Loading deepchem/data/datasets.py +5 −2 Original line number Diff line number Diff line Loading @@ -1049,8 +1049,11 @@ class Databag(object): A utility class to iterate through multiple datasets together. """ def __init__(self): def __init__(self, datasets=None): if datasets is None: self.datasets = dict() else: self.datasets = datasets def add_dataset(self, key, dataset): self.datasets[key] = dataset Loading deepchem/models/tensorgraph/layers.py +109 −6 Original line number Diff line number Diff line Loading @@ -49,6 +49,25 @@ class Layer(object): def __hash__(self): return hash(self.__key()) def __call__(self, *in_layers): if len(in_layers) > 0: layers = [] for in_layer in in_layers: if isinstance(in_layer, Layer): layers.append(layer) elif isinstance(in_layer, tf.Tensor): layers.append(TensorWrapper(in_layer)) else: raise ValueError("Layer must be invoked on layers or tensors") self.in_layers = layers self._create_tensor() class TensorWrapper(Layer): """Used to wrap a tensorflow tensor.""" def __init__(self, out_tensor): self.out_tensor = out_tensor class Conv1DLayer(Layer): Loading Loading @@ -269,7 +288,8 @@ class SoftMax(Layer): class Concat(Layer): def __init__(self, **kwargs): def __init__(self, axis=1, **kwargs): self.axis = axis super(Concat, self).__init__(**kwargs) def _create_tensor(self): Loading @@ -278,10 +298,30 @@ class Concat(Layer): return self.out_tensor out_tensors = [x.out_tensor for x in self.in_layers] self.out_tensor = tf.concat(out_tensors, 1) self.out_tensor = tf.concat(out_tensors, axis=self.axis) return self.out_tensor class InteratomicL2Distances(Layer): """Compute (squared) L2 Distances between atoms given neighbors.""" def _create_tensor(self): if len(self.in_layers) != 2: raise ValueError("InteratomicDistances requires coords,nbr_list") coords, nbr_list = (self.in_layers[0].out_tensor, self.in_layers[1].out_tensor) N_atoms, ndim = coords.get_shape() _, M = nbr_list.get_shape() # Shape (N_atoms, M, ndim) nbr_coords = tf.gather(coords, nbr_list) # Shape (N_atoms, M, ndim) tiled_atom_coords = tf.tile( tf.reshape(atom_coords, (N_atoms, 1, ndim)), (1, M, 1)) # Shape (N_atoms, M) dists = tf.reduce_sum((tiled_atom_coords - nbr_coords)**2, axis=2) class SoftMaxCrossEntropy(Layer): def __init__(self, **kwargs): Loading @@ -299,6 +339,10 @@ class SoftMaxCrossEntropy(Layer): class ReduceMean(Layer): def __init__(self, axis=None, **kwargs): self.axis=axis super(ReduceMean, self).__init__(**kwargs) def _create_tensor(self): if len(self.in_layers) > 1: out_tensors = [x.out_tensor for x in self.in_layers] Loading @@ -309,16 +353,41 @@ class ReduceMean(Layer): self.out_tensor = tf.reduce_mean(self.out_tensor) return self.out_tensor class ToFloat(Layer): def _create_tensor(self): if len(self.in_layers) > 1: raise ValueError("Only one layer supported.") self.out_tensor = tf.to_float(self.in_layers[0].out_tensor) return self.out_tensor class ReduceSum(Layer): def __init__(self, axis=None, **kwargs): self.axis=axis super(ReduceSum, self).__init__(**kwargs) def _create_tensor(self): if len(self.in_layers) > 1: out_tensors = [x.out_tensor for x in self.in_layers] self.out_tensor = tf.stack(out_tensors) else: self.out_tensor = self.in_layers[0].out_tensor self.out_tensor = tf.reduce_sum(self.out_tensor, axis=self.axis) return self.out_tensor class ReduceSquareDifference(Layer): def __init__(self, **kwargs): def __init__(self, axis=None, **kwargs): self.axis=axis super(ReduceSquareDifference, self).__init__(**kwargs) def _create_tensor(self): a = self.in_layers[0].out_tensor b = self.in_layers[1].out_tensor self.out_tensor = tf.reduce_mean(tf.squared_difference(a, b)) self.out_tensor = tf.reduce_mean(tf.squared_difference(a, b), axis=self.axis) return self.out_tensor Loading Loading @@ -651,10 +720,44 @@ class VinaHydrogenBond(Layer): tf.where(d < 0, (1.0 / 0.7) * (0 - d), tf.zeros_like(d))) return self.out_tensor def VinaGaussianFirst(Layer): class VinaGaussianFirst(Layer): """Computes Autodock Vina's first Gaussian interaction term.""" pass def _create_tensor(self): d = self.in_layers[0].out_tensor self.out_tensor = tf.exp(-(d / 0.5)**2) return self.out_tensor class VinaGaussianSecond(Layer): """Computes Autodock Vina's second Gaussian interaction term.""" def _create_tensor(self): d = self.in_layers[0].out_tensor self.out_tensor = tf.exp(-((d - 3) / 2)**2) return self.out_tensor class WeightedLinearCombo(Layer): """Computes a weighted linear combination of input layers.""" def __init__(self, std=.3, **kwargs): self.std = std super(WeightedLinearCombo, self).__init__(**kwargs) def _create_tensor(self): weights = [] out_tensor = None for in_layer in self.in_layers: w = tf.Variable(tf.random_normal([1,], stddev=self.std)) if out_tensor is None: out_tensor = w * in_layer.out_tensor else: out_tensor += w * in_layer.out_tensor self.out_tensor = out_tensor return self.out_tensor class NeighborList(Layer): """Computes a neighbor-list on the GPU. Loading deepchem/models/tensorgraph/tests/test_docking.py 0 → 100644 +106 −0 Original line number Diff line number Diff line import unittest import numpy as np import os from nose.tools import assert_true import deepchem as dc from deepchem.data import NumpyDataset from deepchem.data.datasets import Databag from deepchem.models.tensorgraph.layers import ReduceSum from deepchem.models.tensorgraph.layers import Feature, Label from deepchem.models.tensorgraph.layers import ToFloat from deepchem.models.tensorgraph.layers import NeighborList from deepchem.models.tensorgraph.layers import ReduceSquareDifference from deepchem.models.tensorgraph.layers import WeightedLinearCombo from deepchem.models.tensorgraph.layers import InteratomicL2Distances from deepchem.models.tensorgraph.tensor_graph import TensorGraph class TestDocking(unittest.TestCase): """ Test that tensorgraph docking-style models work. """ def test_neighbor_list(self): """Test that neighbor lists can be constructed.""" N_atoms = 10 start = 0 stop = 12 nbr_cutoff = 3 ndim = 3 M = 6 k = 5 # The number of cells which we should theoretically have n_cells = int(((stop - start) / nbr_cutoff)**ndim) X = np.random.rand(N_atoms, ndim) y = np.random.rand(N_atoms, 1) dataset = NumpyDataset(X, y) features = Feature(shape=(N_atoms, ndim)) labels = Label(shape=(N_atoms,)) nbr_list = NeighborList(N_atoms, M, ndim, n_cells, k, nbr_cutoff, in_layers=[features]) nbr_list = ToFloat(in_layers=[nbr_list]) # This isn't a meaningful loss, but just for test loss = ReduceSum(in_layers=[nbr_list]) tg = dc.models.TensorGraph(use_queue=False) tg.add_output(nbr_list) tg.set_loss(loss) tg.build() def test_weighted_combo(self): """Tests that weighted linear combinations can be built""" N = 10 n_features = 5 X1 = NumpyDataset(np.random.rand(N, n_features)) X2 = NumpyDataset(np.random.rand(N, n_features)) y = NumpyDataset(np.random.rand(N)) features_1 = Feature(shape=(None, n_features)) features_2 = Feature(shape=(None, n_features)) labels = Label(shape=(None,)) combo = WeightedLinearCombo(in_layers=[features_1, features_2]) out = ReduceSum(in_layers=[combo], axis=1) loss = ReduceSquareDifference(in_layers=[out, labels]) databag = Databag({features_1: X1, features_2: X2, labels: y}) tg = dc.models.TensorGraph(learning_rate=0.1, use_queue=False) tg.set_loss(loss) tg.fit_generator(databag.iterbatches(epochs=1)) def test_vina(self): """Test that vina graph can be constructed in TensorGraph.""" prot_coords = Features(shape=(N_protein, 3)) prot_Z = Features(shape=(N_protein,), dtype=tf.int32) ligand_coords = Features(shape=(N_ligand, 3)) ligand_Z = Features(shape=(N_ligand,), dtype=tf.int32) labels = Label(shape=(1,)) coords = Concat(in_layers=[prot_coords, ligand_coords], axis=0) Z = Concat(in_layers=[prot_Z, ligand_Z], axis=0) # Now an (N, M) shape nbr_list = NeighborList(N_protein+N_ligand, M, ndim, n_cells, k, nbr_cutoff, in_layers=[coords]) def test_interatomic_distances(self): """Test that the interatomic distance calculation works.""" N_atoms = 5 M = 2 ndim = 3 coords = np.random.rand(N_atoms, ndim) nbr_list = np.random.randint(0, N_atoms, size=(N_atoms, M)) coords_tensor = tf.convert_to_tensor(coords) nbr_list_tensor = tf.convert_to_tensor(nbr_list) dist_tensor = deepchem/models/tensorgraph/tests/test_tensor_graph.py +0 −28 Original line number Diff line number Diff line Loading @@ -208,31 +208,3 @@ class TestTensorGraph(unittest.TestCase): tg1 = TensorGraph.load_from_dir(tg.model_dir) prediction2 = np.squeeze(tg1.predict_proba_on_batch(X)) assert_true(np.all(np.isclose(prediction, prediction2, atol=0.01))) def test_neighbor_list(self): N_atoms = 10 start = 0 stop = 12 nbr_cutoff = 3 ndim = 3 M = 6 k = 5 # The number of cells which we should theoretically have n_cells = int(((stop - start) / nbr_cutoff)**ndim) X = np.random.rand(N_atoms, ndim) y = np.random.rand(N_atoms, 1) dataset = NumpyDataset(X, y) features = Feature(shape=(N_atoms, ndim)) labels = Label(shape=(N_atoms,)) nbr_list = NeighborList(N_atoms, M, ndim, n_cells, k, nbr_cutoff, in_layers=[features]) # This isn't a meaningful loss, but just for test loss = ReduceMean(in_layers=[nbr_list]) tg = dc.models.TensorGraph(use_queue=False) tg.add_output(nbr_list) tg.set_loss(loss) tg.fit(dataset, nb_epoch=1) Loading
deepchem/data/datasets.py +5 −2 Original line number Diff line number Diff line Loading @@ -1049,8 +1049,11 @@ class Databag(object): A utility class to iterate through multiple datasets together. """ def __init__(self): def __init__(self, datasets=None): if datasets is None: self.datasets = dict() else: self.datasets = datasets def add_dataset(self, key, dataset): self.datasets[key] = dataset Loading
deepchem/models/tensorgraph/layers.py +109 −6 Original line number Diff line number Diff line Loading @@ -49,6 +49,25 @@ class Layer(object): def __hash__(self): return hash(self.__key()) def __call__(self, *in_layers): if len(in_layers) > 0: layers = [] for in_layer in in_layers: if isinstance(in_layer, Layer): layers.append(layer) elif isinstance(in_layer, tf.Tensor): layers.append(TensorWrapper(in_layer)) else: raise ValueError("Layer must be invoked on layers or tensors") self.in_layers = layers self._create_tensor() class TensorWrapper(Layer): """Used to wrap a tensorflow tensor.""" def __init__(self, out_tensor): self.out_tensor = out_tensor class Conv1DLayer(Layer): Loading Loading @@ -269,7 +288,8 @@ class SoftMax(Layer): class Concat(Layer): def __init__(self, **kwargs): def __init__(self, axis=1, **kwargs): self.axis = axis super(Concat, self).__init__(**kwargs) def _create_tensor(self): Loading @@ -278,10 +298,30 @@ class Concat(Layer): return self.out_tensor out_tensors = [x.out_tensor for x in self.in_layers] self.out_tensor = tf.concat(out_tensors, 1) self.out_tensor = tf.concat(out_tensors, axis=self.axis) return self.out_tensor class InteratomicL2Distances(Layer): """Compute (squared) L2 Distances between atoms given neighbors.""" def _create_tensor(self): if len(self.in_layers) != 2: raise ValueError("InteratomicDistances requires coords,nbr_list") coords, nbr_list = (self.in_layers[0].out_tensor, self.in_layers[1].out_tensor) N_atoms, ndim = coords.get_shape() _, M = nbr_list.get_shape() # Shape (N_atoms, M, ndim) nbr_coords = tf.gather(coords, nbr_list) # Shape (N_atoms, M, ndim) tiled_atom_coords = tf.tile( tf.reshape(atom_coords, (N_atoms, 1, ndim)), (1, M, 1)) # Shape (N_atoms, M) dists = tf.reduce_sum((tiled_atom_coords - nbr_coords)**2, axis=2) class SoftMaxCrossEntropy(Layer): def __init__(self, **kwargs): Loading @@ -299,6 +339,10 @@ class SoftMaxCrossEntropy(Layer): class ReduceMean(Layer): def __init__(self, axis=None, **kwargs): self.axis=axis super(ReduceMean, self).__init__(**kwargs) def _create_tensor(self): if len(self.in_layers) > 1: out_tensors = [x.out_tensor for x in self.in_layers] Loading @@ -309,16 +353,41 @@ class ReduceMean(Layer): self.out_tensor = tf.reduce_mean(self.out_tensor) return self.out_tensor class ToFloat(Layer): def _create_tensor(self): if len(self.in_layers) > 1: raise ValueError("Only one layer supported.") self.out_tensor = tf.to_float(self.in_layers[0].out_tensor) return self.out_tensor class ReduceSum(Layer): def __init__(self, axis=None, **kwargs): self.axis=axis super(ReduceSum, self).__init__(**kwargs) def _create_tensor(self): if len(self.in_layers) > 1: out_tensors = [x.out_tensor for x in self.in_layers] self.out_tensor = tf.stack(out_tensors) else: self.out_tensor = self.in_layers[0].out_tensor self.out_tensor = tf.reduce_sum(self.out_tensor, axis=self.axis) return self.out_tensor class ReduceSquareDifference(Layer): def __init__(self, **kwargs): def __init__(self, axis=None, **kwargs): self.axis=axis super(ReduceSquareDifference, self).__init__(**kwargs) def _create_tensor(self): a = self.in_layers[0].out_tensor b = self.in_layers[1].out_tensor self.out_tensor = tf.reduce_mean(tf.squared_difference(a, b)) self.out_tensor = tf.reduce_mean(tf.squared_difference(a, b), axis=self.axis) return self.out_tensor Loading Loading @@ -651,10 +720,44 @@ class VinaHydrogenBond(Layer): tf.where(d < 0, (1.0 / 0.7) * (0 - d), tf.zeros_like(d))) return self.out_tensor def VinaGaussianFirst(Layer): class VinaGaussianFirst(Layer): """Computes Autodock Vina's first Gaussian interaction term.""" pass def _create_tensor(self): d = self.in_layers[0].out_tensor self.out_tensor = tf.exp(-(d / 0.5)**2) return self.out_tensor class VinaGaussianSecond(Layer): """Computes Autodock Vina's second Gaussian interaction term.""" def _create_tensor(self): d = self.in_layers[0].out_tensor self.out_tensor = tf.exp(-((d - 3) / 2)**2) return self.out_tensor class WeightedLinearCombo(Layer): """Computes a weighted linear combination of input layers.""" def __init__(self, std=.3, **kwargs): self.std = std super(WeightedLinearCombo, self).__init__(**kwargs) def _create_tensor(self): weights = [] out_tensor = None for in_layer in self.in_layers: w = tf.Variable(tf.random_normal([1,], stddev=self.std)) if out_tensor is None: out_tensor = w * in_layer.out_tensor else: out_tensor += w * in_layer.out_tensor self.out_tensor = out_tensor return self.out_tensor class NeighborList(Layer): """Computes a neighbor-list on the GPU. Loading
deepchem/models/tensorgraph/tests/test_docking.py 0 → 100644 +106 −0 Original line number Diff line number Diff line import unittest import numpy as np import os from nose.tools import assert_true import deepchem as dc from deepchem.data import NumpyDataset from deepchem.data.datasets import Databag from deepchem.models.tensorgraph.layers import ReduceSum from deepchem.models.tensorgraph.layers import Feature, Label from deepchem.models.tensorgraph.layers import ToFloat from deepchem.models.tensorgraph.layers import NeighborList from deepchem.models.tensorgraph.layers import ReduceSquareDifference from deepchem.models.tensorgraph.layers import WeightedLinearCombo from deepchem.models.tensorgraph.layers import InteratomicL2Distances from deepchem.models.tensorgraph.tensor_graph import TensorGraph class TestDocking(unittest.TestCase): """ Test that tensorgraph docking-style models work. """ def test_neighbor_list(self): """Test that neighbor lists can be constructed.""" N_atoms = 10 start = 0 stop = 12 nbr_cutoff = 3 ndim = 3 M = 6 k = 5 # The number of cells which we should theoretically have n_cells = int(((stop - start) / nbr_cutoff)**ndim) X = np.random.rand(N_atoms, ndim) y = np.random.rand(N_atoms, 1) dataset = NumpyDataset(X, y) features = Feature(shape=(N_atoms, ndim)) labels = Label(shape=(N_atoms,)) nbr_list = NeighborList(N_atoms, M, ndim, n_cells, k, nbr_cutoff, in_layers=[features]) nbr_list = ToFloat(in_layers=[nbr_list]) # This isn't a meaningful loss, but just for test loss = ReduceSum(in_layers=[nbr_list]) tg = dc.models.TensorGraph(use_queue=False) tg.add_output(nbr_list) tg.set_loss(loss) tg.build() def test_weighted_combo(self): """Tests that weighted linear combinations can be built""" N = 10 n_features = 5 X1 = NumpyDataset(np.random.rand(N, n_features)) X2 = NumpyDataset(np.random.rand(N, n_features)) y = NumpyDataset(np.random.rand(N)) features_1 = Feature(shape=(None, n_features)) features_2 = Feature(shape=(None, n_features)) labels = Label(shape=(None,)) combo = WeightedLinearCombo(in_layers=[features_1, features_2]) out = ReduceSum(in_layers=[combo], axis=1) loss = ReduceSquareDifference(in_layers=[out, labels]) databag = Databag({features_1: X1, features_2: X2, labels: y}) tg = dc.models.TensorGraph(learning_rate=0.1, use_queue=False) tg.set_loss(loss) tg.fit_generator(databag.iterbatches(epochs=1)) def test_vina(self): """Test that vina graph can be constructed in TensorGraph.""" prot_coords = Features(shape=(N_protein, 3)) prot_Z = Features(shape=(N_protein,), dtype=tf.int32) ligand_coords = Features(shape=(N_ligand, 3)) ligand_Z = Features(shape=(N_ligand,), dtype=tf.int32) labels = Label(shape=(1,)) coords = Concat(in_layers=[prot_coords, ligand_coords], axis=0) Z = Concat(in_layers=[prot_Z, ligand_Z], axis=0) # Now an (N, M) shape nbr_list = NeighborList(N_protein+N_ligand, M, ndim, n_cells, k, nbr_cutoff, in_layers=[coords]) def test_interatomic_distances(self): """Test that the interatomic distance calculation works.""" N_atoms = 5 M = 2 ndim = 3 coords = np.random.rand(N_atoms, ndim) nbr_list = np.random.randint(0, N_atoms, size=(N_atoms, M)) coords_tensor = tf.convert_to_tensor(coords) nbr_list_tensor = tf.convert_to_tensor(nbr_list) dist_tensor =
deepchem/models/tensorgraph/tests/test_tensor_graph.py +0 −28 Original line number Diff line number Diff line Loading @@ -208,31 +208,3 @@ class TestTensorGraph(unittest.TestCase): tg1 = TensorGraph.load_from_dir(tg.model_dir) prediction2 = np.squeeze(tg1.predict_proba_on_batch(X)) assert_true(np.all(np.isclose(prediction, prediction2, atol=0.01))) def test_neighbor_list(self): N_atoms = 10 start = 0 stop = 12 nbr_cutoff = 3 ndim = 3 M = 6 k = 5 # The number of cells which we should theoretically have n_cells = int(((stop - start) / nbr_cutoff)**ndim) X = np.random.rand(N_atoms, ndim) y = np.random.rand(N_atoms, 1) dataset = NumpyDataset(X, y) features = Feature(shape=(N_atoms, ndim)) labels = Label(shape=(N_atoms,)) nbr_list = NeighborList(N_atoms, M, ndim, n_cells, k, nbr_cutoff, in_layers=[features]) # This isn't a meaningful loss, but just for test loss = ReduceMean(in_layers=[nbr_list]) tg = dc.models.TensorGraph(use_queue=False) tg.add_output(nbr_list) tg.set_loss(loss) tg.fit(dataset, nb_epoch=1)
