Seeds are either provided as parameters or generated at random
     seeds = parameters['seeds'] if parameters['seeds'] is not None \
         else [random.randint(begin_random_seed_range, end_random_seed_range) \
         for i in range(parameters['random_seed_number'])]