Merge pull request #116 from BioSystemsUM/issue_105

[FIX] saving models with dots in the path to save the model + [BUMP] …

Author: jcapels

setup.cfg

@@ -1,6 +1,6 @@
 [metadata]
 name = deepmol
-version = 1.1.14
+version = 1.1.15
 description = DeepMol: a python-based machine and deep learning framework for drug discovery
 keywords = machine-learning, deep-learning, cheminformatics, drug-discovery
 author = DeepMol Team

src/deepmol/__init__.py

@@ -1,2 +1,2 @@
 
-__version__ = '1.1.14'
+__version__ = '1.1.15'

src/deepmol/models/sklearn_models.py

@@ -190,8 +190,8 @@ def save(self, folder_path: str = None):
 
         save_to_disk(self.model, model_path)
 
-        # change file path to keep the extension but add _params
-        parameters_file_path = model_path.split('.')[0] + '_params.' + model_path.split('.')[1]
+        base, ext = os.path.splitext(model_path)
+        parameters_file_path = f"{base}_params{ext}"
         save_to_disk(self.parameters_to_save, parameters_file_path)
 
     @classmethod
@@ -213,7 +213,8 @@ def load(cls, folder_path: str, **kwargs) -> 'SklearnModel':
         model_path = cls.get_model_filename(folder_path)
         model = load_from_disk(model_path)
         # change file path to keep the extension but add _params
-        parameters_file_path = ".".join(model_path.split('.')[:-1]) + '_params.' + model_path.split('.')[-1]
+        base, ext = os.path.splitext(model_path)
+        parameters_file_path = f"{base}_params{ext}"
         params = load_from_disk(parameters_file_path)
         instance = cls(model=model, model_dir=model_path, **params)
         return instance

tests/unit_tests/models/test_deepchem_model.py

@@ -1,4 +1,5 @@
 import os
+import shutil
 from unittest import TestCase
 from unittest.mock import MagicMock
 
@@ -109,6 +110,25 @@ def test_save(self):
         new_predictions = model_graph_loaded.predict(ds_test)
         self.assertTrue(np.array_equal(test_preds, new_predictions))
 
+    def test_save_load_with_dots(self):
+        ds_train = self.multitask_dataset
+        ds_train.X = ConvMolFeaturizer().featurize([MolFromSmiles('CCC')] * 10)
+        ds_test = self.multitask_dataset_test
+        ds_test.X = ConvMolFeaturizer().featurize([MolFromSmiles('CCC')] * 10)
+
+        model_graph = DeepChemModel(GraphConvModel, n_tasks=ds_train.n_tasks, mode='classification', epochs=3)
+
+        model_graph.fit(ds_train)
+        test_preds = model_graph.predict(ds_test)
+
+        model_graph.save("../test_model")
+        model_graph_loaded = DeepChemModel.load("../test_model")
+        self.assertEqual(model_graph.n_tasks, ds_train.n_tasks)
+        self.assertEqual(model_graph.epochs, 3)
+        new_predictions = model_graph_loaded.predict(ds_test)
+        self.assertTrue(np.array_equal(test_preds, new_predictions))
+        shutil.rmtree("../test_model")
+
     def test_cross_validate(self):
         ds_train = self.binary_dataset
         ds_train.X = ConvMolFeaturizer().featurize([MolFromSmiles('CCC')] * 100)

tests/unit_tests/models/test_keras_model.py

@@ -96,6 +96,40 @@ def test_save_model(self):
 
         shutil.rmtree("test_model")
 
+    def test_save_model_with_dots(self):
+        model = KerasModel(model_builder=make_cnn_model,
+                           epochs=2, input_dim=self.binary_dataset.X.shape[1])
+        model.fit(self.binary_dataset)
+
+        first_predictions = model.predict(self.binary_dataset_test)
+
+        model.save("../test_model")
+        loaded_model = KerasModel.load("../test_model")
+        self.assertEqual(2, loaded_model.epochs)
+        self.assertEqual(50, loaded_model.parameters_to_save["input_dim"])
+        loaded_model_predictions = loaded_model.predict(self.binary_dataset_test)
+
+        assert np.array_equal(first_predictions, loaded_model_predictions)
+
+        shutil.rmtree("../test_model")
+
+    def test_load_models_with_dots(self):
+        model = KerasModel(model_builder=make_cnn_model,
+                           epochs=2, input_dim=self.binary_dataset.X.shape[1])
+        model.fit(self.binary_dataset)
+
+        first_predictions = model.predict(self.binary_dataset_test)
+
+        model.save("../test_model")
+        loaded_model = KerasModel.load("../test_model")
+        self.assertEqual(2, loaded_model.epochs)
+        self.assertEqual(50, loaded_model.parameters_to_save["input_dim"])
+        loaded_model_predictions = loaded_model.predict(self.binary_dataset_test)
+
+        assert np.array_equal(first_predictions, loaded_model_predictions)
+
+        shutil.rmtree("../test_model")
+
     def test_baseline_models(self):
         model_kwargs = {'input_dim': 50}
         keras_kwargs = {}

tests/unit_tests/models/test_sklearn_models.py

@@ -150,11 +150,12 @@ def test_save_model(self):
 
         self.assertEqual("classification", model.mode)
 
-        with self.assertRaises(ValueError):
-            model.save("test_model.params.pkl")
-
-        with self.assertRaises(ValueError):
-            model.save("test_model.params.joblib")
+        model.save("test_model.params.pkl")
+        self.assertTrue(os.path.exists("test_model.params.pkl"))
+        shutil.rmtree("test_model.params.pkl")
+        model.save("test_model.params.joblib")
+        self.assertTrue(os.path.exists("test_model.params.joblib"))
+        shutil.rmtree("test_model.params.joblib")
 
         rf = RandomForestClassifier()
         model = SklearnModel(model=rf, mode="classification", model_dir="test_model")
@@ -164,6 +165,34 @@ def test_save_model(self):
 
         shutil.rmtree("test_model")
 
+    def test_save_model_with_dots(self):
+        rf = RandomForestClassifier()
+        model = SklearnModel(model=rf, mode="classification", model_dir="test_model")
+        model.fit(self.binary_dataset)
+        model.save("../test_model")
+        self.assertTrue(os.path.exists("../test_model"))
+        shutil.rmtree("../test_model")
+
+    def test_load_models_with_dots(self):
+        rf = RandomForestClassifier()
+        model = SklearnModel(model=rf, mode="classification", model_dir="../test_model")
+        model.fit(self.binary_dataset)
+        model.save("../test_model")
+        self.assertTrue(os.path.exists("../test_model"))
+
+        predictions_1 = model.predict(self.binary_dataset_test)
+
+        new_model = SklearnModel.load("../test_model")
+        y_test = new_model.predict(self.binary_dataset_test)
+        self.assertEqual(len(y_test), len(self.binary_dataset_test.y))
+        self.assertIsInstance(new_model, SklearnModel)
+        self.assertEqual("classification", new_model.mode)
+
+        assert np.array_equal(predictions_1, y_test)
+
+        shutil.rmtree("../test_model")
+
+
     def test_load_model(self):
         rf = RandomForestClassifier()
         model = SklearnModel(model=rf, mode="classification")