Merge pull request #36 from wwu-mmll/develop

Develop

Author: ksarink

examples/heart_failure/heart_failure.py

@@ -4,16 +4,16 @@
 from photonai.optimization import FloatRange, IntegerRange
 
 # setup training and test workflow
-my_pipe = Hyperpipe('heart_failure_lasso',
-                    outer_cv=ShuffleSplit(n_splits=100, test_size=0.2),
+my_pipe = Hyperpipe('heart_failure',
+                    outer_cv=ShuffleSplit(n_splits=10, test_size=0.2),
                     inner_cv=KFold(n_splits=10, shuffle=True),
                     use_test_set=False,
                     metrics=['balanced_accuracy', 'f1_score', 'matthews_corrcoef',
                              'sensitivity', 'specificity'],
                     best_config_metric='f1_score',
                     optimizer='switch',
                     optimizer_params={'name': 'sk_opt', 'n_configurations': 10},
-                    project_folder='./tmpv2',
+                    project_folder='./tmp',
                     cache_folder='./cache',
                     verbosity=0)
 

photonai/base/hyperpipe.py

@@ -1167,6 +1167,79 @@ def score(self, data: np.ndarray, y: np.ndarray, **kwargs) -> np.ndarray:
             scorer = Scorer.create(self.optimization.best_config_metric)
             return scorer(y, predictions)
 
+    def _calculate_permutation_importances(self, **kwargs):
+        """
+        extracted function from get_feature_importance to improve unit testing
+        """
+
+        importance_list = {'mean': list(), 'std': list()}
+
+        def train_and_get_fimps(pipeline, train_idx, test_idx, data_X, data_y, data_kwargs, fold_str):
+
+            train_X, train_y, train_kwargs = PhotonDataHelper.split_data(data_X, data_y, data_kwargs,
+                                                                         indices=train_idx)
+
+            test_X, test_y, test_kwargs = PhotonDataHelper.split_data(data_X, data_y, data_kwargs,
+                                                                      indices=test_idx)
+
+            # fit fold's best model (again) -> to obtain that model's feature importances
+            logger.photon_system_log("Permutation Importances: Fitting model for " + fold_str)
+            pipeline.fit(train_X, train_y, **train_kwargs)
+
+            # get feature importances
+            logger.photon_system_log("Permutation Importances: Calculating performances for " + fold_str)
+            perm_imps = permutation_importance(pipeline, test_X, test_y, **kwargs)
+
+            # store into list
+            importance_list['mean'].append(perm_imps["importances_mean"])
+            importance_list['std'].append(perm_imps["importances_std"])
+
+            return perm_imps
+
+        for outer_fold in self.results.outer_folds:
+
+            if outer_fold.best_config is None:
+                raise ValueError("Could not find a best config for outer fold " + str(outer_fold.fold_nr))
+
+            pipe_copy = self.optimum_pipe.copy_me()
+
+            # set pipe to config
+            pipe_copy.set_params(**outer_fold.best_config.config_dict)
+
+            if not self.results.hyperpipe_info.eval_final_performance:
+                no_outer_cv_indices = False
+                if outer_fold.best_config.best_config_score is None:
+                    no_outer_cv_indices = True
+                if outer_fold.best_config.best_config_score.training is None or not outer_fold.best_config.best_config_score.training.indices:
+                    no_outer_cv_indices = True
+
+                if no_outer_cv_indices:
+                    data_to_split, y_to_split, kwargs_to_split = self.data.X, self.data.y, self.data.kwargs
+                else:
+
+                    logger.photon_system_log("Permutation Importances: Using inner_cv folds.")
+
+                    # get outer fold data
+                    idx = outer_fold.best_config.best_config_score.training.indices
+                    data_to_split, y_to_split, kwargs_to_split = PhotonDataHelper.split_data(self.data.X,
+                                                                                             self.data.y,
+                                                                                             self.data.kwargs,
+                                                                                             indices=idx)
+
+                for inner_fold in outer_fold.best_config.inner_folds:
+                    train_and_get_fimps(pipe_copy,
+                                        inner_fold.training.indices, inner_fold.validation.indices,
+                                        data_to_split, y_to_split, kwargs_to_split,
+                                        "inner fold " + str(inner_fold.fold_nr))
+
+            else:
+                train_and_get_fimps(pipe_copy,
+                                    outer_fold.best_config.best_config_score.training.indices,
+                                    outer_fold.best_config.best_config_score.validation.indices,
+                                    self.data.X, self.data.y, self.data.kwargs, "outer fold " + str(outer_fold.fold_nr))
+
+        return importance_list
+
     def get_permutation_feature_importances(self, **kwargs):
         """
         Fits a model for the best config of each outer fold (using the training data of that fold).
@@ -1191,41 +1264,13 @@ def get_permutation_feature_importances(self, **kwargs):
 
         """
 
-        importance_list = {'mean': list(), 'std': list()}
-        pipe_copy = self.optimum_pipe.copy_me()
         logger.photon_system_log("")
         logger.photon_system_log("Computing permutation importances. This may take a while.")
         logger.stars()
-        for outer_fold in self.results.outer_folds:
-
-            if outer_fold.best_config.best_config_score is None:
-                raise ValueError("Cannot compute permutation importances when use_test_set is false")
-
-
-            # prepare data
-            train_indices = outer_fold.best_config.best_config_score.training.indices
-            test_indices = outer_fold.best_config.best_config_score.validation.indices
-
-            train_X, train_y, train_kwargs = PhotonDataHelper.split_data(self.data.X,
-                                                                         self.data.y,
-                                                                         self.data.kwargs,
-                                                                         indices=train_indices)
-
-            test_X, test_y, test_kwargs = PhotonDataHelper.split_data(self.data.X,
-                                                                      self.data.y,
-                                                                      self.data.kwargs,
-                                                                      indices=test_indices)
-            # set pipe to config
-            pipe_copy.set_params(**outer_fold.best_config.config_dict)
-            logger.photon_system_log("Permutation Importances: Fitting model for outer fold " + str(outer_fold.fold_nr))
-            pipe_copy.fit(train_X, train_y, **train_kwargs)
-
-            logger.photon_system_log("Permutation Importances: Calculating performances for outer fold "
-                                     + str(outer_fold.fold_nr))
-            outer_fold_perm_imps = permutation_importance(pipe_copy, test_X, test_y, **kwargs)
-            importance_list['mean'].append(outer_fold_perm_imps["importances_mean"])
-            importance_list['std'].append(outer_fold_perm_imps["importances_std"])
-
+        if self.optimum_pipe is None:
+            raise ValueError("Cannot calculate permutation importances when optimum_pipe is None (probably the "
+                             "training and optimization procedure failed)")
+        importance_list = self._calculate_permutation_importances(**kwargs)
         mean_importances = np.mean(np.array(importance_list["mean"]), axis=0)
         std_importances = np.mean(np.array(importance_list["std"]), axis=0)
         logger.stars()

photonai/processing/results_handler.py

@@ -751,9 +751,9 @@ def write_convenience_files(self):
             self.write_predictions_file()
 
     def convert_to_json_serializable(self, value):
-        if isinstance(value, (np.int, np.int32, np.int64)):
+        if isinstance(value, (int, np.int32, np.int64)):
             return int(value)
-        if isinstance(value, (np.float, np.float32, np.float64)):
+        if isinstance(value, (float, np.float32, np.float64)):
             if self.output_settings.reduce_space:
                 return round(float(value), 3)
             return float(value)

test/base_tests/test_hyperpipe.py

@@ -184,7 +184,8 @@ def test_preprocessing(self):
 
     def test_permutation_feature_importances(self):
         hp = Hyperpipe('god',
-                       inner_cv=self.inner_cv_object,
+                       outer_cv=KFold(n_splits=3),
+                       inner_cv=KFold(n_splits=2),
                        metrics=self.metrics,
                        best_config_metric=self.best_config_metric,
                        project_folder=self.tmp_folder_path,
@@ -197,16 +198,60 @@ def test_permutation_feature_importances(self):
         score_element = svc.score(self.__X, self.__y)
         self.assertAlmostEqual(score_photon, score_element)
 
-        permutation_score = hp.get_permutation_feature_importances(n_repeats=5, random_state=0)
-        self.assertTrue("mean" in permutation_score)
-        self.assertTrue("std" in permutation_score)
-        self.assertEqual(permutation_score["mean"].shape, (self.__X.shape[1],))
-        self.assertEqual(permutation_score["std"].shape, (self.__X.shape[1],))
+        # do it on outer folds
+        permutation_list_outer = hp._calculate_permutation_importances(n_repeats=5, random_state=0)
+        self.assertEqual(len(permutation_list_outer["mean"]), 3)
 
+        permutation_score_outer = hp.get_permutation_feature_importances(n_repeats=5, random_state=0)
+        self.assertTrue("mean" in permutation_score_outer)
+        self.assertTrue("std" in permutation_score_outer)
+        self.assertEqual(permutation_score_outer["mean"].shape, (self.__X.shape[1],))
+        self.assertEqual(permutation_score_outer["std"].shape, (self.__X.shape[1],))
+
+        # do it on inner folds but on training sets from outer split
+        hp.cross_validation.use_test_set = False
+        hp.fit(self.__X, self.__y)
+        permutation_list_inner = hp._calculate_permutation_importances(n_repeats=5)
+        self.assertEqual(len(permutation_list_inner["mean"]), 3*2)
+        permutation_score_inner = hp.get_permutation_feature_importances(n_repeats=5)
+        self.assertEqual(permutation_score_inner["mean"].shape, (self.__X.shape[1],))
+        self.assertEqual(permutation_score_inner["std"].shape, (self.__X.shape[1],))
+        # check that validation set permutation importances (inner folds) differ from those of test set (outer folds)
+        self.assertFalse(np.array_equal(permutation_score_outer["mean"], permutation_score_inner["mean"]))
+
+        # do it on inner folds only
+        hp.cross_validation.outer_folds = {}
+        hp.cross_validation.outer_cv = None
         hp.cross_validation.use_test_set = False
         hp.fit(self.__X, self.__y)
+        permutation_list_no_outer = hp._calculate_permutation_importances(n_repeats=5)
+        self.assertEqual(len(permutation_list_no_outer), 2)
+        permutation_score_no_outer = hp.get_permutation_feature_importances(n_repeats=5)
+        self.assertEqual(permutation_score_inner["mean"].shape, (self.__X.shape[1],))
+        self.assertEqual(permutation_score_inner["std"].shape, (self.__X.shape[1],))
+
+        # raise error
+        def fake_metric(y_true, y_pred):
+           return 'a'
+
+        hp = Hyperpipe('god',
+                       outer_cv=KFold(n_splits=3),
+                       inner_cv=KFold(n_splits=2),
+                       metrics=[('fake_metric', fake_metric)],
+                       best_config_metric='fake_metric',
+                       project_folder=self.tmp_folder_path,
+                       verbosity=0)
+        svc = PipelineElement('SVC')
+        hp += svc
+        try:
+            hp.fit(self.__X, self.__y)
+        except Exception as e:
+            # should produce an error so that hp.results.best_config is None.
+            pass
         with self.assertRaises(ValueError):
             hp.get_permutation_feature_importances(n_repeats=5)
+        with self.assertRaises(ValueError):
+            hp._calculate_permutation_importances(n_repeats=5)
 
     def test_estimation_type(self):
         def callback(X, y=None, **kwargs):

test/integration_tests/test_architecture.py

@@ -25,9 +25,8 @@ def setUpClass(cls) -> None:
         if cls.test_multiple_hyperpipes:
             optimizer_list = ['random_grid_search', 'sk_opt']
             eval_final_performance_list = [True, False]
-            inner_cv_list = [KFold(n_splits=3, shuffle=True), ShuffleSplit(n_splits=1, test_size=.2), LeaveOneOut()]
-            outer_cv_list = [None, KFold(n_splits=3, shuffle=True), ShuffleSplit(n_splits=1, test_size=.25),
-                             LeaveOneOut()]
+            inner_cv_list = [KFold(n_splits=3, shuffle=True), ShuffleSplit(n_splits=1, test_size=.2)]
+            outer_cv_list = [None, KFold(n_splits=3, shuffle=True), ShuffleSplit(n_splits=1, test_size=.25)]
             performance_constraints_list = [None]
 
             combinations = list(product(optimizer_list, eval_final_performance_list, inner_cv_list, outer_cv_list,

test/optimization_tests/sk_opt_tests/test_sk_opt.py

@@ -15,7 +15,7 @@ def setUp(self):
                                   PipelineElement("SVC", hyperparameters={'C': FloatRange(1, 100)})]
         self.optimizer = SkOptOptimizer()
         self.optimizer_name = "sk_opt"
-        self.optimizer_params = None
+        self.optimizer_params = {'n_configurations': 10}
 
     def test_ask_advanced(self):
         with self.assertRaises(ValueError):