Loading deepchem/models/tensorflow_models/fcnet.py +6 −6 Original line number Diff line number Diff line Loading @@ -19,8 +19,8 @@ from deepchem.models.tensorflow_models import TensorflowRegressor from deepchem.metrics import to_one_hot from deepchem.models.tensorgraph.tensor_graph import TensorGraph from deepchem.models.tensorgraph.layers import Feature, Label, Weights, WeightedError, Dense, Dropout, Reshape, SoftMaxCrossEntropy, L2Loss, Initializer from deepchem.models.tensorgraph.tensor_graph import TensorGraph, TFWrapper from deepchem.models.tensorgraph.layers import Feature, Label, Weights, WeightedError, Dense, Dropout, Reshape, SoftMaxCrossEntropy, L2Loss class TensorGraphMultiTaskClassifier(TensorGraph): def __init__(self, Loading @@ -46,8 +46,8 @@ class TensorGraphMultiTaskClassifier(TensorGraph): for size, weight_stddev, bias_const, dropout in zip(layer_sizes, weight_init_stddevs, bias_init_consts, dropouts): layer = Dense(in_layers=[prev_layer], out_channels=size, activation_fn=tf.nn.relu, weights_initializer=Initializer(tf.truncated_normal_initializer, stddev=weight_stddev), biases_initializer=Initializer(tf.constant_initializer, value=bias_const)) weights_initializer=TFWrapper(tf.truncated_normal_initializer, stddev=weight_stddev), biases_initializer=TFWrapper(tf.constant_initializer, value=bias_const)) if dropout > 0.0: layer = Dropout(dropout, in_layers=[layer]) prev_layer = layer Loading Loading @@ -107,8 +107,8 @@ class TensorGraphMultiTaskRegressor(TensorGraph): for size, weight_stddev, bias_const, dropout in zip(layer_sizes, weight_init_stddevs, bias_init_consts, dropouts): layer = Dense(in_layers=[prev_layer], out_channels=size, activation_fn=tf.nn.relu, weights_initializer=Initializer(tf.truncated_normal_initializer, stddev=weight_stddev), biases_initializer=Initializer(tf.constant_initializer, value=bias_const)) weights_initializer=TFWrapper(tf.truncated_normal_initializer, stddev=weight_stddev), biases_initializer=TFWrapper(tf.constant_initializer, value=bias_const)) if dropout > 0.0: layer = Dropout(dropout, in_layers=[layer]) prev_layer = layer Loading deepchem/models/tensorgraph/__init__.py +1 −1 Original line number Diff line number Diff line from deepchem.models.tensorgraph.tensor_graph import TensorGraphfrom deepchem.models.tensorgraph import models No newline at end of file from deepchem.models.tensorgraph.tensor_graph import TensorGraph, TFWrapperfrom deepchem.models.tensorgraph import models No newline at end of file deepchem/models/tensorgraph/layers.py +0 −20 Original line number Diff line number Diff line Loading @@ -98,26 +98,6 @@ def convert_to_layers(in_layers): return layers class Initializer(object): """This class exists as a workaround for Tensorflow initializers not being picklable.""" def __init__(self, initializer_class, **kwargs): """Create an Initializer for constructing Tensorflow initializers. Parameters ---------- initializer_class: class the type of initializer to create kwargs: any other arguments will be passed on to the Tensorflow initializer's constructor """ self.initializer_class = initializer_class self.kwargs = kwargs def __call__(self): return self.initializer_class(**self.kwargs) class Conv1D(Layer): def __init__(self, width, out_channels, **kwargs): Loading deepchem/models/tensorgraph/tensor_graph.py +36 −7 Original line number Diff line number Diff line Loading @@ -22,7 +22,6 @@ class TensorGraph(Model): def __init__(self, tensorboard=False, tensorboard_log_frequency=100, learning_rate=0.001, batch_size=100, random_seed=None, use_queue=True, Loading @@ -36,8 +35,6 @@ class TensorGraph(Model): Should we log to model_dir data for tensorboard? tensorboard_log_frequency: int How many training batches before logging tensorboard? learning_rate: float learning rate for optimizer batch_size: int default batch size for training and evaluating use_queue: boolean Loading @@ -63,6 +60,7 @@ class TensorGraph(Model): self.loss = None self.built = False self.queue_installed = False self.optimizer = TFWrapper(tf.train.AdamOptimizer, learning_rate=0.001, beta1=0.9, beta2=0.999) # Singular place to hold Tensor objects which don't serialize # These have to be reconstructed on restoring from pickle Loading @@ -81,7 +79,6 @@ class TensorGraph(Model): self.last_checkpoint = None self.use_queue = use_queue self.learning_rate = learning_rate self.batch_size = batch_size self.random_seed = random_seed super(TensorGraph, self).__init__(**kwargs) Loading Loading @@ -368,6 +365,14 @@ class TensorGraph(Model): self._add_layer(layer) self.outputs.append(layer) def set_optimizer(self, optimizer): """Set the optimizer to use for fitting. The argument should be a callable object (most often a TFWrapper) that constructs a Tensorflow optimizer when called. """ self.optimizer = optimizer def save(self): # Remove out_tensor from the object to be pickled must_restore = False Loading Loading @@ -443,9 +448,7 @@ class TensorGraph(Model): elif obj == "FileWriter": self.tensor_objects['FileWriter'] = tf.summary.FileWriter(self.model_dir) elif obj == 'train_op': self.tensor_objects['train_op'] = tf.train.AdamOptimizer( self.learning_rate, beta1=.9, beta2=.999).minimize(self.loss.out_tensor) self.tensor_objects['train_op'] = self.optimizer().minimize(self.loss.out_tensor) elif obj == 'summary_op': self.tensor_objects['summary_op'] = tf.summary.merge_all( key=tf.GraphKeys.SUMMARIES) Loading Loading @@ -518,3 +521,29 @@ def _enqueue_batch(tg, generator, graph, sess, coord): sess.run(tg.input_queue.close_op) coord.num_samples = num_samples coord.request_stop() class TFWrapper(object): """This class exists as a workaround for Tensorflow objects not being picklable. The job of a TFWrapper is to create Tensorflow objects by passing defined arguments to a constructor. There are cases where we really want to store Tensorflow objects of various sorts (optimizers, initializers, etc.), but we can't because they cannot be pickled. So instead we store a TFWrapper that creates the object when needed. """ def __init__(self, tf_class, **kwargs): """Create a TFWrapper for constructing a Tensorflow object. Parameters ---------- tf_class: class the type of object to create kwargs: any other arguments will be passed on to the object's constructor """ self.tf_class = tf_class self.kwargs = kwargs def __call__(self): return self.tf_class(**self.kwargs) Loading
deepchem/models/tensorflow_models/fcnet.py +6 −6 Original line number Diff line number Diff line Loading @@ -19,8 +19,8 @@ from deepchem.models.tensorflow_models import TensorflowRegressor from deepchem.metrics import to_one_hot from deepchem.models.tensorgraph.tensor_graph import TensorGraph from deepchem.models.tensorgraph.layers import Feature, Label, Weights, WeightedError, Dense, Dropout, Reshape, SoftMaxCrossEntropy, L2Loss, Initializer from deepchem.models.tensorgraph.tensor_graph import TensorGraph, TFWrapper from deepchem.models.tensorgraph.layers import Feature, Label, Weights, WeightedError, Dense, Dropout, Reshape, SoftMaxCrossEntropy, L2Loss class TensorGraphMultiTaskClassifier(TensorGraph): def __init__(self, Loading @@ -46,8 +46,8 @@ class TensorGraphMultiTaskClassifier(TensorGraph): for size, weight_stddev, bias_const, dropout in zip(layer_sizes, weight_init_stddevs, bias_init_consts, dropouts): layer = Dense(in_layers=[prev_layer], out_channels=size, activation_fn=tf.nn.relu, weights_initializer=Initializer(tf.truncated_normal_initializer, stddev=weight_stddev), biases_initializer=Initializer(tf.constant_initializer, value=bias_const)) weights_initializer=TFWrapper(tf.truncated_normal_initializer, stddev=weight_stddev), biases_initializer=TFWrapper(tf.constant_initializer, value=bias_const)) if dropout > 0.0: layer = Dropout(dropout, in_layers=[layer]) prev_layer = layer Loading Loading @@ -107,8 +107,8 @@ class TensorGraphMultiTaskRegressor(TensorGraph): for size, weight_stddev, bias_const, dropout in zip(layer_sizes, weight_init_stddevs, bias_init_consts, dropouts): layer = Dense(in_layers=[prev_layer], out_channels=size, activation_fn=tf.nn.relu, weights_initializer=Initializer(tf.truncated_normal_initializer, stddev=weight_stddev), biases_initializer=Initializer(tf.constant_initializer, value=bias_const)) weights_initializer=TFWrapper(tf.truncated_normal_initializer, stddev=weight_stddev), biases_initializer=TFWrapper(tf.constant_initializer, value=bias_const)) if dropout > 0.0: layer = Dropout(dropout, in_layers=[layer]) prev_layer = layer Loading
deepchem/models/tensorgraph/__init__.py +1 −1 Original line number Diff line number Diff line from deepchem.models.tensorgraph.tensor_graph import TensorGraphfrom deepchem.models.tensorgraph import models No newline at end of file from deepchem.models.tensorgraph.tensor_graph import TensorGraph, TFWrapperfrom deepchem.models.tensorgraph import models No newline at end of file
deepchem/models/tensorgraph/layers.py +0 −20 Original line number Diff line number Diff line Loading @@ -98,26 +98,6 @@ def convert_to_layers(in_layers): return layers class Initializer(object): """This class exists as a workaround for Tensorflow initializers not being picklable.""" def __init__(self, initializer_class, **kwargs): """Create an Initializer for constructing Tensorflow initializers. Parameters ---------- initializer_class: class the type of initializer to create kwargs: any other arguments will be passed on to the Tensorflow initializer's constructor """ self.initializer_class = initializer_class self.kwargs = kwargs def __call__(self): return self.initializer_class(**self.kwargs) class Conv1D(Layer): def __init__(self, width, out_channels, **kwargs): Loading
deepchem/models/tensorgraph/tensor_graph.py +36 −7 Original line number Diff line number Diff line Loading @@ -22,7 +22,6 @@ class TensorGraph(Model): def __init__(self, tensorboard=False, tensorboard_log_frequency=100, learning_rate=0.001, batch_size=100, random_seed=None, use_queue=True, Loading @@ -36,8 +35,6 @@ class TensorGraph(Model): Should we log to model_dir data for tensorboard? tensorboard_log_frequency: int How many training batches before logging tensorboard? learning_rate: float learning rate for optimizer batch_size: int default batch size for training and evaluating use_queue: boolean Loading @@ -63,6 +60,7 @@ class TensorGraph(Model): self.loss = None self.built = False self.queue_installed = False self.optimizer = TFWrapper(tf.train.AdamOptimizer, learning_rate=0.001, beta1=0.9, beta2=0.999) # Singular place to hold Tensor objects which don't serialize # These have to be reconstructed on restoring from pickle Loading @@ -81,7 +79,6 @@ class TensorGraph(Model): self.last_checkpoint = None self.use_queue = use_queue self.learning_rate = learning_rate self.batch_size = batch_size self.random_seed = random_seed super(TensorGraph, self).__init__(**kwargs) Loading Loading @@ -368,6 +365,14 @@ class TensorGraph(Model): self._add_layer(layer) self.outputs.append(layer) def set_optimizer(self, optimizer): """Set the optimizer to use for fitting. The argument should be a callable object (most often a TFWrapper) that constructs a Tensorflow optimizer when called. """ self.optimizer = optimizer def save(self): # Remove out_tensor from the object to be pickled must_restore = False Loading Loading @@ -443,9 +448,7 @@ class TensorGraph(Model): elif obj == "FileWriter": self.tensor_objects['FileWriter'] = tf.summary.FileWriter(self.model_dir) elif obj == 'train_op': self.tensor_objects['train_op'] = tf.train.AdamOptimizer( self.learning_rate, beta1=.9, beta2=.999).minimize(self.loss.out_tensor) self.tensor_objects['train_op'] = self.optimizer().minimize(self.loss.out_tensor) elif obj == 'summary_op': self.tensor_objects['summary_op'] = tf.summary.merge_all( key=tf.GraphKeys.SUMMARIES) Loading Loading @@ -518,3 +521,29 @@ def _enqueue_batch(tg, generator, graph, sess, coord): sess.run(tg.input_queue.close_op) coord.num_samples = num_samples coord.request_stop() class TFWrapper(object): """This class exists as a workaround for Tensorflow objects not being picklable. The job of a TFWrapper is to create Tensorflow objects by passing defined arguments to a constructor. There are cases where we really want to store Tensorflow objects of various sorts (optimizers, initializers, etc.), but we can't because they cannot be pickled. So instead we store a TFWrapper that creates the object when needed. """ def __init__(self, tf_class, **kwargs): """Create a TFWrapper for constructing a Tensorflow object. Parameters ---------- tf_class: class the type of object to create kwargs: any other arguments will be passed on to the object's constructor """ self.tf_class = tf_class self.kwargs = kwargs def __call__(self): return self.tf_class(**self.kwargs)
