Formatting (1d8a4d54) · Commits · 钟慕尧 / deepchem

deepchem/models/tensorgraph/layers.py

+67 −32

Original line number	Diff line number	Diff line
		@@ -5,7 +5,9 @@ import tensorflow as tf

		from deepchem.nn import model_ops, initializations


		class Layer(object):

		def __init__(self, **kwargs):
		if "name" not in kwargs:
		self.name = "%s%s" % (self.__class__.__name__, self._random_name())
		@@ -30,6 +32,7 @@ class Layer(object):


		class Conv1DLayer(Layer):

		def __init__(self, width, out_channels, **kwargs):
		self.width = width
		self.out_channels = out_channels
		@@ -54,6 +57,7 @@ class Conv1DLayer(Layer):


		class Dense(Layer):

		def __init__(self, out_channels, activation_fn=None, **kwargs):
		self.out_channels = out_channels
		self.out_tensor = None
		@@ -76,6 +80,7 @@ class Dense(Layer):


		class Flatten(Layer):

		def __init__(self, **kwargs):
		super().__init__(**kwargs)

		@@ -93,6 +98,7 @@ class Flatten(Layer):


		class Reshape(Layer):

		def __init__(self, shape, **kwargs):
		self.shape = shape
		super().__init__(**kwargs)
		@@ -103,6 +109,7 @@ class Reshape(Layer):


		class CombineMeanStd(Layer):

		def __init__(self, **kwargs):
		super().__init__(**kwargs)

		@@ -117,6 +124,7 @@ class CombineMeanStd(Layer):


		class Repeat(Layer):

		def __init__(self, n_times, **kwargs):
		self.n_times = n_times
		super().__init__(**kwargs)
		@@ -131,6 +139,7 @@ class Repeat(Layer):


		class GRU(Layer):

		def __init__(self, n_hidden, out_channels, batch_size, **kwargs):
		self.n_hidden = n_hidden
		self.out_channels = out_channels
		@@ -153,6 +162,7 @@ class GRU(Layer):


		class TimeSeriesDense(Layer):

		def __init__(self, out_channels, **kwargs):
		self.out_channels = out_channels
		super().__init__(**kwargs)
		@@ -167,6 +177,7 @@ class TimeSeriesDense(Layer):


		class Input(Layer):

		def __init__(self, shape, dtype=tf.float32, pre_queue=False, **kwargs):
		self.shape = shape
		self.pre_queue = pre_queue
		@@ -186,7 +197,11 @@ class Input(Layer):
		if self.pre_queue:
		raise ValueError("Input is already pre_q")
		q_shape = (batch_size,) + self.shape[1:]
		return Input(shape=q_shape, name="%s_pre_q" % self.name, dtype=self.dtype, pre_queue=True)
		return Input(
		shape=q_shape,
		name="%s_pre_q" % self.name,
		dtype=self.dtype,
		pre_queue=True)

		def get_pre_q_name(self):
		if self.pre_queue:
		@@ -195,6 +210,7 @@ class Input(Layer):


		class LossLayer(Layer):

		def __init__(self, **kwargs):
		super().__init__(**kwargs)

		@@ -206,6 +222,7 @@ class LossLayer(Layer):


		class SoftMax(Layer):

		def __init__(self, **kwargs):
		super().__init__(**kwargs)

		@@ -218,6 +235,7 @@ class SoftMax(Layer):


		class Concat(Layer):

		def __init__(self, **kwargs):
		super().__init__(**kwargs)

		@@ -232,6 +250,7 @@ class Concat(Layer):


		class SoftMaxCrossEntropy(Layer):

		def __init__(self, **kwargs):
		super().__init__(**kwargs)

		@@ -246,6 +265,7 @@ class SoftMaxCrossEntropy(Layer):


		class ReduceMean(Layer):

		def __call__(self, *parents):
		parent_tensor = parents[0].out_tensor
		self.out_tensor = tf.reduce_mean(parent_tensor)
		@@ -253,6 +273,7 @@ class ReduceMean(Layer):


		class Conv2d(Layer):

		def __init__(self, num_outputs, kernel_size=5, **kwargs):
		self.num_outputs = num_outputs
		self.kernel_size = kernel_size
		@@ -271,6 +292,7 @@ class Conv2d(Layer):


		class MaxPool(Layer):

		def __init__(self,
		ksize=[1, 2, 2, 1],
		strides=[1, 2, 2, 1],
		@@ -323,7 +345,13 @@ class InputFifoQueue(Layer):


		class GraphConvLayer(Layer):
		def __init__(self, out_channel, min_deg=0, max_deg=10, activation_fn=None, **kwargs):

		def __init__(self,
		out_channel,
		min_deg=0,
		max_deg=10,
		activation_fn=None,
		**kwargs):
		self.out_channel = out_channel
		self.min_degree = min_deg
		self.max_degree = max_deg
		@@ -428,6 +456,7 @@ class GraphConvLayer(Layer):


		class GraphPoolLayer(Layer):

		def __init__(self, min_degree=0, max_degree=10, **kwargs):
		self.min_degree = min_degree
		self.max_degree = max_degree
		@@ -473,6 +502,7 @@ class GraphPoolLayer(Layer):


		class GraphGather(Layer):

		def __init__(self, batch_size, activation_fn=None, **kwargs):
		self.batch_size = batch_size
		self.activation_fn = activation_fn
		@@ -490,14 +520,17 @@ class GraphGather(Layer):
		assert (self.batch_size > 1, "graph_gather requires batches larger than 1")

		# Obtain the partitions for each of the molecules
		activated_par = tf.dynamic_partition(atom_features, membership, self.batch_size)
		activated_par = tf.dynamic_partition(atom_features, membership,
		self.batch_size)

		# Sum over atoms for each molecule
		sparse_reps = [
		tf.reduce_sum(activated, 0, keep_dims=True) for activated in activated_par
		tf.reduce_sum(activated, 0, keep_dims=True)
		for activated in activated_par
		]
		max_reps = [
		tf.reduce_max(activated, 0, keep_dims=True) for activated in activated_par
		tf.reduce_max(activated, 0, keep_dims=True)
		for activated in activated_par
		]

		# Get the final sparse representations
		@@ -512,6 +545,7 @@ class GraphGather(Layer):


		class BatchNormLayer(Layer):

		def __call__(self, *parents):
		parent_tensor = parents[0].out_tensor
		self.out_tensor = tf.layers.batch_normalization(parent_tensor)
		@@ -519,6 +553,7 @@ class BatchNormLayer(Layer):


		class WeightedError(Layer):

		def __call__(self, *parents):
		entropy, weights = parents[0], parents[1]
		self.out_tensor = tf.reduce_sum(entropy.out_tensor * weights.out_tensor)

deepchem/models/tensorgraph/models/graph_conv.py

+18 −12

Original line number	Diff line number	Diff line
		@@ -7,10 +7,10 @@ from deepchem.feat.mol_graphs import ConvMol
		import time



		class GraphConvTensorGraph(TensorGraph):
		"""
		"""

		def __init__(self, **kwargs):
		super().__init__(**kwargs)
		self.min_degree = 0
		@@ -29,7 +29,8 @@ class GraphConvTensorGraph(TensorGraph):
		feed_dict[self.features[1].out_tensor] = multiConvMol.deg_slice
		feed_dict[self.features[2].out_tensor] = multiConvMol.membership
		for i in range(self.max_degree):
		feed_dict[self.features[i + 3].out_tensor] = multiConvMol.get_deg_adjacency_lists()[i + 1]
		feed_dict[self.features[i + 3]
		.out_tensor] = multiConvMol.get_deg_adjacency_lists()[i + 1]
		return feed_dict

		def fit(self,
		@@ -63,7 +64,8 @@ class GraphConvTensorGraph(TensorGraph):
		avg_loss, n_batches = 0.0, 0.0
		for epoch in range(nb_epoch):
		for ind, (X_b, y_b, w_b, ids_b) in enumerate(
		dataset.iterbatches(self.batch_size, deterministic=True, pad_batches=True)):
		dataset.iterbatches(
		self.batch_size, deterministic=True, pad_batches=True)):
		feed_dict = self._construct_feed_dict(X_b, y_b, w_b, ids_b)
		output_tensors = [x.out_tensor for x in self.outputs]
		fetches = output_tensors + [train_op, self.loss.out_tensor]
		@@ -85,8 +87,7 @@ class GraphConvTensorGraph(TensorGraph):


		def graph_conv_model(batch_size, num_tasks):
		model = GraphConvTensorGraph(batch_size=batch_size,
		use_queue=False)
		model = GraphConvTensorGraph(batch_size=batch_size, use_queue=False)
		atom_features = Input(shape=(None, 75))
		model.add_feature(atom_features)

		@@ -103,13 +104,15 @@ def graph_conv_model(batch_size, num_tasks):
		deg_adjs.append(deg_adj)

		gc1 = GraphConvLayer(64, activation_fn=tf.nn.relu)
		model.add_layer(gc1, parents=[atom_features, degree_slice, membership] + deg_adjs)
		model.add_layer(
		gc1, parents=[atom_features, degree_slice, membership] + deg_adjs)

		batch_norm1 = BatchNormLayer()
		model.add_layer(batch_norm1, parents=[gc1])

		gp1 = GraphPoolLayer()
		model.add_layer(gp1, parents=[batch_norm1, degree_slice, membership] + deg_adjs)
		model.add_layer(
		gp1, parents=[batch_norm1, degree_slice, membership] + deg_adjs)

		gc2 = GraphConvLayer(64, activation_fn=tf.nn.relu)
		model.add_layer(gc2, parents=[gp1, degree_slice, membership] + deg_adjs)
		@@ -118,7 +121,8 @@ def graph_conv_model(batch_size, num_tasks):
		model.add_layer(batch_norm2, parents=[gc2])

		gp2 = GraphPoolLayer()
		model.add_layer(gp2, parents=[batch_norm2, degree_slice, membership] + deg_adjs)
		model.add_layer(
		gp2, parents=[batch_norm2, degree_slice, membership] + deg_adjs)

		dense = Dense(out_channels=128, activation_fn=None)
		model.add_layer(dense, parents=[gp2])
		@@ -127,11 +131,13 @@ def graph_conv_model(batch_size, num_tasks):
		model.add_layer(batch_norm3, parents=[dense])

		gg1 = GraphGather(batch_size=batch_size, activation_fn=tf.nn.tanh)
		model.add_layer(gg1, parents=[batch_norm3, degree_slice, membership] + deg_adjs)
		model.add_layer(
		gg1, parents=[batch_norm3, degree_slice, membership] + deg_adjs)

		costs = []
		for task in range(num_tasks):
		classification = Dense(out_channels=2, name="GUESS%s" % task, activation_fn=None)
		classification = Dense(
		out_channels=2, name="GUESS%s" % task, activation_fn=None)
		model.add_layer(classification, parents=[gg1])

		softmax = SoftMax(name="SOFTMAX%s" % task)

deepchem/models/tensorgraph/tensor_graph.py

+3 −1

Original line number	Diff line number	Diff line
		@@ -176,6 +176,7 @@ class TensorGraph(Model):
		return retval

		def predict_proba_on_batch(self, X, sess=None):

		def predict():
		out_tensors = [x.out_tensor for x in self.outputs]
		fetches = out_tensors
		@@ -376,7 +377,8 @@ class TensorGraph(Model):
		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.learning_rate, beta1=.9,
		beta2=.999).minimize(self.loss.out_tensor)
		elif obj == 'summary_op':
		self.tensor_objects['summary_op'] = tf.summary.merge_all(
		key=tf.GraphKeys.SUMMARIES)

examples/benchmark.py

+2 −1

Original line number	Diff line number	Diff line
		@@ -495,7 +495,8 @@ def benchmark_classification(train_dataset,
		nb_epoch = hyper_parameters['nb_epoch']
		learning_rate = hyper_parameters['learning_rate']

		model_graphconv = dc.models.tensorgraph.models.graph_conv_model(batch_size, len(tasks))
		model_graphconv = dc.models.tensorgraph.models.graph_conv_model(batch_size,
		len(tasks))
		model_graphconv.fit(train_dataset, nb_epoch=nb_epoch)
		# Evaluating graph convolution model
		train_scores['graphconv'] = model_graphconv.evaluate(

examples/tox21/tox21_graph_conv.py

+4 −3

Original line number	Diff line number	Diff line
		@@ -24,7 +24,8 @@ metric = dc.metrics.Metric(
		# Batch size of models
		batch_size = 50

		model = dc.models.tensorgraph.models.graph_conv_model(batch_size, len(tox21_tasks))
		model = dc.models.tensorgraph.models.graph_conv_model(batch_size,
		len(tox21_tasks))

		model.fit(train_dataset, nb_epoch=10, checkpoint_interval=10)

Admin message