Loading deepchem/models/tf_new_models/graph_models.py +8 −5 Original line number Diff line number Diff line Loading @@ -166,15 +166,14 @@ class SequentialWeaveGraph(SequentialGraph): """SequentialGraph for Weave models """ def __init__(self, batch_size, n_atom_feat=75, n_pair_feat=14, max_atoms=100): def __init__(self, max_atoms=50, n_atom_feat=75, n_pair_feat=14): self.graph = tf.Graph() self.batch_size = batch_size self.max_atoms = max_atoms self.n_atom_feat = n_atom_feat self.n_pair_feat = n_pair_feat with self.graph.as_default(): self.graph_topology = WeaveGraphTopology(self.batch_size, self.n_atom_feat, self.n_pair_feat, self.max_atoms) self.graph_topology = WeaveGraphTopology(self.max_atoms, self.n_atom_feat, self.n_pair_feat) self.output = self.graph_topology.get_atom_features_placeholder() self.output_P = self.graph_topology.get_pair_features_placeholder() self.layers = [] Loading @@ -186,14 +185,18 @@ class SequentialWeaveGraph(SequentialGraph): self.output, self.output_P = layer([ self.output, self.output_P ] + self.graph_topology.get_topology_placeholders()) elif type(layer).__name__ in ['WeaveConcat']: self.output = layer( [self.output, self.graph_topology.atom_mask_placeholder]) elif type(layer).__name__ in ['WeaveGather']: self.output = layer( [self.output, self.graph_topology.atom_split_placeholder]) [self.output, self.graph_topology.membership_placeholder]) else: self.output = layer(self.output) self.layers.append(layer) class SequentialSupportGraph(object): """An analog of Keras Sequential model for test/support models.""" Loading deepchem/models/tf_new_models/graph_topology.py +39 −53 Original line number Diff line number Diff line Loading @@ -397,8 +397,8 @@ class DAGGraphTopology(GraphTopology): class WeaveGraphTopology(GraphTopology): """Manages placeholders associated with batch of graphs and their topology""" def __init__(self, batch_size, n_atom_feat, n_pair_feat, max_atoms=100, name='Weave_topology'): def __init__(self, max_atoms, n_atom_feat, n_pair_feat, name='Weave_topology'): """ Parameters ---------- Loading @@ -412,36 +412,30 @@ class WeaveGraphTopology(GraphTopology): #self.n_atoms = n_atoms self.name = name self.batch_size = batch_size self.max_atoms = max_atoms self.n_atom_feat = n_atom_feat self.n_pair_feat = n_pair_feat self.max_atoms = max_atoms * batch_size self.atom_features_placeholder = tf.placeholder( dtype='float32', shape=(None, self.n_atom_feat), shape=(None, self.max_atoms, self.n_atom_feat), name=self.name + '_atom_features') self.atom_mask_placeholder = tf.placeholder( dtype='float32', shape=(None, self.max_atoms), name=self.name + '_atom_mask') self.pair_features_placeholder = tf.placeholder( dtype='float32', shape=(None, self.n_pair_feat), shape=(None, self.max_atoms, self.max_atoms, self.n_pair_feat), name=self.name + '_pair_features') self.pair_split_placeholder = tf.placeholder( dtype='int32', shape=(self.max_atoms,), name=self.name + '_pair_split') self.pair_membership_placeholder = tf.placeholder( dtype='bool', shape=(self.max_atoms,), name=self.name + '_pair_membership') self.atom_split_placeholder = tf.placeholder( dtype='int32', shape=(self.batch_size,), name=self.name + '_atom_split') self.atom_to_pair_placeholder = tf.placeholder( dtype='int32', shape=(None,2), name=self.name + '_atom_to_pair') self.pair_mask_placeholder = tf.placeholder( dtype='float32', shape=(None, self.max_atoms, self.max_atoms), name=self.name + '_pair_mask') self.membership_placeholder = tf.placeholder( dtype='int32', shape=(None,), name=self.name + '_membership') # Define the list of tensors to be used as topology self.topology = [self.pair_split_placeholder, self.pair_membership_placeholder, self.atom_split_placeholder, self.atom_to_pair_placeholder] self.topology = [self.atom_mask_placeholder, self.pair_mask_placeholder] self.inputs = [self.atom_features_placeholder] self.inputs += self.topology Loading @@ -467,42 +461,34 @@ class WeaveGraphTopology(GraphTopology): # Extract atom numbers atom_feat = [] pair_feat = [] atom_split = [] atom_to_pair = [] pair_split = [] atom_mask = [] pair_mask = [] membership = [] max_atoms = self.max_atoms start = 0 for im, mol in enumerate(batch): n_atoms = mol.get_num_atoms() # number of atoms in each molecule atom_split.append(n_atoms) # index of pair features C0, C1 = np.meshgrid(np.arange(n_atoms), np.arange(n_atoms)) atom_to_pair.append(np.transpose(np.array([C1.flatten()+start, C0.flatten()+start]))) start = start + n_atoms # number of pairs for each atom pair_split.extend([n_atoms]*n_atoms) # atom features atom_feat.append(mol.get_atom_features()) # pair features pair_feat.append(np.reshape(mol.get_pair_features(), (n_atoms*n_atoms, self.n_pair_feat))) atom_feat = np.concatenate(atom_feat, axis=0) pair_feat = np.concatenate(pair_feat, axis=0) atom_to_pair = np.concatenate(atom_to_pair, axis=0) atom_split = np.array(atom_split) n_pair = len(pair_split) pair_split = np.pad(pair_split, ((0, max_atoms-n_pair)), 'constant') pair_membership = np.array([True]*n_pair + [False]*(max_atoms-n_pair)) atom_feat.append( np.pad(mol.get_atom_features(), ((0, max_atoms - n_atoms), (0, 0)), 'constant')) atom_mask.append( np.array([1] * n_atoms + [0] * (max_atoms - n_atoms), dtype=float)) pair_feat.append( np.pad(mol.get_pair_features(), ((0, max_atoms - n_atoms), ( 0, max_atoms - n_atoms), (0, 0)), 'constant')) pair_mask.append(np.array([[1]*n_atoms + [0]*(max_atoms-n_atoms)]*n_atoms + \ [[0]*max_atoms]*(max_atoms-n_atoms), dtype=float)) membership.extend([im] * n_atoms) atom_feat = np.stack(atom_feat) pair_feat = np.stack(pair_feat) atom_mask = np.stack(atom_mask) pair_mask = np.stack(pair_mask) membership = np.array(membership) # Generate dicts dict_DTNN = { self.atom_features_placeholder: atom_feat, self.pair_features_placeholder: pair_feat, self.pair_split_placeholder: pair_split, self.pair_membership_placeholder: pair_membership, self.atom_split_placeholder: atom_split, self.atom_to_pair_placeholder: atom_to_pair self.atom_mask_placeholder: atom_mask, self.pair_mask_placeholder: pair_mask, self.membership_placeholder: membership } return dict_DTNN deepchem/molnet/run_benchmark_models.py +20 −10 Original line number Diff line number Diff line Loading @@ -266,11 +266,16 @@ def benchmark_classification(train_dataset, n_graph_feat = hyper_parameters['n_graph_feat'] n_pair_feat = hyper_parameters['n_pair_feat'] graph_model = deepchem.nn.SequentialWeaveGraph( batch_size, n_atom_feat=n_features, n_pair_feat=n_pair_feat, max_atoms=120) graph_model.add(deepchem.nn.WeaveLayer(75, 14)) graph_model.add(deepchem.nn.WeaveLayer(50, 50)) graph_model.add(deepchem.nn.Dense(n_graph_feat, 50, activation='tanh')) max_atoms_train = max([mol.get_num_atoms() for mol in train_dataset.X]) max_atoms_valid = max([mol.get_num_atoms() for mol in valid_dataset.X]) max_atoms_test = max([mol.get_num_atoms() for mol in test_dataset.X]) max_atoms = max([max_atoms_train, max_atoms_valid, max_atoms_test]) graph_model = deepchem.nn.SequentialWeaveGraph(max_atoms=max_atoms, n_atom_feat=n_features, n_pair_feat=n_pair_feat) graph_model.add(deepchem.nn.WeaveLayer(max_atoms, 75, 14)) graph_model.add(deepchem.nn.WeaveLayer(max_atoms, 50, 50, update_pair=False)) graph_model.add(deepchem.nn.WeaveConcat(batch_size, n_output=n_graph_feat)) graph_model.add(deepchem.nn.BatchNormalization(epsilon=1e-5, mode=1)) graph_model.add( deepchem.nn.WeaveGather( Loading Loading @@ -585,11 +590,16 @@ def benchmark_regression(train_dataset, n_graph_feat = hyper_parameters['n_graph_feat'] n_pair_feat = hyper_parameters['n_pair_feat'] graph_model = deepchem.nn.SequentialWeaveGraph( batch_size, n_atom_feat=n_features, n_pair_feat=n_pair_feat, max_atoms=80) graph_model.add(deepchem.nn.WeaveLayer(75, 14)) graph_model.add(deepchem.nn.WeaveLayer(50, 50)) graph_model.add(deepchem.nn.Dense(n_graph_feat, 50, activation='tanh')) max_atoms_train = max([mol.get_num_atoms() for mol in train_dataset.X]) max_atoms_valid = max([mol.get_num_atoms() for mol in valid_dataset.X]) max_atoms_test = max([mol.get_num_atoms() for mol in test_dataset.X]) max_atoms = max([max_atoms_train, max_atoms_valid, max_atoms_test]) graph_model = deepchem.nn.SequentialWeaveGraph(max_atoms=max_atoms, n_atom_feat=n_features, n_pair_feat=n_pair_feat) graph_model.add(deepchem.nn.WeaveLayer(max_atoms, 75, 14)) graph_model.add(deepchem.nn.WeaveLayer(max_atoms, 50, 50, update_pair=False)) graph_model.add(deepchem.nn.WeaveConcat(batch_size, n_output=n_graph_feat)) graph_model.add(deepchem.nn.BatchNormalization(epsilon=1e-5, mode=1)) graph_model.add( deepchem.nn.WeaveGather( Loading deepchem/nn/__init__.py +1 −0 Original line number Diff line number Diff line Loading @@ -21,6 +21,7 @@ from deepchem.nn.layers import DAGLayer from deepchem.nn.layers import DAGGather from deepchem.nn.weave_layers import WeaveLayer from deepchem.nn.weave_layers import WeaveConcat from deepchem.nn.weave_layers import WeaveGather from deepchem.nn.model_ops import weight_decay Loading deepchem/nn/weave_layers.py +122 −47 Original line number Diff line number Diff line Loading @@ -29,6 +29,7 @@ class WeaveLayer(Layer): """ def __init__(self, max_atoms, n_atom_input_feat=75, n_pair_input_feat=14, n_atom_output_feat=50, Loading @@ -37,6 +38,7 @@ class WeaveLayer(Layer): n_hidden_PA=50, n_hidden_AP=50, n_hidden_PP=50, update_pair=True, init='glorot_uniform', activation='relu', dropout=None, Loading @@ -63,8 +65,10 @@ class WeaveLayer(Layer): """ super(WeaveLayer, self).__init__(**kwargs) self.max_atoms = max_atoms self.init = initializations.get(init) # Set weight initialization self.activation = activations.get(activation) # Get activations self.update_pair = update_pair # last weave layer does not need to update self.n_hidden_AA = n_hidden_AA self.n_hidden_PA = n_hidden_PA self.n_hidden_AP = n_hidden_AP Loading Loading @@ -96,30 +100,27 @@ class WeaveLayer(Layer): self.n_atom_output_feat, ]) self.trainable_weights = [ self.W_AA, self.b_AA, self.W_PA, self.b_PA, self.W_A, self.b_A ] if self.update_pair: self.W_AP = self.init([self.n_atom_input_feat * 2, self.n_hidden_AP]) self.b_AP = model_ops.zeros(shape=[ self.n_hidden_AP, ]) self.b_AP = model_ops.zeros(shape=[self.n_hidden_AP,]) self.W_PP = self.init([self.n_pair_input_feat, self.n_hidden_PP]) self.b_PP = model_ops.zeros(shape=[ self.n_hidden_PP, ]) self.b_PP = model_ops.zeros(shape=[self.n_hidden_PP,]) self.W_P = self.init([self.n_hidden_P, self.n_pair_output_feat]) self.b_P = model_ops.zeros(shape=[ self.n_pair_output_feat, ]) self.b_P = model_ops.zeros(shape=[self.n_pair_output_feat,]) self.trainable_weights = [ self.W_AA, self.b_AA, self.W_PA, self.b_PA, self.W_A, self.b_A, self.trainable_weights.extend([ self.W_AP, self.b_AP, self.W_PP, self.b_PP, self.W_P, self.b_P ] ]) def call(self, x, mask=None): """Execute this layer on input tensors. x = [atom_features, pair_features, pair_split, pair_membership, atom_split] x = [atom_features, pair_features, atom_mask, pair_mask] Parameters ---------- Loading @@ -141,39 +142,113 @@ class WeaveLayer(Layer): atom_features = x[0] pair_features = x[1] pair_split = x[2] pair_membership = x[3] atom_split = x[4] atom_to_pair = x[5] atom_mask = x[2] pair_mask = x[3] max_atoms = self.max_atoms AA = tf.matmul(atom_features, self.W_AA) + self.b_AA AA = tf.tensordot(atom_features, self.W_AA, [[2], [0]]) + self.b_AA AA = self.activation(AA) PA = tf.matmul(pair_features, self.W_PA) + self.b_PA PA = tf.reduce_sum( tf.tensordot(pair_features, self.W_PA, [[3], [0]]) + self.b_PA, axis=2) PA = self.activation(PA) PAs = tf.split(PA, pair_split, axis=0) PA = [tf.reduce_sum(molecule, 0) for molecule in PAs] PA = tf.boolean_mask(PA, pair_membership) A = tf.matmul(tf.concat([AA, PA], 1), self.W_A) + self.b_A A = tf.tensordot(tf.concat([AA, PA], 2), self.W_A, [[2], [0]]) + self.b_A A = self.activation(A) AP_ij = tf.matmul(tf.reshape(tf.gather(atom_features, atom_to_pair), [-1, 2*self.n_atom_input_feat]), self.W_AP) + self.b_AP AP_ij = self.activation(AP_ij) AP_ji = tf.matmul(tf.reshape(tf.gather(atom_features, tf.reverse(atom_to_pair, [1])), [-1, 2*self.n_atom_input_feat]), self.W_AP) + self.b_AP AP_ji = self.activation(AP_ji) PP = tf.matmul(pair_features, self.W_PP) + self.b_PP A = tf.multiply(A, tf.expand_dims(atom_mask, axis=2)) if self.update_pair: AP_combine = tf.concat([ tf.stack([atom_features] * max_atoms, axis=2), tf.stack([atom_features] * max_atoms, axis=1) ], 3) AP_combine_t = tf.transpose(AP_combine, perm=[0, 2, 1, 3]) AP = tf.tensordot(AP_combine + AP_combine_t, self.W_AP, [[3], [0]]) + self.b_AP AP = self.activation(AP) PP = tf.tensordot(pair_features, self.W_PP, [[3], [0]]) + self.b_PP PP = self.activation(PP) P = tf.matmul(tf.concat([AP_ij + AP_ji, PP], 1), self.W_P) + self.b_P P = tf.tensordot(tf.concat([AP, PP], 3), self.W_P, [[3], [0]]) + self.b_P P = self.activation(P) P = tf.multiply(P, tf.expand_dims(pair_mask, axis=3)) else: P = pair_features return A, P class WeaveConcat(Layer): """" Concat a batch of molecules into a batch of atoms """ def __init__(self, batch_size, n_atom_input_feat=50, n_output=128, init='glorot_uniform', activation='tanh', **kwargs): """ Parameters ---------- batch_size: int number of molecules in a batch n_atom_input_feat: int, optional Number of features for each atom in input. n_output: int, optional Number of output features for each atom(concatenated) init: str, optional Weight initialization for filters. activation: str, optional Activation function applied """ self.batch_size = batch_size self.n_atom_input_feat = n_atom_input_feat self.n_output = n_output self.init = initializations.get(init) # Set weight initialization self.activation = activations.get(activation) # Get activations super(WeaveConcat, self).__init__(**kwargs) def build(self): """"Construct internal trainable weights. """ self.W = self.init([self.n_atom_input_feat, self.n_output]) self.b = model_ops.zeros(shape=[ self.n_output, ]) self.trainable_weights = self.W + self.b def call(self, x, mask=None): """Execute this layer on input tensors. x = [atom_features, atom_mask] Parameters ---------- x: list Tensors as listed above 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.split(atom_features, self.batch_size, axis=0) A_mask = tf.split( tf.cast(atom_masks, dtype=tf.bool), self.batch_size, axis=0) outputs = tf.concat( [tf.boolean_mask(A[i], A_mask[i]) for i in range(len(A))], axis=0) outputs = tf.matmul(outputs, self.W) + self.b outputs = self.activation(outputs) return outputs class WeaveGather(Layer): """" Gather layer of Weave model a batch of normalized atom features go through a hidden layer, Loading Loading @@ -220,7 +295,7 @@ class WeaveGather(Layer): def call(self, x, mask=None): """Execute this layer on input tensors. x = [atom_features, atom_split] x = [atom_features, membership] Parameters ---------- Loading @@ -237,12 +312,12 @@ class WeaveGather(Layer): # Add trainable weights self.build() outputs = x[0] atom_split = x[1] membership = x[1] if self.gaussian_expand: outputs = self.gaussian_histogram(outputs) outputs = tf.split(outputs, atom_split, axis=0) outputs = tf.dynamic_partition(outputs, membership, self.batch_size) output_molecules = [tf.reduce_sum(molecule, 0) for molecule in outputs] Loading Loading
deepchem/models/tf_new_models/graph_models.py +8 −5 Original line number Diff line number Diff line Loading @@ -166,15 +166,14 @@ class SequentialWeaveGraph(SequentialGraph): """SequentialGraph for Weave models """ def __init__(self, batch_size, n_atom_feat=75, n_pair_feat=14, max_atoms=100): def __init__(self, max_atoms=50, n_atom_feat=75, n_pair_feat=14): self.graph = tf.Graph() self.batch_size = batch_size self.max_atoms = max_atoms self.n_atom_feat = n_atom_feat self.n_pair_feat = n_pair_feat with self.graph.as_default(): self.graph_topology = WeaveGraphTopology(self.batch_size, self.n_atom_feat, self.n_pair_feat, self.max_atoms) self.graph_topology = WeaveGraphTopology(self.max_atoms, self.n_atom_feat, self.n_pair_feat) self.output = self.graph_topology.get_atom_features_placeholder() self.output_P = self.graph_topology.get_pair_features_placeholder() self.layers = [] Loading @@ -186,14 +185,18 @@ class SequentialWeaveGraph(SequentialGraph): self.output, self.output_P = layer([ self.output, self.output_P ] + self.graph_topology.get_topology_placeholders()) elif type(layer).__name__ in ['WeaveConcat']: self.output = layer( [self.output, self.graph_topology.atom_mask_placeholder]) elif type(layer).__name__ in ['WeaveGather']: self.output = layer( [self.output, self.graph_topology.atom_split_placeholder]) [self.output, self.graph_topology.membership_placeholder]) else: self.output = layer(self.output) self.layers.append(layer) class SequentialSupportGraph(object): """An analog of Keras Sequential model for test/support models.""" Loading
deepchem/models/tf_new_models/graph_topology.py +39 −53 Original line number Diff line number Diff line Loading @@ -397,8 +397,8 @@ class DAGGraphTopology(GraphTopology): class WeaveGraphTopology(GraphTopology): """Manages placeholders associated with batch of graphs and their topology""" def __init__(self, batch_size, n_atom_feat, n_pair_feat, max_atoms=100, name='Weave_topology'): def __init__(self, max_atoms, n_atom_feat, n_pair_feat, name='Weave_topology'): """ Parameters ---------- Loading @@ -412,36 +412,30 @@ class WeaveGraphTopology(GraphTopology): #self.n_atoms = n_atoms self.name = name self.batch_size = batch_size self.max_atoms = max_atoms self.n_atom_feat = n_atom_feat self.n_pair_feat = n_pair_feat self.max_atoms = max_atoms * batch_size self.atom_features_placeholder = tf.placeholder( dtype='float32', shape=(None, self.n_atom_feat), shape=(None, self.max_atoms, self.n_atom_feat), name=self.name + '_atom_features') self.atom_mask_placeholder = tf.placeholder( dtype='float32', shape=(None, self.max_atoms), name=self.name + '_atom_mask') self.pair_features_placeholder = tf.placeholder( dtype='float32', shape=(None, self.n_pair_feat), shape=(None, self.max_atoms, self.max_atoms, self.n_pair_feat), name=self.name + '_pair_features') self.pair_split_placeholder = tf.placeholder( dtype='int32', shape=(self.max_atoms,), name=self.name + '_pair_split') self.pair_membership_placeholder = tf.placeholder( dtype='bool', shape=(self.max_atoms,), name=self.name + '_pair_membership') self.atom_split_placeholder = tf.placeholder( dtype='int32', shape=(self.batch_size,), name=self.name + '_atom_split') self.atom_to_pair_placeholder = tf.placeholder( dtype='int32', shape=(None,2), name=self.name + '_atom_to_pair') self.pair_mask_placeholder = tf.placeholder( dtype='float32', shape=(None, self.max_atoms, self.max_atoms), name=self.name + '_pair_mask') self.membership_placeholder = tf.placeholder( dtype='int32', shape=(None,), name=self.name + '_membership') # Define the list of tensors to be used as topology self.topology = [self.pair_split_placeholder, self.pair_membership_placeholder, self.atom_split_placeholder, self.atom_to_pair_placeholder] self.topology = [self.atom_mask_placeholder, self.pair_mask_placeholder] self.inputs = [self.atom_features_placeholder] self.inputs += self.topology Loading @@ -467,42 +461,34 @@ class WeaveGraphTopology(GraphTopology): # Extract atom numbers atom_feat = [] pair_feat = [] atom_split = [] atom_to_pair = [] pair_split = [] atom_mask = [] pair_mask = [] membership = [] max_atoms = self.max_atoms start = 0 for im, mol in enumerate(batch): n_atoms = mol.get_num_atoms() # number of atoms in each molecule atom_split.append(n_atoms) # index of pair features C0, C1 = np.meshgrid(np.arange(n_atoms), np.arange(n_atoms)) atom_to_pair.append(np.transpose(np.array([C1.flatten()+start, C0.flatten()+start]))) start = start + n_atoms # number of pairs for each atom pair_split.extend([n_atoms]*n_atoms) # atom features atom_feat.append(mol.get_atom_features()) # pair features pair_feat.append(np.reshape(mol.get_pair_features(), (n_atoms*n_atoms, self.n_pair_feat))) atom_feat = np.concatenate(atom_feat, axis=0) pair_feat = np.concatenate(pair_feat, axis=0) atom_to_pair = np.concatenate(atom_to_pair, axis=0) atom_split = np.array(atom_split) n_pair = len(pair_split) pair_split = np.pad(pair_split, ((0, max_atoms-n_pair)), 'constant') pair_membership = np.array([True]*n_pair + [False]*(max_atoms-n_pair)) atom_feat.append( np.pad(mol.get_atom_features(), ((0, max_atoms - n_atoms), (0, 0)), 'constant')) atom_mask.append( np.array([1] * n_atoms + [0] * (max_atoms - n_atoms), dtype=float)) pair_feat.append( np.pad(mol.get_pair_features(), ((0, max_atoms - n_atoms), ( 0, max_atoms - n_atoms), (0, 0)), 'constant')) pair_mask.append(np.array([[1]*n_atoms + [0]*(max_atoms-n_atoms)]*n_atoms + \ [[0]*max_atoms]*(max_atoms-n_atoms), dtype=float)) membership.extend([im] * n_atoms) atom_feat = np.stack(atom_feat) pair_feat = np.stack(pair_feat) atom_mask = np.stack(atom_mask) pair_mask = np.stack(pair_mask) membership = np.array(membership) # Generate dicts dict_DTNN = { self.atom_features_placeholder: atom_feat, self.pair_features_placeholder: pair_feat, self.pair_split_placeholder: pair_split, self.pair_membership_placeholder: pair_membership, self.atom_split_placeholder: atom_split, self.atom_to_pair_placeholder: atom_to_pair self.atom_mask_placeholder: atom_mask, self.pair_mask_placeholder: pair_mask, self.membership_placeholder: membership } return dict_DTNN
deepchem/molnet/run_benchmark_models.py +20 −10 Original line number Diff line number Diff line Loading @@ -266,11 +266,16 @@ def benchmark_classification(train_dataset, n_graph_feat = hyper_parameters['n_graph_feat'] n_pair_feat = hyper_parameters['n_pair_feat'] graph_model = deepchem.nn.SequentialWeaveGraph( batch_size, n_atom_feat=n_features, n_pair_feat=n_pair_feat, max_atoms=120) graph_model.add(deepchem.nn.WeaveLayer(75, 14)) graph_model.add(deepchem.nn.WeaveLayer(50, 50)) graph_model.add(deepchem.nn.Dense(n_graph_feat, 50, activation='tanh')) max_atoms_train = max([mol.get_num_atoms() for mol in train_dataset.X]) max_atoms_valid = max([mol.get_num_atoms() for mol in valid_dataset.X]) max_atoms_test = max([mol.get_num_atoms() for mol in test_dataset.X]) max_atoms = max([max_atoms_train, max_atoms_valid, max_atoms_test]) graph_model = deepchem.nn.SequentialWeaveGraph(max_atoms=max_atoms, n_atom_feat=n_features, n_pair_feat=n_pair_feat) graph_model.add(deepchem.nn.WeaveLayer(max_atoms, 75, 14)) graph_model.add(deepchem.nn.WeaveLayer(max_atoms, 50, 50, update_pair=False)) graph_model.add(deepchem.nn.WeaveConcat(batch_size, n_output=n_graph_feat)) graph_model.add(deepchem.nn.BatchNormalization(epsilon=1e-5, mode=1)) graph_model.add( deepchem.nn.WeaveGather( Loading Loading @@ -585,11 +590,16 @@ def benchmark_regression(train_dataset, n_graph_feat = hyper_parameters['n_graph_feat'] n_pair_feat = hyper_parameters['n_pair_feat'] graph_model = deepchem.nn.SequentialWeaveGraph( batch_size, n_atom_feat=n_features, n_pair_feat=n_pair_feat, max_atoms=80) graph_model.add(deepchem.nn.WeaveLayer(75, 14)) graph_model.add(deepchem.nn.WeaveLayer(50, 50)) graph_model.add(deepchem.nn.Dense(n_graph_feat, 50, activation='tanh')) max_atoms_train = max([mol.get_num_atoms() for mol in train_dataset.X]) max_atoms_valid = max([mol.get_num_atoms() for mol in valid_dataset.X]) max_atoms_test = max([mol.get_num_atoms() for mol in test_dataset.X]) max_atoms = max([max_atoms_train, max_atoms_valid, max_atoms_test]) graph_model = deepchem.nn.SequentialWeaveGraph(max_atoms=max_atoms, n_atom_feat=n_features, n_pair_feat=n_pair_feat) graph_model.add(deepchem.nn.WeaveLayer(max_atoms, 75, 14)) graph_model.add(deepchem.nn.WeaveLayer(max_atoms, 50, 50, update_pair=False)) graph_model.add(deepchem.nn.WeaveConcat(batch_size, n_output=n_graph_feat)) graph_model.add(deepchem.nn.BatchNormalization(epsilon=1e-5, mode=1)) graph_model.add( deepchem.nn.WeaveGather( Loading
deepchem/nn/__init__.py +1 −0 Original line number Diff line number Diff line Loading @@ -21,6 +21,7 @@ from deepchem.nn.layers import DAGLayer from deepchem.nn.layers import DAGGather from deepchem.nn.weave_layers import WeaveLayer from deepchem.nn.weave_layers import WeaveConcat from deepchem.nn.weave_layers import WeaveGather from deepchem.nn.model_ops import weight_decay Loading
deepchem/nn/weave_layers.py +122 −47 Original line number Diff line number Diff line Loading @@ -29,6 +29,7 @@ class WeaveLayer(Layer): """ def __init__(self, max_atoms, n_atom_input_feat=75, n_pair_input_feat=14, n_atom_output_feat=50, Loading @@ -37,6 +38,7 @@ class WeaveLayer(Layer): n_hidden_PA=50, n_hidden_AP=50, n_hidden_PP=50, update_pair=True, init='glorot_uniform', activation='relu', dropout=None, Loading @@ -63,8 +65,10 @@ class WeaveLayer(Layer): """ super(WeaveLayer, self).__init__(**kwargs) self.max_atoms = max_atoms self.init = initializations.get(init) # Set weight initialization self.activation = activations.get(activation) # Get activations self.update_pair = update_pair # last weave layer does not need to update self.n_hidden_AA = n_hidden_AA self.n_hidden_PA = n_hidden_PA self.n_hidden_AP = n_hidden_AP Loading Loading @@ -96,30 +100,27 @@ class WeaveLayer(Layer): self.n_atom_output_feat, ]) self.trainable_weights = [ self.W_AA, self.b_AA, self.W_PA, self.b_PA, self.W_A, self.b_A ] if self.update_pair: self.W_AP = self.init([self.n_atom_input_feat * 2, self.n_hidden_AP]) self.b_AP = model_ops.zeros(shape=[ self.n_hidden_AP, ]) self.b_AP = model_ops.zeros(shape=[self.n_hidden_AP,]) self.W_PP = self.init([self.n_pair_input_feat, self.n_hidden_PP]) self.b_PP = model_ops.zeros(shape=[ self.n_hidden_PP, ]) self.b_PP = model_ops.zeros(shape=[self.n_hidden_PP,]) self.W_P = self.init([self.n_hidden_P, self.n_pair_output_feat]) self.b_P = model_ops.zeros(shape=[ self.n_pair_output_feat, ]) self.b_P = model_ops.zeros(shape=[self.n_pair_output_feat,]) self.trainable_weights = [ self.W_AA, self.b_AA, self.W_PA, self.b_PA, self.W_A, self.b_A, self.trainable_weights.extend([ self.W_AP, self.b_AP, self.W_PP, self.b_PP, self.W_P, self.b_P ] ]) def call(self, x, mask=None): """Execute this layer on input tensors. x = [atom_features, pair_features, pair_split, pair_membership, atom_split] x = [atom_features, pair_features, atom_mask, pair_mask] Parameters ---------- Loading @@ -141,39 +142,113 @@ class WeaveLayer(Layer): atom_features = x[0] pair_features = x[1] pair_split = x[2] pair_membership = x[3] atom_split = x[4] atom_to_pair = x[5] atom_mask = x[2] pair_mask = x[3] max_atoms = self.max_atoms AA = tf.matmul(atom_features, self.W_AA) + self.b_AA AA = tf.tensordot(atom_features, self.W_AA, [[2], [0]]) + self.b_AA AA = self.activation(AA) PA = tf.matmul(pair_features, self.W_PA) + self.b_PA PA = tf.reduce_sum( tf.tensordot(pair_features, self.W_PA, [[3], [0]]) + self.b_PA, axis=2) PA = self.activation(PA) PAs = tf.split(PA, pair_split, axis=0) PA = [tf.reduce_sum(molecule, 0) for molecule in PAs] PA = tf.boolean_mask(PA, pair_membership) A = tf.matmul(tf.concat([AA, PA], 1), self.W_A) + self.b_A A = tf.tensordot(tf.concat([AA, PA], 2), self.W_A, [[2], [0]]) + self.b_A A = self.activation(A) AP_ij = tf.matmul(tf.reshape(tf.gather(atom_features, atom_to_pair), [-1, 2*self.n_atom_input_feat]), self.W_AP) + self.b_AP AP_ij = self.activation(AP_ij) AP_ji = tf.matmul(tf.reshape(tf.gather(atom_features, tf.reverse(atom_to_pair, [1])), [-1, 2*self.n_atom_input_feat]), self.W_AP) + self.b_AP AP_ji = self.activation(AP_ji) PP = tf.matmul(pair_features, self.W_PP) + self.b_PP A = tf.multiply(A, tf.expand_dims(atom_mask, axis=2)) if self.update_pair: AP_combine = tf.concat([ tf.stack([atom_features] * max_atoms, axis=2), tf.stack([atom_features] * max_atoms, axis=1) ], 3) AP_combine_t = tf.transpose(AP_combine, perm=[0, 2, 1, 3]) AP = tf.tensordot(AP_combine + AP_combine_t, self.W_AP, [[3], [0]]) + self.b_AP AP = self.activation(AP) PP = tf.tensordot(pair_features, self.W_PP, [[3], [0]]) + self.b_PP PP = self.activation(PP) P = tf.matmul(tf.concat([AP_ij + AP_ji, PP], 1), self.W_P) + self.b_P P = tf.tensordot(tf.concat([AP, PP], 3), self.W_P, [[3], [0]]) + self.b_P P = self.activation(P) P = tf.multiply(P, tf.expand_dims(pair_mask, axis=3)) else: P = pair_features return A, P class WeaveConcat(Layer): """" Concat a batch of molecules into a batch of atoms """ def __init__(self, batch_size, n_atom_input_feat=50, n_output=128, init='glorot_uniform', activation='tanh', **kwargs): """ Parameters ---------- batch_size: int number of molecules in a batch n_atom_input_feat: int, optional Number of features for each atom in input. n_output: int, optional Number of output features for each atom(concatenated) init: str, optional Weight initialization for filters. activation: str, optional Activation function applied """ self.batch_size = batch_size self.n_atom_input_feat = n_atom_input_feat self.n_output = n_output self.init = initializations.get(init) # Set weight initialization self.activation = activations.get(activation) # Get activations super(WeaveConcat, self).__init__(**kwargs) def build(self): """"Construct internal trainable weights. """ self.W = self.init([self.n_atom_input_feat, self.n_output]) self.b = model_ops.zeros(shape=[ self.n_output, ]) self.trainable_weights = self.W + self.b def call(self, x, mask=None): """Execute this layer on input tensors. x = [atom_features, atom_mask] Parameters ---------- x: list Tensors as listed above 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.split(atom_features, self.batch_size, axis=0) A_mask = tf.split( tf.cast(atom_masks, dtype=tf.bool), self.batch_size, axis=0) outputs = tf.concat( [tf.boolean_mask(A[i], A_mask[i]) for i in range(len(A))], axis=0) outputs = tf.matmul(outputs, self.W) + self.b outputs = self.activation(outputs) return outputs class WeaveGather(Layer): """" Gather layer of Weave model a batch of normalized atom features go through a hidden layer, Loading Loading @@ -220,7 +295,7 @@ class WeaveGather(Layer): def call(self, x, mask=None): """Execute this layer on input tensors. x = [atom_features, atom_split] x = [atom_features, membership] Parameters ---------- Loading @@ -237,12 +312,12 @@ class WeaveGather(Layer): # Add trainable weights self.build() outputs = x[0] atom_split = x[1] membership = x[1] if self.gaussian_expand: outputs = self.gaussian_histogram(outputs) outputs = tf.split(outputs, atom_split, axis=0) outputs = tf.dynamic_partition(outputs, membership, self.batch_size) output_molecules = [tf.reduce_sum(molecule, 0) for molecule in outputs] Loading
