Move invalid value checking to load_one_graph

DeepRank · Aug 4, 2023 · 2150ef3 · 2150ef3
1 parent 7fa7bfa
commit 2150ef3
Show file tree

Hide file tree

Showing 2 changed files with 22 additions and 25 deletions.
diff --git a/deeprankcore/dataset.py b/deeprankcore/dataset.py
@@ -286,30 +286,16 @@ def hdf5_to_pandas( # noqa: MC0001, pylint: disable=too-many-locals
                                 df_dict[feat + '_' + str(i)] = [f[entry_name][feat_type][feat][:][:,i] for entry_name in entry_names]
                                 #apply transformation for each channel in this feature
                                 if transform:
-                                    for row in df_dict[feat + '_' + str(i)]:
-                                        if(np.isnan(transform(row)).any()):
-                                            raise ValueError(f"NaN value occurs when applying {self.features_transform} for feature {feat}."
-                                                            "Please set an appropriate features_transform function.")
-                                        if(np.isneginf(transform(row)).any() or np.isposinf(transform(row)).any()):
-                                            raise ValueError(f"Infinte value occurs when applying {self.features_transform} for feature {feat}."
-                                                            "Please set an appropriate features_transform function.")
-                                        df_dict[feat + '_' + str(i)] = transform(row)
+                                    df_dict[feat + '_' + str(i)] = [transform(row) for row in df_dict[feat + '_' + str(i)]]
                         else:
                             df_dict[feat] = [
                                 f[entry_name][feat_type][feat][:]
                                 if f[entry_name][feat_type][feat][()].ndim == 1
                                 else f[entry_name][feat_type][feat][()] for entry_name in entry_names]
                             #apply transformation
                             if transform:
-                                for row in df_dict[feat]:
-                                    if(np.isnan(transform(row)).any()):
-                                        raise ValueError(f"NaN value occurs when applying {self.features_transform} for feature {feat}."
-                                                        "Please set an appropriate features_transform function.")
-                                    if(np.isneginf(transform(row)).any() or np.isposinf(transform(row)).any()):
-                                        raise ValueError(f"Infinte value occurs when applying {self.features_transform} for feature {feat}."
-                                                        "Please set an appropriate features_transform function.")    
-                                    df_dict[feat] = transform(row)
-
+                                df_dict[feat]=[transform(row) for row in df_dict[feat]]
+
                 df = pd.DataFrame(data=df_dict)
 
             df_final = pd.concat([df_final, df])
@@ -823,7 +809,11 @@ def load_one_graph(self, fname: str, entry_name: str)  -> Data: # pylint: disabl
 
                         # apply transformation
                         if transform:
-                            vals = transform(vals)
+                            with warnings.catch_warnings(record=True) as w:
+                                vals = transform(vals)
+                                if (len(w)):
+                                    raise ValueError(f"Invalid value occurs when applying {self.features_transform} for feature {feat}."
+                                                    "Please set an appropriate features_transform function.")
 
                         if vals.ndim == 1: # features with only one channel
                             vals = vals.reshape(-1, 1)   
@@ -869,7 +859,11 @@ def load_one_graph(self, fname: str, entry_name: str)  -> Data: # pylint: disabl
 
                         # apply transformation
                         if transform:
-                            vals = transform(vals)
+                            with warnings.catch_warnings(record=True) as w:
+                                vals = transform(vals)
+                                if (len(w)):
+                                    raise ValueError(f"Invalid value occurs when applying {self.features_transform} for feature {feat}."
+                                                    "Please set an appropriate features_transform function.")
 
                         if vals.ndim == 1:
                             vals = vals.reshape(-1, 1)

diff --git a/tests/test_dataset.py b/tests/test_dataset.py
@@ -913,14 +913,17 @@ def test_invalid_value_features_transform(self):
 
         hdf5_path = "tests/data/hdf5/train.hdf5"
         features_transform = {'all': {'transform': lambda t: np.log(t+10), 'standardize': True}}
+
+        transf_dataset = GraphDataset(
+            hdf5_path = hdf5_path,
+            target = 'binary',
+            features_transform = features_transform
+        )
         with pytest.raises(ValueError):
             with warnings.catch_warnings():
-                warnings.filterwarnings('ignore', r'invalid value encountered in log')
-                GraphDataset(
-                    hdf5_path = hdf5_path,
-                    target = 'binary',
-                    features_transform = features_transform
-                )
+                warnings.filterwarnings('ignore', r'divide by zero encountered in divide')
+                _compute_features_with_get(hdf5_path, transf_dataset)     
+
     def test_inherit_info_training_graphdataset(self):
         hdf5_path = "tests/data/hdf5/train.hdf5"
         feature_transform = {'all': {'transform': None, 'standardize': True}}