Switched back to an older version of yapf (73015898) · Commits · 钟慕尧 / deepchem

deepchem/models/tensorgraph/graph_layers.py

+10 −9

Original line number	Diff line number	Diff line
		@@ -660,7 +660,8 @@ class DAGLayer(Layer):
		batch_atom_features = tf.gather(atom_features, current_round)

		# generating index for graph features used in the inputs
		index = tf.stack([
		index = tf.stack(
		[
		tf.reshape(
		tf.stack(
		[tf.boolean_mask(tf.range(n_atoms), mask)] *

deepchem/models/tensorgraph/layers.py

+16 −15

Original line number	Diff line number	Diff line
		@@ -2648,8 +2648,8 @@ class GraphConv(Layer):
		def _create_variables(self, in_channels):
		# Generate the nb_affine weights and biases
		W_list = [
		initializations.glorot_uniform([in_channels, self.out_channel],
		name='kernel')
		initializations.glorot_uniform(
		[in_channels, self.out_channel], name='kernel')
		for k in range(self.num_deg)
		]
		b_list = [
		@@ -4278,7 +4278,8 @@ class ANIFeat(Layer):
		angular_sym = self.angular_symmetry(d_angular_cutoff, d, atom_numbers,
		coordinates)

		out_tensor = tf.concat([
		out_tensor = tf.concat(
		[
		tf.cast(tf.expand_dims(atom_numbers, 2), tf.float32), radial_sym,
		angular_sym
		],
		@@ -4509,8 +4510,8 @@ class GraphEmbedPoolLayer(Layer):

		def _create_variables(self, no_features, no_filters, name):
		W = tf.Variable(
		tf.truncated_normal([no_features, no_filters],
		stddev=1.0 / math.sqrt(no_features)),
		tf.truncated_normal(
		[no_features, no_filters], stddev=1.0 / math.sqrt(no_features)),
		name='%s_weights' % name,
		dtype=tf.float32)
		b = tf.Variable(
		@@ -4599,15 +4600,15 @@ class GraphCNN(Layer):

		def _create_variables(self, no_features, no_A):
		W = tf.Variable(
		tf.truncated_normal([no_features * no_A, self.num_filters],
		stddev=math.sqrt(
		1.0 / (no_features * (no_A + 1) * 1.0))),
		tf.truncated_normal(
		[no_features * no_A, self.num_filters],
		stddev=math.sqrt(1.0 / (no_features * (no_A + 1) * 1.0))),
		name='%s_weights' % self.name,
		dtype=tf.float32)
		W_I = tf.Variable(
		tf.truncated_normal([no_features, self.num_filters],
		stddev=math.sqrt(
		1.0 / (no_features * (no_A + 1) * 1.0))),
		tf.truncated_normal(
		[no_features, self.num_filters],
		stddev=math.sqrt(1.0 / (no_features * (no_A + 1) * 1.0))),
		name='%s_weights_I' % self.name,
		dtype=tf.float32)
		b = tf.Variable(

deepchem/models/tensorgraph/models/seqtoseq.py

+10 −12

Original line number	Diff line number	Diff line
		@@ -240,10 +240,9 @@ class SeqToSeq(TensorGraph):
		for batch in self._batch_elements(sequences):
		feed_dict = {}
		feed_dict[self._features] = self._create_input_array(batch)
		feed_dict[self._gather_indices] = [
		(i, len(batch[i]) if i < len(batch) else 0)
		for i in range(self.batch_size)
		]
		feed_dict[self._gather_indices] = [(i, len(batch[i])
		if i < len(batch) else 0)
		for i in range(self.batch_size)]
		feed_dict[self._training_placeholder] = 0.0
		for initial, zero in zip(self.rnn_initial_states, self.rnn_zero_states):
		feed_dict[initial] = zero
		@@ -267,8 +266,8 @@ class SeqToSeq(TensorGraph):
		result = []
		with self._get_tf("Graph").as_default():
		for batch in self._batch_elements(embeddings):
		embedding_array = np.zeros((self.batch_size, self._embedding_dimension),
		dtype=np.float32)
		embedding_array = np.zeros(
		(self.batch_size, self._embedding_dimension), dtype=np.float32)
		for i, e in enumerate(batch):
		embedding_array[i] = e
		feed_dict = {}
		@@ -294,10 +293,9 @@ class SeqToSeq(TensorGraph):
		for batch in self._batch_elements(sequences):
		feed_dict = {}
		feed_dict[self._features] = self._create_input_array(batch)
		feed_dict[self._gather_indices] = [
		(i, len(batch[i]) if i < len(batch) else 0)
		for i in range(self.batch_size)
		]
		feed_dict[self._gather_indices] = [(i, len(batch[i])
		if i < len(batch) else 0)
		for i in range(self.batch_size)]
		feed_dict[self._training_placeholder] = 0.0
		for initial, zero in zip(self.rnn_initial_states, self.rnn_zero_states):
		feed_dict[initial] = zero
		@@ -357,8 +355,8 @@ class SeqToSeq(TensorGraph):
		for i, sequence in enumerate(sequences):
		for j, token in enumerate(sequence):
		features[i, j, self._input_dict[token]] = 1
		features[np.arange(len(sequences)), lengths, self.
		_input_dict[SeqToSeq.sequence_end]] = 1
		features[np.arange(len(sequences)), lengths, self._input_dict[
		SeqToSeq.sequence_end]] = 1
		return features

		def _create_output_array(self, sequences):

deepchem/models/tensorgraph/symmetry_functions.py

+2 −2

Original line number	Diff line number	Diff line
		@@ -390,8 +390,8 @@ class BPFeatureMerge(Layer):
		angular_symmetry = in_layers[2].out_tensor
		atom_flags = in_layers[3].out_tensor

		out_tensor = tf.concat([atom_embedding, radial_symmetry, angular_symmetry],
		axis=2)
		out_tensor = tf.concat(
		[atom_embedding, radial_symmetry, angular_symmetry], axis=2)
		out_tensor = out_tensor * atom_flags[:, :, 0:1]

		if set_tensors:

deepchem/models/tensorgraph/tensor_graph.py

+3 −2

Original line number	Diff line number	Diff line
		@@ -624,8 +624,9 @@ class TensorGraph(Model):
		n = y.shape[0]
		loop_vars = [tf.constant(0, tf.int32), tf.TensorArray(tf.float32, size=n)]
		_, jacobian = tf.while_loop(
		lambda j, _: j < n, lambda j, result: (
		j + 1, result.write(j, tf.gradients(y[j], x))), loop_vars)
		lambda j, _: j < n,
		lambda j, result: (j + 1, result.write(j, tf.gradients(y[j], x))),
		loop_vars)
		return jacobian.stack()

		if not self.built:

Admin message