Loading deepchem/models/keras_model.py +22 −21 Original line number Diff line number Diff line Loading @@ -1020,8 +1020,6 @@ class KerasModel(Model): ---------- source_model: dc.models.KerasModel Source model to create value map from include_top: bool, default True if true, copies the last dense layer """ value_map = {} source_vars = source_model.model.trainable_variables Loading @@ -1044,37 +1042,40 @@ class KerasModel(Model): model_dir=None, include_top=True, **kwargs): """Copies variable values between pretrained model and current model. This method assumes that the variable values of the pretrained model are saved to disk. source_model is a dc.KerasModel with the same architecture as the pretrained model. The variable values are then restored to the source_model. assignment_map is a dictionary mapping the variables from the source model to those in the current model. If no assignment_map is provided, one is made from scratch and assumes the model is composed of several different layers, with the final one being a dense layer. include_top is used to control whether or not the final dense layer is used. The default assignment map is useful in cases where the type of task is different (classification vs regression) and/or number of tasks in the setting. """Copies variable values from a pretrained model. `source_model` can either be a pretrained model or a model with the same architecture. `value_map` is a variable-value dictionary. If no `value_map` is provided, the variable values are restored to the `source_model` from a checkpoint and a default `value_map` is created. `assignment_map` is a dictionary mapping variables from the `source_model` to the current model. If no `assignment_map` is provided, one is made from scratch and assumes the model is composed of several different layers, with the final one being a dense layer. include_top is used to control whether or not the final dense layer is used. The default assignment map is useful in cases where the type of task is different (classification vs regression) and/or number of tasks in the setting. Parameters ---------- source_model: dc.KerasModel, required Model which has the same topology and architecture as the pretrained model. source_model can either be the pretrained model or a dc.KerasModel with the same architecture as the pretrained model. It is used to restore from a checkpoint, if value_map is None and to create a default assignment map if assignment_map is None assignment_map: Dict, default None Dictionary containing layer mapping between source and current model layers Dictionary mapping the source_model variables and current model variables value_map: Dict, default None Dictionary containing source model trainable variables mapped to numpy arrays Dictionary containing source_model trainable variables mapped to numpy arrays. If value_map is None, the values are restored and a default variable map is created using the restored values checkpoint: str, default None the path to the checkpoint file to load. If this is None, the most recent checkpoint will be chosen automatically. Call get_checkpoints() to get a list of all available checkpoints. list of all available checkpoints model_dir: str, default None Restore model from custom model directory if needed include_top: bool, default True if True, copies the weights and bias associated with the final dense layer. Used only when assignment map is None. if True, copies the weights and bias associated with the final dense layer. Used only when assignment map is None """ self._ensure_built() Loading deepchem/models/tests/test_pretrained.py +17 −67 Original line number Diff line number Diff line Loading @@ -36,7 +36,6 @@ class MLP(dc.models.KerasModel): class TestPretrained(unittest.TestCase): def setUp(self): model_dir = "./MLP/" self.feature_dim = 2 self.hidden_layer_size = 10 data_points = 10 Loading @@ -46,22 +45,11 @@ class TestPretrained(unittest.TestCase): self.dataset = dc.data.NumpyDataset(X, y) model = MLP( hidden_layer_size=self.hidden_layer_size, feature_dim=self.feature_dim, model_dir=model_dir, batch_size=10) model.fit(self.dataset, nb_epoch=1000) predictions = np.squeeze(model.predict_on_batch(self.dataset.X)) np.testing.assert_array_almost_equal(self.dataset.y, np.round(predictions)) def test_load_from_pretrained_graph_mode(self): """Tests loading pretrained model in graph mode.""" source_model = MLP( hidden_layer_size=self.hidden_layer_size, feature_dim=self.feature_dim, model_dir=None, batch_size=10) source_model.fit(self.dataset, nb_epoch=1000, checkpoint_interval=0) Loading @@ -78,6 +66,9 @@ class TestPretrained(unittest.TestCase): for idx, dest_var in enumerate(dest_vars): source_var = source_model.model.trainable_variables[idx] assignment_map[source_var] = dest_var if tf.executing_eagerly(): value_map[source_var] = source_var.numpy() else: value_map[source_var] = source_var.eval(session=source_model.session) dest_model.load_from_pretrained( Loading @@ -86,50 +77,25 @@ class TestPretrained(unittest.TestCase): value_map=value_map) for source_var, dest_var in assignment_map.items(): np.testing.assert_array_almost_equal( source_var.eval(session=source_model.session), dest_var.eval(session=dest_model.session)) if tf.executing_eagerly(): source_val = source_var.numpy() dest_val = dest_var.numpy() else: source_val = source_var.eval(session=source_model.session) dest_val = dest_var.eval(session=dest_model.session) np.testing.assert_array_almost_equal(source_val, dest_val) def test_load_from_pretrained_eager_mode(self): """Tests loading pretrained model in eager execution mode.""" with context.eager_mode(): source_model = MLP( hidden_layer_size=self.hidden_layer_size, feature_dim=self.feature_dim, model_dir=None, batch_size=10) source_model.fit(self.dataset, nb_epoch=1000, checkpoint_interval=0) dest_model = MLP( feature_dim=self.feature_dim, hidden_layer_size=self.hidden_layer_size, n_tasks=10) assignment_map = dict() value_map = dict() dest_vars = dest_model.model.trainable_variables[:-2] for idx, dest_var in enumerate(dest_vars): source_var = source_model.model.trainable_variables[idx] assignment_map[source_var] = dest_var value_map[source_var] = source_var.numpy() dest_model.load_from_pretrained( source_model=source_model, value_map=value_map, assignment_map=assignment_map) for source_var, dest_var in assignment_map.items(): np.testing.assert_array_almost_equal(source_var.numpy(), dest_var.numpy()) self.test_load_from_pretrained_graph_mode() def test_restore_equivalency_graph_mode(self): """Test for restore based pretrained model loading in graph mode.""" source_model = MLP( model_dir="./MLP/", feature_dim=self.feature_dim, hidden_layer_size=self.hidden_layer_size) feature_dim=self.feature_dim, hidden_layer_size=self.hidden_layer_size) source_model.fit(self.dataset, nb_epoch=1000) dest_model = MLP( feature_dim=self.feature_dim, hidden_layer_size=self.hidden_layer_size) Loading @@ -138,6 +104,7 @@ class TestPretrained(unittest.TestCase): source_model=source_model, assignment_map=None, value_map=None, model_dir=None, include_top=True) predictions = np.squeeze(dest_model.predict_on_batch(self.dataset.X)) Loading @@ -146,21 +113,4 @@ class TestPretrained(unittest.TestCase): def test_restore_equivalency_eager_mode(self): """Test for restore based pretrained model loading in eager mode.""" with context.eager_mode(): source_model = MLP( model_dir="./MLP/", feature_dim=self.feature_dim, hidden_layer_size=self.hidden_layer_size) dest_model = MLP( feature_dim=self.feature_dim, hidden_layer_size=self.hidden_layer_size) dest_model.load_from_pretrained( source_model=source_model, assignment_map=None, value_map=None, include_top=True) predictions = np.squeeze(dest_model.predict_on_batch(self.dataset.X)) np.testing.assert_array_almost_equal(self.dataset.y, np.round(predictions)) self.test_restore_equivalency_graph_mode() Loading
deepchem/models/keras_model.py +22 −21 Original line number Diff line number Diff line Loading @@ -1020,8 +1020,6 @@ class KerasModel(Model): ---------- source_model: dc.models.KerasModel Source model to create value map from include_top: bool, default True if true, copies the last dense layer """ value_map = {} source_vars = source_model.model.trainable_variables Loading @@ -1044,37 +1042,40 @@ class KerasModel(Model): model_dir=None, include_top=True, **kwargs): """Copies variable values between pretrained model and current model. This method assumes that the variable values of the pretrained model are saved to disk. source_model is a dc.KerasModel with the same architecture as the pretrained model. The variable values are then restored to the source_model. assignment_map is a dictionary mapping the variables from the source model to those in the current model. If no assignment_map is provided, one is made from scratch and assumes the model is composed of several different layers, with the final one being a dense layer. include_top is used to control whether or not the final dense layer is used. The default assignment map is useful in cases where the type of task is different (classification vs regression) and/or number of tasks in the setting. """Copies variable values from a pretrained model. `source_model` can either be a pretrained model or a model with the same architecture. `value_map` is a variable-value dictionary. If no `value_map` is provided, the variable values are restored to the `source_model` from a checkpoint and a default `value_map` is created. `assignment_map` is a dictionary mapping variables from the `source_model` to the current model. If no `assignment_map` is provided, one is made from scratch and assumes the model is composed of several different layers, with the final one being a dense layer. include_top is used to control whether or not the final dense layer is used. The default assignment map is useful in cases where the type of task is different (classification vs regression) and/or number of tasks in the setting. Parameters ---------- source_model: dc.KerasModel, required Model which has the same topology and architecture as the pretrained model. source_model can either be the pretrained model or a dc.KerasModel with the same architecture as the pretrained model. It is used to restore from a checkpoint, if value_map is None and to create a default assignment map if assignment_map is None assignment_map: Dict, default None Dictionary containing layer mapping between source and current model layers Dictionary mapping the source_model variables and current model variables value_map: Dict, default None Dictionary containing source model trainable variables mapped to numpy arrays Dictionary containing source_model trainable variables mapped to numpy arrays. If value_map is None, the values are restored and a default variable map is created using the restored values checkpoint: str, default None the path to the checkpoint file to load. If this is None, the most recent checkpoint will be chosen automatically. Call get_checkpoints() to get a list of all available checkpoints. list of all available checkpoints model_dir: str, default None Restore model from custom model directory if needed include_top: bool, default True if True, copies the weights and bias associated with the final dense layer. Used only when assignment map is None. if True, copies the weights and bias associated with the final dense layer. Used only when assignment map is None """ self._ensure_built() Loading
deepchem/models/tests/test_pretrained.py +17 −67 Original line number Diff line number Diff line Loading @@ -36,7 +36,6 @@ class MLP(dc.models.KerasModel): class TestPretrained(unittest.TestCase): def setUp(self): model_dir = "./MLP/" self.feature_dim = 2 self.hidden_layer_size = 10 data_points = 10 Loading @@ -46,22 +45,11 @@ class TestPretrained(unittest.TestCase): self.dataset = dc.data.NumpyDataset(X, y) model = MLP( hidden_layer_size=self.hidden_layer_size, feature_dim=self.feature_dim, model_dir=model_dir, batch_size=10) model.fit(self.dataset, nb_epoch=1000) predictions = np.squeeze(model.predict_on_batch(self.dataset.X)) np.testing.assert_array_almost_equal(self.dataset.y, np.round(predictions)) def test_load_from_pretrained_graph_mode(self): """Tests loading pretrained model in graph mode.""" source_model = MLP( hidden_layer_size=self.hidden_layer_size, feature_dim=self.feature_dim, model_dir=None, batch_size=10) source_model.fit(self.dataset, nb_epoch=1000, checkpoint_interval=0) Loading @@ -78,6 +66,9 @@ class TestPretrained(unittest.TestCase): for idx, dest_var in enumerate(dest_vars): source_var = source_model.model.trainable_variables[idx] assignment_map[source_var] = dest_var if tf.executing_eagerly(): value_map[source_var] = source_var.numpy() else: value_map[source_var] = source_var.eval(session=source_model.session) dest_model.load_from_pretrained( Loading @@ -86,50 +77,25 @@ class TestPretrained(unittest.TestCase): value_map=value_map) for source_var, dest_var in assignment_map.items(): np.testing.assert_array_almost_equal( source_var.eval(session=source_model.session), dest_var.eval(session=dest_model.session)) if tf.executing_eagerly(): source_val = source_var.numpy() dest_val = dest_var.numpy() else: source_val = source_var.eval(session=source_model.session) dest_val = dest_var.eval(session=dest_model.session) np.testing.assert_array_almost_equal(source_val, dest_val) def test_load_from_pretrained_eager_mode(self): """Tests loading pretrained model in eager execution mode.""" with context.eager_mode(): source_model = MLP( hidden_layer_size=self.hidden_layer_size, feature_dim=self.feature_dim, model_dir=None, batch_size=10) source_model.fit(self.dataset, nb_epoch=1000, checkpoint_interval=0) dest_model = MLP( feature_dim=self.feature_dim, hidden_layer_size=self.hidden_layer_size, n_tasks=10) assignment_map = dict() value_map = dict() dest_vars = dest_model.model.trainable_variables[:-2] for idx, dest_var in enumerate(dest_vars): source_var = source_model.model.trainable_variables[idx] assignment_map[source_var] = dest_var value_map[source_var] = source_var.numpy() dest_model.load_from_pretrained( source_model=source_model, value_map=value_map, assignment_map=assignment_map) for source_var, dest_var in assignment_map.items(): np.testing.assert_array_almost_equal(source_var.numpy(), dest_var.numpy()) self.test_load_from_pretrained_graph_mode() def test_restore_equivalency_graph_mode(self): """Test for restore based pretrained model loading in graph mode.""" source_model = MLP( model_dir="./MLP/", feature_dim=self.feature_dim, hidden_layer_size=self.hidden_layer_size) feature_dim=self.feature_dim, hidden_layer_size=self.hidden_layer_size) source_model.fit(self.dataset, nb_epoch=1000) dest_model = MLP( feature_dim=self.feature_dim, hidden_layer_size=self.hidden_layer_size) Loading @@ -138,6 +104,7 @@ class TestPretrained(unittest.TestCase): source_model=source_model, assignment_map=None, value_map=None, model_dir=None, include_top=True) predictions = np.squeeze(dest_model.predict_on_batch(self.dataset.X)) Loading @@ -146,21 +113,4 @@ class TestPretrained(unittest.TestCase): def test_restore_equivalency_eager_mode(self): """Test for restore based pretrained model loading in eager mode.""" with context.eager_mode(): source_model = MLP( model_dir="./MLP/", feature_dim=self.feature_dim, hidden_layer_size=self.hidden_layer_size) dest_model = MLP( feature_dim=self.feature_dim, hidden_layer_size=self.hidden_layer_size) dest_model.load_from_pretrained( source_model=source_model, assignment_map=None, value_map=None, include_top=True) predictions = np.squeeze(dest_model.predict_on_batch(self.dataset.X)) np.testing.assert_array_almost_equal(self.dataset.y, np.round(predictions)) self.test_restore_equivalency_graph_mode()
