Loading deepchem/nn/weave_layers.py +53 −18 Original line number Diff line number Diff line Loading @@ -19,10 +19,9 @@ from deepchem.nn.copy import Layer class WeaveLayer(Layer): """" Main layer of DAG model For a molecule with n atoms, n different graphs are generated and run through The final outputs of each graph become the graph features of corresponding atom, which will be summed and put into another network in DAGGather Layer """" Main layer of Weave model For each molecule, atom features and pair features are recombined to generate new atom(pair) features """ def __init__(self, Loading Loading @@ -160,21 +159,55 @@ class WeaveLayer(Layer): P = tf.multiply(P, tf.expand_dims(pair_mask, axis=3)) return A, P class WeaveConcat(Layer): """" Concat a batch of molecules into a batch of atoms """ def __init__(self, **kwargs): super(WeaveConcat, self).__init__(**kwargs) def build(self, shape): pass def call(self, x, mask=None): """Execute this layer on input tensors. Parameters ---------- x: list Tensors: atom_features, atom_masks mask: bool, optional Ignored. Present only to shadow superclass call() method. Returns ------- outputs: Tensor Tensor of concatenated atom features """ self.build() atom_features = x[0] atom_masks = x[1] A = tf.unstack(atom_features, axis=0) A_mask = tf.unstack(tf.cast(atom_masks, dtype=tf.bool), axis=0) outputs = tf.concat([tf.boolean_mask(A[i], A_mask[i]) for i in range(len(A))], axis=0) return outputs class WeaveGather(Layer): """" Main layer of DAG model For a molecule with n atoms, n different graphs are generated and run through The final outputs of each graph become the graph features of corresponding atom, which will be summed and put into another network in DAGGather Layer """" Gather layer of Weave model a batch of normalized atom features go through a hidden layer, then summed to form molecular features """ def __init__(self, n_atom_input_feat=50, n_hidden=128, init='glorot_uniform', activation='relu', activation='tanh', gaussian_expand=True, dropout=None, epsilon=1e-3, momentum=0.99, **kwargs): """ Parameters Loading Loading @@ -210,15 +243,15 @@ class WeaveGather(Layer): self.n_outputs = self.n_hidden * 11 else: self.n_outputs = self.n_hidden self.epsilon = epsilon self.momentum = momentum def build(self): def build(self, shape): """"Construct internal trainable weights. """ self.W = self.init([self.n_atom_input_feat, self.n_hidden]) self.b = model_ops.zeros(shape=[ self.n_hidden, ]) self.b = model_ops.zeros(shape=[self.n_hidden,]) self.trainable_weights = self.W + self.b Loading @@ -227,8 +260,8 @@ class WeaveGather(Layer): Parameters ---------- x: Tensor Tensors of atom features x: list Tensors: atom_features, atom_masks mask: bool, optional Ignored. Present only to shadow superclass call() method. Loading @@ -239,10 +272,12 @@ class WeaveGather(Layer): """ # Add trainable weights self.build() outputs = tf.tensordot(x, self.W, [[2], [0]]) + self.b atom_features = x[0] atom_masks = x[1] outputs = tf.matmul(atom_features, self.W) + self.b if self.gaussian_expand: outputs = self.gaussian_histogram(outputs) outputs = tf.reduce_sum(outputs, axis=1) return outputs Loading Loading
deepchem/nn/weave_layers.py +53 −18 Original line number Diff line number Diff line Loading @@ -19,10 +19,9 @@ from deepchem.nn.copy import Layer class WeaveLayer(Layer): """" Main layer of DAG model For a molecule with n atoms, n different graphs are generated and run through The final outputs of each graph become the graph features of corresponding atom, which will be summed and put into another network in DAGGather Layer """" Main layer of Weave model For each molecule, atom features and pair features are recombined to generate new atom(pair) features """ def __init__(self, Loading Loading @@ -160,21 +159,55 @@ class WeaveLayer(Layer): P = tf.multiply(P, tf.expand_dims(pair_mask, axis=3)) return A, P class WeaveConcat(Layer): """" Concat a batch of molecules into a batch of atoms """ def __init__(self, **kwargs): super(WeaveConcat, self).__init__(**kwargs) def build(self, shape): pass def call(self, x, mask=None): """Execute this layer on input tensors. Parameters ---------- x: list Tensors: atom_features, atom_masks mask: bool, optional Ignored. Present only to shadow superclass call() method. Returns ------- outputs: Tensor Tensor of concatenated atom features """ self.build() atom_features = x[0] atom_masks = x[1] A = tf.unstack(atom_features, axis=0) A_mask = tf.unstack(tf.cast(atom_masks, dtype=tf.bool), axis=0) outputs = tf.concat([tf.boolean_mask(A[i], A_mask[i]) for i in range(len(A))], axis=0) return outputs class WeaveGather(Layer): """" Main layer of DAG model For a molecule with n atoms, n different graphs are generated and run through The final outputs of each graph become the graph features of corresponding atom, which will be summed and put into another network in DAGGather Layer """" Gather layer of Weave model a batch of normalized atom features go through a hidden layer, then summed to form molecular features """ def __init__(self, n_atom_input_feat=50, n_hidden=128, init='glorot_uniform', activation='relu', activation='tanh', gaussian_expand=True, dropout=None, epsilon=1e-3, momentum=0.99, **kwargs): """ Parameters Loading Loading @@ -210,15 +243,15 @@ class WeaveGather(Layer): self.n_outputs = self.n_hidden * 11 else: self.n_outputs = self.n_hidden self.epsilon = epsilon self.momentum = momentum def build(self): def build(self, shape): """"Construct internal trainable weights. """ self.W = self.init([self.n_atom_input_feat, self.n_hidden]) self.b = model_ops.zeros(shape=[ self.n_hidden, ]) self.b = model_ops.zeros(shape=[self.n_hidden,]) self.trainable_weights = self.W + self.b Loading @@ -227,8 +260,8 @@ class WeaveGather(Layer): Parameters ---------- x: Tensor Tensors of atom features x: list Tensors: atom_features, atom_masks mask: bool, optional Ignored. Present only to shadow superclass call() method. Loading @@ -239,10 +272,12 @@ class WeaveGather(Layer): """ # Add trainable weights self.build() outputs = tf.tensordot(x, self.W, [[2], [0]]) + self.b atom_features = x[0] atom_masks = x[1] outputs = tf.matmul(atom_features, self.W) + self.b if self.gaussian_expand: outputs = self.gaussian_histogram(outputs) outputs = tf.reduce_sum(outputs, axis=1) return outputs Loading
