Loading deepchem/feat/smiles_featurizers.py +2 −2 Original line number Diff line number Diff line Loading @@ -174,8 +174,8 @@ class SmilesToImage(Featurizer): self.max_len = max_len self.res = res self.img_spec = img_spec self.embed = int(dims * res / 2) super(SmilesToImage, self).__init__(**kwargs) self.embed = int(img_size * res / 2) super(SmilesToImage, self).__init__() def _featurize(self, mol): """Featurizes a single SMILE sequence.""" Loading deepchem/models/__init__.py +1 −0 Original line number Diff line number Diff line Loading @@ -29,6 +29,7 @@ from deepchem.models.tensorgraph.sequential import Sequential from deepchem.models.tensorgraph.models.sequence_dnn import SequenceDNN from deepchem.models.tensorgraph.models.ontology import OntologyModel, OntologyNode, create_gene_ontology from deepchem.models.tensorgraph.models.atomic_conv import AtomicConvModel from deepchem.models.tensorgraph.models.chemnet_models import Smiles2Vec, ChemCeption #################### Compatibility imports for renamed TensorGraph models. Remove below with DeepChem 3.0. #################### Loading deepchem/models/tensorgraph/chemnet_layers.py 0 → 100644 +542 −0 Original line number Diff line number Diff line """ Contains implementations of layers used in ChemCeption and Smiles2Vec models. """ from __future__ import division from __future__ import unicode_literals __author__ = "Vignesh Ram Somnath" __license__ = "MIT" import numpy as np import tensorflow as tf from tensorflow.keras.layers import Conv2D, Concatenate, ReLU, Add, MaxPool2D class Stem(tf.keras.layers.Layer): """ Stem Layer as defined in https://arxiv.org/abs/1710.02238. The structure is significantly altered from the original Inception-ResNet architecture, (https://arxiv.org/abs/1602.07261) but the idea behind this layer is to downsample the image as a preprocessing step for the Inception-ResNet layers, and reduce computational complexity. """ def __init__(self, num_filters, **kwargs): """ Parameters ---------- num_filters: int, Number of convolutional filters """ self.num_filters = num_filters self._build_layer_components() super(Stem, self).__init__(**kwargs) def _build_layer_components(self): """Builds the layers components and set _layers attribute.""" self.conv_layer = Conv2D( filters=self.num_filters, kernel_size=4, strides=2, activation=tf.nn.relu) self.activation_layer = ReLU() self._layers = [self.conv_layer, self.activation_layer] def call(self, inputs): """Invoked when __call__ method of the layer is used.""" conv1 = self.conv_layer(inputs) return self.activation_layer(conv1) class InceptionResnetA(tf.keras.layers.Layer): """ Variant A of the three InceptionResNet layers described in https://arxiv.org/abs/1710.02238. All variants use multiple convolutional blocks with varying kernel sizes and number of filters. This allows capturing patterns over different scales in the inputs. Residual connections are additionally used and have been shown previously to improve convergence and training in deep networks. A 1x1 convolution is used on the concatenated feature maps from the different convolutional blocks, to ensure shapes of inputs and feature maps are same for the residual connection. """ def __init__(self, num_filters, input_dim, **kwargs): """ Parameters ---------- num_filters: int, Number of convolutional filters input_dim: int, Number of channels in the input. """ self.num_filters = num_filters self.input_dim = input_dim self._build_layer_components() super(InceptionResnetA, self).__init__(**kwargs) def _build_layer_components(self): """Builds the layers components and set _layers attribute.""" self.conv_block1 = [ Conv2D( self.num_filters, kernel_size=(1, 1), strides=1, padding="same", activation=tf.nn.relu) ] self.conv_block2 = [ Conv2D( filters=self.num_filters, kernel_size=(1, 1), strides=1, activation=tf.nn.relu, padding="same") ] self.conv_block2.append( Conv2D( filters=self.num_filters, kernel_size=(3, 3), strides=1, activation=tf.nn.relu, padding="same")) self.conv_block3 = [ Conv2D( filters=self.num_filters, kernel_size=1, strides=1, activation=tf.nn.relu, padding="same") ] self.conv_block3.append( Conv2D( filters=int(self.num_filters * 1.5), kernel_size=(3, 3), strides=1, activation=tf.nn.relu, padding="same")) self.conv_block3.append( Conv2D( filters=self.num_filters * 2, kernel_size=(3, 3), strides=1, activation=tf.nn.relu, padding="same")) self.conv_block4 = [ Conv2D( filters=self.input_dim, kernel_size=(1, 1), strides=1, padding="same") ] self.concat_layer = Concatenate() self.add_layer = Add() self.activation_layer = ReLU() self._layers = self.conv_block1 + self.conv_block2 + self.conv_block3 + self.conv_block4 self._layers.extend( [self.concat_layer, self.add_layer, self.activation_layer]) def call(self, inputs): """Invoked when __call__ method of the layer is used.""" conv1 = inputs for layer in self.conv_block1: conv1 = layer(conv1) conv2 = inputs for layer in self.conv_block2: conv2 = layer(conv2) conv3 = inputs for layer in self.conv_block3: conv3 = layer(conv3) concat_conv = self.concat_layer([conv1, conv2, conv3]) conv4 = concat_conv for layer in self.conv_block4: conv4 = layer(conv4) output = self.add_layer([conv4, inputs]) return self.activation_layer(output) class InceptionResnetB(tf.keras.layers.Layer): """ Variant B of the three InceptionResNet layers described in https://arxiv.org/abs/1710.02238. All variants use multiple convolutional blocks with varying kernel sizes and number of filters. This allows capturing patterns over different scales in the inputs. Residual connections are additionally used and have been shown previously to improve convergence and training in deep networks. A 1x1 convolution is used on the concatenated feature maps from the different convolutional blocks, to ensure shapes of inputs and feature maps are same for the residual connection. """ def __init__(self, num_filters, input_dim, **kwargs): """ Parameters ---------- num_filters: int, Number of convolutional filters input_dim: int, Number of channels in the input. """ self.num_filters = num_filters self.input_dim = input_dim self._build_layer_components() super(InceptionResnetB, self).__init__(**kwargs) def _build_layer_components(self): """Builds the layers components and set _layers attribute.""" self.conv_block1 = [ Conv2D( self.num_filters, kernel_size=1, strides=1, padding="same", activation=tf.nn.relu) ] self.conv_block2 = [ Conv2D( filters=self.num_filters, kernel_size=(1, 1), strides=1, activation=tf.nn.relu, padding="same") ] self.conv_block2.append( Conv2D( filters=int(self.num_filters * 1.25), kernel_size=(1, 7), strides=1, activation=tf.nn.relu, padding="same")) self.conv_block2.append( Conv2D( filters=int(self.num_filters * 1.5), kernel_size=(7, 1), strides=1, activation=tf.nn.relu, padding="same")) self.conv_block3 = [ Conv2D( filters=self.input_dim, kernel_size=1, strides=1, padding="same") ] self.concat_layer = Concatenate() self.add_layer = Add() self.activation_layer = ReLU() self._layers = self.conv_block1 + self.conv_block2 + self.conv_block3 self._layers.extend( [self.concat_layer, self.add_layer, self.activation_layer]) def call(self, inputs): """Invoked when __call__ method of the layer is used.""" conv1 = inputs for layer in self.conv_block1: conv1 = layer(conv1) conv2 = inputs for layer in self.conv_block2: conv2 = layer(conv2) concat_conv = self.concat_layer([conv1, conv2]) conv3 = concat_conv for layer in self.conv_block3: conv3 = layer(conv3) output = self.add_layer([conv3, inputs]) output = self.activation_layer(output) return output class InceptionResnetC(tf.keras.layers.Layer): """ Variant C of the three InceptionResNet layers described in https://arxiv.org/abs/1710.02238. All variants use multiple convolutional blocks with varying kernel sizes and number of filters. This allows capturing patterns over different scales in the inputs. Residual connections are additionally used and have been shown previously to improve convergence and training in deep networks. A 1x1 convolution is used on the concatenated feature maps from the different convolutional blocks, to ensure shapes of inputs and feature maps are same for the residual connection. """ def __init__(self, num_filters, input_dim, **kwargs): """ Parameters ---------- num_filters: int, Number of convolutional filters input_dim: int, Number of channels in the input. """ self.num_filters = num_filters self.input_dim = input_dim self._build_layer_components() super(InceptionResnetC, self).__init__(**kwargs) def _build_layer_components(self): """Builds the layers components and set _layers attribute.""" self.conv_block1 = [ Conv2D( self.num_filters, kernel_size=(1, 1), strides=1, padding="same", activation=tf.nn.relu) ] self.conv_block2 = [ Conv2D( filters=self.num_filters, kernel_size=1, strides=1, activation=tf.nn.relu, padding="same") ] self.conv_block2.append( Conv2D( filters=int(self.num_filters * 1.16), kernel_size=(1, 3), strides=1, activation=tf.nn.relu, padding="same")) self.conv_block2.append( Conv2D( filters=int(self.num_filters * 1.33), kernel_size=(3, 1), strides=1, activation=tf.nn.relu, padding="same")) self.conv_block3 = [ Conv2D( filters=self.input_dim, kernel_size=(1, 1), strides=1, padding="same") ] self.concat_layer = Concatenate() self.add_layer = Add() self.activation_layer = ReLU() self._layers = self.conv_block1 + self.conv_block2 + self.conv_block3 self._layers.extend( [self.concat_layer, self.add_layer, self.activation_layer]) def call(self, inputs): """Invoked when __call__ method of the layer is used.""" conv1 = inputs for layer in self.conv_block1: conv1 = layer(conv1) conv2 = inputs for layer in self.conv_block2: conv2 = layer(conv2) concat_conv = self.concat_layer([conv1, conv2]) conv3 = concat_conv for layer in self.conv_block3: conv3 = layer(conv3) output = self.add_layer([conv3, inputs]) output = self.activation_layer(output) return output class ReductionA(tf.keras.layers.Layer): """ Variant A of the two Reduction layers described in https://arxiv.org/abs/1710.02238. All variants use multiple convolutional blocks with varying kernel sizes and number of filters, to reduce the spatial extent of the image and reduce computational complexity for downstream layers. """ def __init__(self, num_filters, **kwargs): """ Parameters ---------- num_filters: int, Number of convolutional filters """ self.num_filters = num_filters self._build_layer_components() super(ReductionA, self).__init__(**kwargs) def _build_layer_components(self): """Builds the layers components and set _layers attribute.""" self.max_pool1 = MaxPool2D(pool_size=(3, 3), strides=2, padding="valid") self.conv_block1 = [ Conv2D( int(self.num_filters * 1.5), kernel_size=(3, 3), strides=2, padding="valid", activation=tf.nn.relu) ] self.conv_block2 = [ Conv2D( filters=self.num_filters, kernel_size=1, strides=1, activation=tf.nn.relu, padding="same") ] self.conv_block2.append( Conv2D( filters=self.num_filters, kernel_size=3, strides=1, activation=tf.nn.relu, padding="same")) self.conv_block2.append( Conv2D( filters=int(self.num_filters * 1.5), kernel_size=3, strides=2, activation=tf.nn.relu, padding="valid")) self.concat_layer = Concatenate() self.activation_layer = ReLU() self._layers = self.conv_block1 + self.conv_block2 self._layers.extend( [self.max_pool1, self.concat_layer, self.activation_layer]) def call(self, inputs): """Invoked when __call__ method of the layer is used.""" maxpool1 = self.max_pool1(inputs) conv1 = inputs for layer in self.conv_block1: conv1 = layer(conv1) conv2 = inputs for layer in self.conv_block2: conv2 = layer(conv2) output = self.concat_layer([maxpool1, conv1, conv2]) output = self.activation_layer(output) return output class ReductionB(tf.keras.layers.Layer): """ Variant B of the two Reduction layers described in https://arxiv.org/abs/1710.02238. All variants use multiple convolutional blocks with varying kernel sizes and number of filters, to reduce the spatial extent of the image and reduce computational complexity for downstream layers. """ def __init__(self, num_filters, **kwargs): """ Parameters ---------- num_filters: int, Number of convolutional filters """ self.num_filters = num_filters self._build_layer_components() super(ReductionB, self).__init__(**kwargs) def _build_layer_components(self): """Builds the layers components and set _layers attribute.""" self.max_pool1 = MaxPool2D(pool_size=(3, 3), strides=2, padding="valid") self.conv_block1 = [ Conv2D( self.num_filters, kernel_size=1, strides=1, padding="same", activation=tf.nn.relu) ] self.conv_block1.append( Conv2D( int(self.num_filters * 1.5), kernel_size=3, strides=2, padding="valid", activation=tf.nn.relu)) self.conv_block2 = [ Conv2D( filters=self.num_filters, kernel_size=1, strides=1, activation=tf.nn.relu, padding="same") ] self.conv_block2.append( Conv2D( filters=int(self.num_filters * 1.125), kernel_size=3, strides=2, activation=tf.nn.relu, padding="valid")) self.conv_block3 = [ Conv2D( filters=self.num_filters, kernel_size=1, strides=1, activation=tf.nn.relu, padding="same") ] self.conv_block3.append( Conv2D( filters=int(self.num_filters * 1.125), kernel_size=(3, 1), strides=1, activation=tf.nn.relu, padding="same")) self.conv_block3.append( Conv2D( filters=int(self.num_filters * 1.25), kernel_size=(3, 3), strides=2, activation=tf.nn.relu, padding="valid")) self.concat_layer = Concatenate() self.activation_layer = ReLU() self._layers = self.conv_block1 + self.conv_block2 + self.conv_block3 self._layers.extend( [self.max_pool1, self.concat_layer, self.activation_layer]) def call(self, inputs): """Invoked when __call__ method of the layer is used.""" maxpool1 = self.max_pool1(inputs) conv1 = inputs for layer in self.conv_block1: conv1 = layer(conv1) conv2 = inputs for layer in self.conv_block2: conv2 = layer(conv2) conv3 = inputs for layer in self.conv_block3: conv3 = layer(conv3) concat = self.concat_layer([maxpool1, conv1, conv2, conv3]) output = self.activation_layer(concat) return output deepchem/models/tensorgraph/models/chemnet_models.py 0 → 100644 +349 −0 File added.Preview size limit exceeded, changes collapsed. Show changes deepchem/models/tensorgraph/optimizers.py +41 −3 Original line number Diff line number Diff line Loading @@ -80,6 +80,44 @@ class Adam(Optimizer): epsilon=self.epsilon) class RMSProp(Optimizer): """RMSProp Optimization algorithm.""" def __init__(self, learning_rate=0.001, momentum=0.0, decay=0.9, epsilon=1e-10): """Construct an RMSProp Optimizer. Parameters ---------- learning_rate: float or LearningRateSchedule the learning_rate used for optimization momentum: float, default 0.0 a parameter of the RMSProp algorithm decay: float, default 0.9 a parameter of the RMSProp algorithm epsilon: float, default 1e-10 a parameter of the RMSProp algorithm """ self.learning_rate = learning_rate self.momentum = momentum self.decay = decay self.epsilon = epsilon def _create_optimizer(self, global_step): if isinstance(self.learning_rate, LearningRateSchedule): learning_rate = self.learning_rate._create_tensor(global_step) else: learning_rate = self.learning_rate return tf.train.RMSPropOptimizer( learning_rate=learning_rate, momentum=self.momentum, decay=self.decay, epsilon=self.epsilon) class GradientDescent(Optimizer): """The gradient descent optimization algorithm.""" Loading Loading
deepchem/feat/smiles_featurizers.py +2 −2 Original line number Diff line number Diff line Loading @@ -174,8 +174,8 @@ class SmilesToImage(Featurizer): self.max_len = max_len self.res = res self.img_spec = img_spec self.embed = int(dims * res / 2) super(SmilesToImage, self).__init__(**kwargs) self.embed = int(img_size * res / 2) super(SmilesToImage, self).__init__() def _featurize(self, mol): """Featurizes a single SMILE sequence.""" Loading
deepchem/models/__init__.py +1 −0 Original line number Diff line number Diff line Loading @@ -29,6 +29,7 @@ from deepchem.models.tensorgraph.sequential import Sequential from deepchem.models.tensorgraph.models.sequence_dnn import SequenceDNN from deepchem.models.tensorgraph.models.ontology import OntologyModel, OntologyNode, create_gene_ontology from deepchem.models.tensorgraph.models.atomic_conv import AtomicConvModel from deepchem.models.tensorgraph.models.chemnet_models import Smiles2Vec, ChemCeption #################### Compatibility imports for renamed TensorGraph models. Remove below with DeepChem 3.0. #################### Loading
deepchem/models/tensorgraph/chemnet_layers.py 0 → 100644 +542 −0 Original line number Diff line number Diff line """ Contains implementations of layers used in ChemCeption and Smiles2Vec models. """ from __future__ import division from __future__ import unicode_literals __author__ = "Vignesh Ram Somnath" __license__ = "MIT" import numpy as np import tensorflow as tf from tensorflow.keras.layers import Conv2D, Concatenate, ReLU, Add, MaxPool2D class Stem(tf.keras.layers.Layer): """ Stem Layer as defined in https://arxiv.org/abs/1710.02238. The structure is significantly altered from the original Inception-ResNet architecture, (https://arxiv.org/abs/1602.07261) but the idea behind this layer is to downsample the image as a preprocessing step for the Inception-ResNet layers, and reduce computational complexity. """ def __init__(self, num_filters, **kwargs): """ Parameters ---------- num_filters: int, Number of convolutional filters """ self.num_filters = num_filters self._build_layer_components() super(Stem, self).__init__(**kwargs) def _build_layer_components(self): """Builds the layers components and set _layers attribute.""" self.conv_layer = Conv2D( filters=self.num_filters, kernel_size=4, strides=2, activation=tf.nn.relu) self.activation_layer = ReLU() self._layers = [self.conv_layer, self.activation_layer] def call(self, inputs): """Invoked when __call__ method of the layer is used.""" conv1 = self.conv_layer(inputs) return self.activation_layer(conv1) class InceptionResnetA(tf.keras.layers.Layer): """ Variant A of the three InceptionResNet layers described in https://arxiv.org/abs/1710.02238. All variants use multiple convolutional blocks with varying kernel sizes and number of filters. This allows capturing patterns over different scales in the inputs. Residual connections are additionally used and have been shown previously to improve convergence and training in deep networks. A 1x1 convolution is used on the concatenated feature maps from the different convolutional blocks, to ensure shapes of inputs and feature maps are same for the residual connection. """ def __init__(self, num_filters, input_dim, **kwargs): """ Parameters ---------- num_filters: int, Number of convolutional filters input_dim: int, Number of channels in the input. """ self.num_filters = num_filters self.input_dim = input_dim self._build_layer_components() super(InceptionResnetA, self).__init__(**kwargs) def _build_layer_components(self): """Builds the layers components and set _layers attribute.""" self.conv_block1 = [ Conv2D( self.num_filters, kernel_size=(1, 1), strides=1, padding="same", activation=tf.nn.relu) ] self.conv_block2 = [ Conv2D( filters=self.num_filters, kernel_size=(1, 1), strides=1, activation=tf.nn.relu, padding="same") ] self.conv_block2.append( Conv2D( filters=self.num_filters, kernel_size=(3, 3), strides=1, activation=tf.nn.relu, padding="same")) self.conv_block3 = [ Conv2D( filters=self.num_filters, kernel_size=1, strides=1, activation=tf.nn.relu, padding="same") ] self.conv_block3.append( Conv2D( filters=int(self.num_filters * 1.5), kernel_size=(3, 3), strides=1, activation=tf.nn.relu, padding="same")) self.conv_block3.append( Conv2D( filters=self.num_filters * 2, kernel_size=(3, 3), strides=1, activation=tf.nn.relu, padding="same")) self.conv_block4 = [ Conv2D( filters=self.input_dim, kernel_size=(1, 1), strides=1, padding="same") ] self.concat_layer = Concatenate() self.add_layer = Add() self.activation_layer = ReLU() self._layers = self.conv_block1 + self.conv_block2 + self.conv_block3 + self.conv_block4 self._layers.extend( [self.concat_layer, self.add_layer, self.activation_layer]) def call(self, inputs): """Invoked when __call__ method of the layer is used.""" conv1 = inputs for layer in self.conv_block1: conv1 = layer(conv1) conv2 = inputs for layer in self.conv_block2: conv2 = layer(conv2) conv3 = inputs for layer in self.conv_block3: conv3 = layer(conv3) concat_conv = self.concat_layer([conv1, conv2, conv3]) conv4 = concat_conv for layer in self.conv_block4: conv4 = layer(conv4) output = self.add_layer([conv4, inputs]) return self.activation_layer(output) class InceptionResnetB(tf.keras.layers.Layer): """ Variant B of the three InceptionResNet layers described in https://arxiv.org/abs/1710.02238. All variants use multiple convolutional blocks with varying kernel sizes and number of filters. This allows capturing patterns over different scales in the inputs. Residual connections are additionally used and have been shown previously to improve convergence and training in deep networks. A 1x1 convolution is used on the concatenated feature maps from the different convolutional blocks, to ensure shapes of inputs and feature maps are same for the residual connection. """ def __init__(self, num_filters, input_dim, **kwargs): """ Parameters ---------- num_filters: int, Number of convolutional filters input_dim: int, Number of channels in the input. """ self.num_filters = num_filters self.input_dim = input_dim self._build_layer_components() super(InceptionResnetB, self).__init__(**kwargs) def _build_layer_components(self): """Builds the layers components and set _layers attribute.""" self.conv_block1 = [ Conv2D( self.num_filters, kernel_size=1, strides=1, padding="same", activation=tf.nn.relu) ] self.conv_block2 = [ Conv2D( filters=self.num_filters, kernel_size=(1, 1), strides=1, activation=tf.nn.relu, padding="same") ] self.conv_block2.append( Conv2D( filters=int(self.num_filters * 1.25), kernel_size=(1, 7), strides=1, activation=tf.nn.relu, padding="same")) self.conv_block2.append( Conv2D( filters=int(self.num_filters * 1.5), kernel_size=(7, 1), strides=1, activation=tf.nn.relu, padding="same")) self.conv_block3 = [ Conv2D( filters=self.input_dim, kernel_size=1, strides=1, padding="same") ] self.concat_layer = Concatenate() self.add_layer = Add() self.activation_layer = ReLU() self._layers = self.conv_block1 + self.conv_block2 + self.conv_block3 self._layers.extend( [self.concat_layer, self.add_layer, self.activation_layer]) def call(self, inputs): """Invoked when __call__ method of the layer is used.""" conv1 = inputs for layer in self.conv_block1: conv1 = layer(conv1) conv2 = inputs for layer in self.conv_block2: conv2 = layer(conv2) concat_conv = self.concat_layer([conv1, conv2]) conv3 = concat_conv for layer in self.conv_block3: conv3 = layer(conv3) output = self.add_layer([conv3, inputs]) output = self.activation_layer(output) return output class InceptionResnetC(tf.keras.layers.Layer): """ Variant C of the three InceptionResNet layers described in https://arxiv.org/abs/1710.02238. All variants use multiple convolutional blocks with varying kernel sizes and number of filters. This allows capturing patterns over different scales in the inputs. Residual connections are additionally used and have been shown previously to improve convergence and training in deep networks. A 1x1 convolution is used on the concatenated feature maps from the different convolutional blocks, to ensure shapes of inputs and feature maps are same for the residual connection. """ def __init__(self, num_filters, input_dim, **kwargs): """ Parameters ---------- num_filters: int, Number of convolutional filters input_dim: int, Number of channels in the input. """ self.num_filters = num_filters self.input_dim = input_dim self._build_layer_components() super(InceptionResnetC, self).__init__(**kwargs) def _build_layer_components(self): """Builds the layers components and set _layers attribute.""" self.conv_block1 = [ Conv2D( self.num_filters, kernel_size=(1, 1), strides=1, padding="same", activation=tf.nn.relu) ] self.conv_block2 = [ Conv2D( filters=self.num_filters, kernel_size=1, strides=1, activation=tf.nn.relu, padding="same") ] self.conv_block2.append( Conv2D( filters=int(self.num_filters * 1.16), kernel_size=(1, 3), strides=1, activation=tf.nn.relu, padding="same")) self.conv_block2.append( Conv2D( filters=int(self.num_filters * 1.33), kernel_size=(3, 1), strides=1, activation=tf.nn.relu, padding="same")) self.conv_block3 = [ Conv2D( filters=self.input_dim, kernel_size=(1, 1), strides=1, padding="same") ] self.concat_layer = Concatenate() self.add_layer = Add() self.activation_layer = ReLU() self._layers = self.conv_block1 + self.conv_block2 + self.conv_block3 self._layers.extend( [self.concat_layer, self.add_layer, self.activation_layer]) def call(self, inputs): """Invoked when __call__ method of the layer is used.""" conv1 = inputs for layer in self.conv_block1: conv1 = layer(conv1) conv2 = inputs for layer in self.conv_block2: conv2 = layer(conv2) concat_conv = self.concat_layer([conv1, conv2]) conv3 = concat_conv for layer in self.conv_block3: conv3 = layer(conv3) output = self.add_layer([conv3, inputs]) output = self.activation_layer(output) return output class ReductionA(tf.keras.layers.Layer): """ Variant A of the two Reduction layers described in https://arxiv.org/abs/1710.02238. All variants use multiple convolutional blocks with varying kernel sizes and number of filters, to reduce the spatial extent of the image and reduce computational complexity for downstream layers. """ def __init__(self, num_filters, **kwargs): """ Parameters ---------- num_filters: int, Number of convolutional filters """ self.num_filters = num_filters self._build_layer_components() super(ReductionA, self).__init__(**kwargs) def _build_layer_components(self): """Builds the layers components and set _layers attribute.""" self.max_pool1 = MaxPool2D(pool_size=(3, 3), strides=2, padding="valid") self.conv_block1 = [ Conv2D( int(self.num_filters * 1.5), kernel_size=(3, 3), strides=2, padding="valid", activation=tf.nn.relu) ] self.conv_block2 = [ Conv2D( filters=self.num_filters, kernel_size=1, strides=1, activation=tf.nn.relu, padding="same") ] self.conv_block2.append( Conv2D( filters=self.num_filters, kernel_size=3, strides=1, activation=tf.nn.relu, padding="same")) self.conv_block2.append( Conv2D( filters=int(self.num_filters * 1.5), kernel_size=3, strides=2, activation=tf.nn.relu, padding="valid")) self.concat_layer = Concatenate() self.activation_layer = ReLU() self._layers = self.conv_block1 + self.conv_block2 self._layers.extend( [self.max_pool1, self.concat_layer, self.activation_layer]) def call(self, inputs): """Invoked when __call__ method of the layer is used.""" maxpool1 = self.max_pool1(inputs) conv1 = inputs for layer in self.conv_block1: conv1 = layer(conv1) conv2 = inputs for layer in self.conv_block2: conv2 = layer(conv2) output = self.concat_layer([maxpool1, conv1, conv2]) output = self.activation_layer(output) return output class ReductionB(tf.keras.layers.Layer): """ Variant B of the two Reduction layers described in https://arxiv.org/abs/1710.02238. All variants use multiple convolutional blocks with varying kernel sizes and number of filters, to reduce the spatial extent of the image and reduce computational complexity for downstream layers. """ def __init__(self, num_filters, **kwargs): """ Parameters ---------- num_filters: int, Number of convolutional filters """ self.num_filters = num_filters self._build_layer_components() super(ReductionB, self).__init__(**kwargs) def _build_layer_components(self): """Builds the layers components and set _layers attribute.""" self.max_pool1 = MaxPool2D(pool_size=(3, 3), strides=2, padding="valid") self.conv_block1 = [ Conv2D( self.num_filters, kernel_size=1, strides=1, padding="same", activation=tf.nn.relu) ] self.conv_block1.append( Conv2D( int(self.num_filters * 1.5), kernel_size=3, strides=2, padding="valid", activation=tf.nn.relu)) self.conv_block2 = [ Conv2D( filters=self.num_filters, kernel_size=1, strides=1, activation=tf.nn.relu, padding="same") ] self.conv_block2.append( Conv2D( filters=int(self.num_filters * 1.125), kernel_size=3, strides=2, activation=tf.nn.relu, padding="valid")) self.conv_block3 = [ Conv2D( filters=self.num_filters, kernel_size=1, strides=1, activation=tf.nn.relu, padding="same") ] self.conv_block3.append( Conv2D( filters=int(self.num_filters * 1.125), kernel_size=(3, 1), strides=1, activation=tf.nn.relu, padding="same")) self.conv_block3.append( Conv2D( filters=int(self.num_filters * 1.25), kernel_size=(3, 3), strides=2, activation=tf.nn.relu, padding="valid")) self.concat_layer = Concatenate() self.activation_layer = ReLU() self._layers = self.conv_block1 + self.conv_block2 + self.conv_block3 self._layers.extend( [self.max_pool1, self.concat_layer, self.activation_layer]) def call(self, inputs): """Invoked when __call__ method of the layer is used.""" maxpool1 = self.max_pool1(inputs) conv1 = inputs for layer in self.conv_block1: conv1 = layer(conv1) conv2 = inputs for layer in self.conv_block2: conv2 = layer(conv2) conv3 = inputs for layer in self.conv_block3: conv3 = layer(conv3) concat = self.concat_layer([maxpool1, conv1, conv2, conv3]) output = self.activation_layer(concat) return output
deepchem/models/tensorgraph/models/chemnet_models.py 0 → 100644 +349 −0 File added.Preview size limit exceeded, changes collapsed. Show changes
deepchem/models/tensorgraph/optimizers.py +41 −3 Original line number Diff line number Diff line Loading @@ -80,6 +80,44 @@ class Adam(Optimizer): epsilon=self.epsilon) class RMSProp(Optimizer): """RMSProp Optimization algorithm.""" def __init__(self, learning_rate=0.001, momentum=0.0, decay=0.9, epsilon=1e-10): """Construct an RMSProp Optimizer. Parameters ---------- learning_rate: float or LearningRateSchedule the learning_rate used for optimization momentum: float, default 0.0 a parameter of the RMSProp algorithm decay: float, default 0.9 a parameter of the RMSProp algorithm epsilon: float, default 1e-10 a parameter of the RMSProp algorithm """ self.learning_rate = learning_rate self.momentum = momentum self.decay = decay self.epsilon = epsilon def _create_optimizer(self, global_step): if isinstance(self.learning_rate, LearningRateSchedule): learning_rate = self.learning_rate._create_tensor(global_step) else: learning_rate = self.learning_rate return tf.train.RMSPropOptimizer( learning_rate=learning_rate, momentum=self.momentum, decay=self.decay, epsilon=self.epsilon) class GradientDescent(Optimizer): """The gradient descent optimization algorithm.""" Loading
