Loading deepchem/models/tensorgraph/layers.py +151 −55 Original line number Diff line number Diff line Loading @@ -8,6 +8,7 @@ import tensorflow as tf import numpy as np from deepchem.models.tensorgraph import model_ops, initializations, regularizers, activations from deepchem.models.tensorgraph.model_ops import create_variable import tensorflow.contrib.eager as tfe import math Loading Loading @@ -88,9 +89,8 @@ class Layer(object): return self.clone(in_layers) raise ValueError('%s does not implement shared()' % self.__class__.__name__) def __call__(self, *in_layers, training=False, **kwargs): return self.create_tensor( in_layers=in_layers, set_tensors=False, training=training, **kwargs) def __call__(self, *in_layers, **kwargs): return self.create_tensor(in_layers=in_layers, set_tensors=False, **kwargs) @property def shape(self): Loading Loading @@ -508,8 +508,6 @@ class Conv1D(Layer): if tfe.in_eager_mode(): if not self._built: self._layer = self._build_layer() self.variables = self._layer.variables self._built = True layer = self._layer else: layer = self._build_layer() Loading @@ -517,6 +515,9 @@ class Conv1D(Layer): if set_tensors: self._record_variable_scope(self.name) self.out_tensor = out_tensor if tfe.in_eager_mode() and not self._built: self._built = True self.variables = self._layer.variables return out_tensor Loading Loading @@ -871,7 +872,7 @@ class CombineMeanStd(Layer): mean_parent, std_parent = inputs[0], inputs[1] sample_noise = tf.random_normal( mean_parent.get_shape(), 0, self.noise_epsilon, dtype=tf.float32) if self.training_only: if self.training_only and 'training' in kwargs: sample_noise *= kwargs['training'] out_tensor = mean_parent + tf.exp(std_parent * 0.5) * sample_noise if set_tensors: Loading Loading @@ -2772,7 +2773,7 @@ class LSTMStep(Layer): self.W = init((self.input_dim, 4 * self.output_dim)) self.U = inner_init((self.output_dim, 4 * self.output_dim)) self.b = model_ops.create_variable( self.b = create_variable( np.hstack((np.zeros(self.output_dim), np.ones(self.output_dim), np.zeros(self.output_dim), np.zeros(self.output_dim))), dtype=tf.float32) Loading Loading @@ -3129,20 +3130,41 @@ class IterRefLSTMEmbedding(Layer): class BatchNorm(Layer): def __init__(self, in_layers=None, **kwargs): def __init__(self, in_layers=None, axis=-1, momentum=0.99, epsilon=1e-3, **kwargs): super(BatchNorm, self).__init__(in_layers, **kwargs) self.axis = axis self.momentum = momentum self.epsilon = epsilon try: parent_shape = self.in_layers[0].shape self._shape = tuple(self.in_layers[0].shape) except: pass def _build_layer(self): return tf.layers.BatchNormalization( axis=self.axis, momentum=self.momentum, epsilon=self.epsilon) def create_tensor(self, in_layers=None, set_tensors=True, **kwargs): inputs = self._get_input_tensors(in_layers) parent_tensor = inputs[0] out_tensor = tf.layers.batch_normalization(parent_tensor) if tfe.in_eager_mode(): if not self._built: self._layer = self._build_layer() layer = self._layer else: layer = self._build_layer() out_tensor = layer(parent_tensor) if set_tensors: self.out_tensor = out_tensor if tfe.in_eager_mode() and not self._built: self._built = True self.variables = self._layer.variables return out_tensor Loading @@ -3155,6 +3177,9 @@ class BatchNormalization(Layer): beta_init='zero', gamma_init='one', **kwargs): warnings.warn( 'BatchNormalization is deprecated and will be removed in a future release. Use BatchNorm instead.', DeprecationWarning) self.beta_init = initializations.get(beta_init) self.gamma_init = initializations.get(gamma_init) self.epsilon = epsilon Loading Loading @@ -3242,13 +3267,17 @@ class VinaFreeEnergy(Layer): self.stop = stop super(VinaFreeEnergy, self).__init__(**kwargs) def _build_layers(self): weighted_combo = WeightedLinearCombo() w = create_variable(tf.random_normal((1,), stddev=self.stddev)) return (weighted_combo, w) def cutoff(self, d, x): out_tensor = tf.where(d < 8, x, tf.zeros_like(x)) return out_tensor def nonlinearity(self, c): def nonlinearity(self, c, w): """Computes non-linearity used in Vina.""" w = tf.Variable(tf.random_normal((1,), stddev=self.stddev)) out_tensor = c / (1 + w * self.Nrot) return w, out_tensor Loading Loading @@ -3298,6 +3327,14 @@ class VinaFreeEnergy(Layer): X = inputs[0] Z = inputs[1] if tfe.in_eager_mode(): if not self._built: self._weighted_combo, self._w = self._build_layers() weighted_combo = self._weighted_combo w = self._w else: weighted_combo, w = self._build_layers() # TODO(rbharath): This layer shouldn't be neighbor-listing. Make # neighbors lists an argument instead of a part of this layer. nbr_list = NeighborList(self.N_atoms, self.M_nbrs, self.ndim, Loading @@ -3314,20 +3351,22 @@ class VinaFreeEnergy(Layer): gauss_2 = self.gaussian_second(dists) # Shape (N, M) weighted_combo = WeightedLinearCombo() interactions = weighted_combo(repulsion, hydrophobic, hbond, gauss_1, gauss_2) # Shape (N, M) thresholded = self.cutoff(dists, interactions) weight, free_energies = self.nonlinearity(thresholded) weight, free_energies = self.nonlinearity(thresholded, w) free_energy = ReduceSum()(free_energies) out_tensor = free_energy if set_tensors: self._record_variable_scope(self.name) self.out_tensor = out_tensor if tfe.in_eager_mode() and not self._built: self.variables = weighted_combo.variables + [w] self._built = True return out_tensor Loading @@ -3342,14 +3381,23 @@ class WeightedLinearCombo(Layer): except: pass def _create_variables(self, inputs): return [ create_variable(tf.random_normal([1], stddev=self.std)) for i in range(len(inputs)) ] def create_tensor(self, in_layers=None, set_tensors=True, **kwargs): inputs = self._get_input_tensors(in_layers, True) weights = [] if tfe.in_eager_mode(): if not self._built: self.variables = self._create_variables(inputs) self._built = True weights = self.variables else: weights = self._create_variables(inputs) out_tensor = None for in_tensor in inputs: w = tf.Variable(tf.random_normal([ 1, ], stddev=self.std)) for in_tensor, w in zip(inputs, weights): if out_tensor is None: out_tensor = w * in_tensor else: Loading Loading @@ -3659,7 +3707,8 @@ class Dropout(Layer): def create_tensor(self, in_layers=None, set_tensors=True, **kwargs): inputs = self._get_input_tensors(in_layers) parent_tensor = inputs[0] keep_prob = 1.0 - self.dropout_prob * kwargs['training'] training = kwargs['training'] if 'training' in kwargs else 1.0 keep_prob = 1.0 - self.dropout_prob * training out_tensor = tf.nn.dropout(parent_tensor, keep_prob) if set_tensors: self.out_tensor = out_tensor Loading Loading @@ -3763,11 +3812,18 @@ class AtomicConvolution(Layer): R = self.distance_matrix(D) sym = [] rsf_zeros = tf.zeros((B, N, M)) for param in self.radial_params: for i, param in enumerate(self.radial_params): if tfe.in_eager_mode(): if not self._built: self.variables += self._create_radial_variables(*param) param_variables = self.variables[3 * i:3 * i + 3] else: param_variables = self._create_radial_variables(*param) # We apply the radial pooling filter before atom type conv # to reduce computation param_variables, rsf = self.radial_symmetry_function(R, *param) rsf = self.radial_symmetry_function(R, *param_variables) if not self.atom_types: cond = tf.not_equal(Nbrs_Z, 0.0) Loading @@ -3784,8 +3840,17 @@ class AtomicConvolution(Layer): if set_tensors: self._record_variable_scope(self.name) self.out_tensor = out_tensor if tfe.in_eager_mode() and not self._built: self._built = True return out_tensor def _create_radial_variables(self, rc, rs, e): with tf.name_scope(None, "NbrRadialSymmetryFunction", [rc, rs, e]): rc = create_variable(rc) rs = create_variable(rs) e = create_variable(e) return (rc, rs, e) def radial_symmetry_function(self, R, rc, rs, e): """Calculates radial symmetry function. Loading @@ -3809,13 +3874,9 @@ class AtomicConvolution(Layer): """ with tf.name_scope(None, "NbrRadialSymmetryFunction", [rc, rs, e]): rc = tf.Variable(rc) rs = tf.Variable(rs) e = tf.Variable(e) K = self.gaussian_distance_matrix(R, rs, e) FC = self.radial_cutoff(R, rc) return [rc, rs, e], tf.multiply(K, FC) return tf.multiply(K, FC) def radial_cutoff(self, R, rc): """Calculates radial cutoff matrix. Loading Loading @@ -4021,7 +4082,17 @@ class AlphaShareLayer(Layer): subspaces = tf.reshape(tf.stack(subspaces), [n_alphas, -1]) # create the alpha learnable parameters alphas = tf.Variable(tf.random_normal([n_alphas, n_alphas]), name='alphas') if tfe.in_eager_mode(): if not self._built: self.variables = [ create_variable( tf.random_normal([n_alphas, n_alphas]), name='alphas') ] self._built = True alphas = self.variables[0] else: alphas = create_variable( tf.random_normal([n_alphas, n_alphas]), name='alphas') subspaces = tf.matmul(alphas, subspaces) Loading Loading @@ -4113,7 +4184,15 @@ class BetaShare(Layer): n_betas = len(inputs) subspaces = tf.reshape(tf.stack(subspaces), [n_betas, -1]) betas = tf.Variable(tf.random_normal([1, n_betas]), name='betas') if tfe.in_eager_mode(): if not self._built: self.variables = [ create_variable(tf.random_normal([1, n_betas]), name='betas') ] self._built = True betas = self.variables[0] else: betas = create_variable(tf.random_normal([1, n_betas]), name='betas') out_tensor = tf.matmul(betas, subspaces) self.betas = betas self.out_tensor = tf.reshape(out_tensor, [-1, original_cols]) Loading @@ -4134,7 +4213,7 @@ class ANIFeat(Layer): """ def __init__(self, in_layers, in_layers=None, max_atoms=23, radial_cutoff=4.6, angular_cutoff=3.1, Loading Loading @@ -4405,17 +4484,25 @@ class GraphEmbedPoolLayer(Layer): self.out_tensors = [result, result_A] return result, result_A def _create_variables(self, no_features, no_filters, name): W = create_variable( tf.truncated_normal( [no_features, no_filters], stddev=1.0 / math.sqrt(no_features)), name='%s_weights' % name, dtype=tf.float32) b = create_variable( tf.constant(0.1), name='%s_bias' % self.name, dtype=tf.float32) return [W, b] def embedding_factors(self, V, no_filters, name="default"): no_features = V.get_shape()[-1].value W = tf.get_variable( '%s_weights' % name, [no_features, no_filters], initializer=tf.truncated_normal_initializer( stddev=1.0 / math.sqrt(no_features)), dtype=tf.float32) b = tf.get_variable( '%s_bias' % self.name, [no_filters], initializer=tf.constant_initializer(0.1), dtype=tf.float32) if tfe.in_eager_mode(): if not self._built: self.variables = self._create_variables(no_features, no_filters, name) self._built = True W, b = self.variables else: W, b = self._create_variables(no_features, no_filters, name) V_reshape = tf.reshape(V, (-1, no_features)) s = tf.slice(tf.shape(V), [0], [len(V.get_shape()) - 1]) s = tf.concat([s, tf.stack([no_filters])], 0) Loading Loading @@ -4487,6 +4574,23 @@ class GraphCNN(Layer): self.num_filters = num_filters super(GraphCNN, self).__init__(**kwargs) def _create_variables(self, no_features, no_A): W = create_variable( 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 = create_variable( 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 = create_variable( tf.constant(0.1), name='%s_bias' % self.name, dtype=tf.float32) return [W, W_I, b] def create_tensor(self, in_layers=None, set_tensors=True, **kwargs): inputs = self._get_input_tensors(in_layers) if len(inputs) == 3: Loading @@ -4495,21 +4599,13 @@ class GraphCNN(Layer): V, A = inputs no_A = A.get_shape()[2].value no_features = V.get_shape()[2].value W = tf.get_variable( '%s_weights' % self.name, [no_features * no_A, self.num_filters], initializer=tf.truncated_normal_initializer( stddev=math.sqrt(1.0 / (no_features * (no_A + 1) * 1.0))), dtype=tf.float32) W_I = tf.get_variable( '%s_weights_I' % self.name, [no_features, self.num_filters], initializer=tf.truncated_normal_initializer( stddev=math.sqrt(1.0 / (no_features * (no_A + 1) * 1.0))), dtype=tf.float32) b = tf.get_variable( '%s_bias' % self.name, [self.num_filters], initializer=tf.constant_initializer(0.1), dtype=tf.float32) if tfe.in_eager_mode(): if not self._built: self.variables = self._create_variables(no_features, no_A) self._built = True W, W_I, b = self.variables else: W, W_I, b = self._create_variables(no_features, no_A) n = self.graphConvolution(V, A) A_shape = tf.shape(A) Loading deepchem/models/tensorgraph/models/graph_models.py +3 −3 Original line number Diff line number Diff line Loading @@ -14,7 +14,7 @@ from deepchem.models.tensorgraph.graph_layers import WeaveGather, \ from deepchem.models.tensorgraph.graph_layers import WeaveLayerFactory from deepchem.models.tensorgraph.layers import Dense, SoftMax, \ SoftMaxCrossEntropy, GraphConv, BatchNorm, \ GraphPool, GraphGather, WeightedError, Dropout, BatchNormalization, Stack, Flatten, GraphCNN, GraphCNNPool GraphPool, GraphGather, WeightedError, Dropout, BatchNorm, Stack, Flatten, GraphCNN, GraphCNNPool from deepchem.models.tensorgraph.layers import L2Loss, Label, Weights, Feature from deepchem.models.tensorgraph.tensor_graph import TensorGraph from deepchem.trans import undo_transforms Loading Loading @@ -86,7 +86,7 @@ class WeaveTensorGraph(TensorGraph): out_channels=self.n_graph_feat, activation_fn=tf.nn.tanh, in_layers=weave_layer2A) batch_norm1 = BatchNormalization(epsilon=1e-5, mode=1, in_layers=[dense1]) batch_norm1 = BatchNorm(epsilon=1e-5, in_layers=[dense1]) weave_gather = WeaveGather( self.batch_size, n_input=self.n_graph_feat, Loading deepchem/models/tensorgraph/models/text_cnn.py +1 −1 Original line number Diff line number Diff line Loading @@ -9,7 +9,7 @@ import copy from deepchem.metrics import to_one_hot, from_one_hot from deepchem.models.tensorgraph.layers import Dense, Concat, SoftMax, \ SoftMaxCrossEntropy, BatchNorm, WeightedError, Dropout, BatchNormalization, \ SoftMaxCrossEntropy, BatchNorm, WeightedError, Dropout, \ Conv1D, ReduceMax, Squeeze, Stack, Highway from deepchem.models.tensorgraph.graph_layers import DTNNEmbedding Loading deepchem/models/tensorgraph/tests/test_layers_eager.py +249 −1 Original line number Diff line number Diff line Loading @@ -28,6 +28,7 @@ class TestLayersEager(test_util.TensorFlowTestCase): result = layer(input) self.assertEqual(result.shape[0], batch_size) self.assertEqual(result.shape[2], filters) assert len(layer.variables) == 2 # Creating a second layer should produce different results, since it has # different random weights. Loading Loading @@ -137,7 +138,7 @@ class TestLayersEager(test_util.TensorFlowTestCase): mean = np.random.rand(5, 3).astype(np.float32) std = np.random.rand(5, 3).astype(np.float32) layer = layers.CombineMeanStd(training_only=True, noise_epsilon=0.01) result1 = layer(mean, std) result1 = layer(mean, std, training=False) assert np.array_equal(result1, mean) # No noise in test mode result2 = layer(mean, std, training=True) assert not np.array_equal(result2, mean) Loading Loading @@ -730,3 +731,250 @@ class TestLayersEager(test_util.TensorFlowTestCase): assert test_out.shape == (n_test, n_feat) assert support_out.shape == (n_support, n_feat) assert len(layer.variables) == 8 def test_batch_norm(self): """Test invoking BatchNorm in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): batch_size = 10 n_features = 5 input = np.random.rand(batch_size, n_features).astype(np.float32) layer = layers.BatchNorm() result = layer(input) assert result.shape == (batch_size, n_features) assert len(layer.variables) == 4 def test_weighted_error(self): """Test invoking WeightedError in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): input1 = np.random.rand(5, 10).astype(np.float32) input2 = np.random.rand(5, 10).astype(np.float32) result = layers.WeightedError()(input1, input2) expected = np.sum(input1 * input2) assert np.allclose(result, expected) def test_vina_free_energy(self): """Test invoking VinaFreeEnergy in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): n_atoms = 5 m_nbrs = 1 ndim = 3 nbr_cutoff = 1 start = 0 stop = 4 X = np.random.rand(n_atoms, ndim).astype(np.float32) Z = np.random.randint(0, 2, (n_atoms)).astype(np.float32) layer = layers.VinaFreeEnergy(n_atoms, m_nbrs, ndim, nbr_cutoff, start, stop) result = layer(X, Z) assert len(layer.variables) == 6 assert result.shape == tuple() # Creating a second layer should produce different results, since it has # different random weights. layer2 = layers.VinaFreeEnergy(n_atoms, m_nbrs, ndim, nbr_cutoff, start, stop) result2 = layer2(X, Z) assert not np.allclose(result, result2) # But evaluating the first layer again should produce the same result as before. result3 = layer(X, Z) assert np.allclose(result, result3) def test_weighted_linear_combo(self): """Test invoking WeightedLinearCombo in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): input1 = np.random.rand(5, 10).astype(np.float32) input2 = np.random.rand(5, 10).astype(np.float32) layer = layers.WeightedLinearCombo() result = layer(input1, input2) assert len(layer.variables) == 2 expected = input1 * layer.variables[0] + input2 * layer.variables[1] assert np.allclose(result, expected) def test_neighbor_list(self): """Test invoking NeighborList in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): N_atoms = 5 start = 0 stop = 12 nbr_cutoff = 3 ndim = 3 M_nbrs = 2 coords = start + np.random.rand(N_atoms, ndim) * (stop - start) coords = tf.to_float(tf.stack(coords)) layer = layers.NeighborList(N_atoms, M_nbrs, ndim, nbr_cutoff, start, stop) result = layer(coords) assert result.shape == (N_atoms, M_nbrs) def test_dropout(self): """Test invoking Dropout in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): rate = 0.5 input = np.random.rand(5, 10).astype(np.float32) layer = layers.Dropout(rate) result1 = layer(input, training=False) assert np.allclose(result1, input) result2 = layer(input, training=True) assert not np.allclose(result2, input) nonzero = result2.numpy() != 0 assert np.allclose(result2.numpy()[nonzero], input[nonzero] / rate) def test_atomic_convolution(self): """Test invoking AtomicConvolution in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): batch_size = 4 max_atoms = 5 max_neighbors = 2 dimensions = 3 params = [[5.0, 2.0, 0.5], [10.0, 2.0, 0.5]] input1 = np.random.rand(batch_size, max_atoms, dimensions).astype( np.float32) input2 = np.random.randint( max_atoms, size=(batch_size, max_atoms, max_neighbors)) input3 = np.random.randint( 1, 10, size=(batch_size, max_atoms, max_neighbors)) layer = layers.AtomicConvolution(radial_params=params) result = layer(input1, input2, input3) assert result.shape == (batch_size, max_atoms, len(params)) assert len(layer.variables) == 3 * len(params) def test_alpha_share_layer(self): """Test invoking AlphaShareLayer in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): batch_size = 10 length = 6 input1 = np.random.rand(batch_size, length).astype(np.float32) input2 = np.random.rand(batch_size, length).astype(np.float32) layer = layers.AlphaShareLayer() result = layer(input1, input2) assert input1.shape == result[0].shape assert input2.shape == result[1].shape # Creating a second layer should produce different results, since it has # different random weights. layer2 = layers.AlphaShareLayer() result2 = layer2(input1, input2) assert not np.allclose(result[0], result2[0]) assert not np.allclose(result[1], result2[1]) # But evaluating the first layer again should produce the same result as before. result3 = layer(input1, input2) assert np.allclose(result[0], result3[0]) assert np.allclose(result[1], result3[1]) def test_sluice_loss(self): """Test invoking SluiceLoss in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): input1 = np.ones((3, 4)).astype(np.float32) input2 = np.ones((2, 2)).astype(np.float32) result = layers.SluiceLoss()(input1, input2) assert np.allclose(result, 40.0) def test_beta_share(self): """Test invoking BetaShare in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): batch_size = 10 length = 6 input1 = np.random.rand(batch_size, length).astype(np.float32) input2 = np.random.rand(batch_size, length).astype(np.float32) layer = layers.BetaShare() result = layer(input1, input2) assert input1.shape == result.shape assert input2.shape == result.shape # Creating a second layer should produce different results, since it has # different random weights. layer2 = layers.BetaShare() result2 = layer2(input1, input2) assert not np.allclose(result, result2) # But evaluating the first layer again should produce the same result as before. result3 = layer(input1, input2) assert np.allclose(result, result3) def test_ani_feat(self): """Test invoking ANIFeat in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): batch_size = 10 max_atoms = 5 input = np.random.rand(batch_size, max_atoms, 4).astype(np.float32) layer = layers.ANIFeat(max_atoms=max_atoms) result = layer(input) # TODO What should the output shape be? It's not documented, and there # are no other test cases for it. def test_graph_embed_pool_layer(self): """Test invoking GraphEmbedPoolLayer in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): V = np.random.uniform(size=(10, 100, 50)).astype(np.float32) adjs = np.random.uniform(size=(10, 100, 5, 100)).astype(np.float32) layer = layers.GraphEmbedPoolLayer(num_vertices=6) result = layer(V, adjs) assert result[0].shape == (10, 6, 50) assert result[1].shape == (10, 6, 5, 6) # Creating a second layer should produce different results, since it has # different random weights. layer2 = layers.GraphEmbedPoolLayer(num_vertices=6) result2 = layer2(V, adjs) assert not np.allclose(result[0], result2[0]) assert not np.allclose(result[1], result2[1]) # But evaluating the first layer again should produce the same result as before. result3 = layer(V, adjs) assert np.allclose(result[0], result3[0]) assert np.allclose(result[1], result3[1]) def test_graph_cnn(self): """Test invoking GraphCNN in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): V = np.random.uniform(size=(10, 100, 50)).astype(np.float32) adjs = np.random.uniform(size=(10, 100, 5, 100)).astype(np.float32) layer = layers.GraphCNN(num_filters=6) result = layer(V, adjs) assert result.shape == (10, 100, 6) # Creating a second layer should produce different results, since it has # different random weights. layer2 = layers.GraphCNN(num_filters=6) result2 = layer2(V, adjs) assert not np.allclose(result, result2) # But evaluating the first layer again should produce the same result as before. result3 = layer(V, adjs) assert np.allclose(result, result3) def test_hinge_loss(self): """Test invoking HingeLoss in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): n_labels = 1 n_logits = 1 logits = np.random.rand(n_logits).astype(np.float32) labels = np.random.rand(n_labels).astype(np.float32) result = layers.Hingeloss()(labels, logits) assert result.shape == (n_labels,) Loading
deepchem/models/tensorgraph/layers.py +151 −55 Original line number Diff line number Diff line Loading @@ -8,6 +8,7 @@ import tensorflow as tf import numpy as np from deepchem.models.tensorgraph import model_ops, initializations, regularizers, activations from deepchem.models.tensorgraph.model_ops import create_variable import tensorflow.contrib.eager as tfe import math Loading Loading @@ -88,9 +89,8 @@ class Layer(object): return self.clone(in_layers) raise ValueError('%s does not implement shared()' % self.__class__.__name__) def __call__(self, *in_layers, training=False, **kwargs): return self.create_tensor( in_layers=in_layers, set_tensors=False, training=training, **kwargs) def __call__(self, *in_layers, **kwargs): return self.create_tensor(in_layers=in_layers, set_tensors=False, **kwargs) @property def shape(self): Loading Loading @@ -508,8 +508,6 @@ class Conv1D(Layer): if tfe.in_eager_mode(): if not self._built: self._layer = self._build_layer() self.variables = self._layer.variables self._built = True layer = self._layer else: layer = self._build_layer() Loading @@ -517,6 +515,9 @@ class Conv1D(Layer): if set_tensors: self._record_variable_scope(self.name) self.out_tensor = out_tensor if tfe.in_eager_mode() and not self._built: self._built = True self.variables = self._layer.variables return out_tensor Loading Loading @@ -871,7 +872,7 @@ class CombineMeanStd(Layer): mean_parent, std_parent = inputs[0], inputs[1] sample_noise = tf.random_normal( mean_parent.get_shape(), 0, self.noise_epsilon, dtype=tf.float32) if self.training_only: if self.training_only and 'training' in kwargs: sample_noise *= kwargs['training'] out_tensor = mean_parent + tf.exp(std_parent * 0.5) * sample_noise if set_tensors: Loading Loading @@ -2772,7 +2773,7 @@ class LSTMStep(Layer): self.W = init((self.input_dim, 4 * self.output_dim)) self.U = inner_init((self.output_dim, 4 * self.output_dim)) self.b = model_ops.create_variable( self.b = create_variable( np.hstack((np.zeros(self.output_dim), np.ones(self.output_dim), np.zeros(self.output_dim), np.zeros(self.output_dim))), dtype=tf.float32) Loading Loading @@ -3129,20 +3130,41 @@ class IterRefLSTMEmbedding(Layer): class BatchNorm(Layer): def __init__(self, in_layers=None, **kwargs): def __init__(self, in_layers=None, axis=-1, momentum=0.99, epsilon=1e-3, **kwargs): super(BatchNorm, self).__init__(in_layers, **kwargs) self.axis = axis self.momentum = momentum self.epsilon = epsilon try: parent_shape = self.in_layers[0].shape self._shape = tuple(self.in_layers[0].shape) except: pass def _build_layer(self): return tf.layers.BatchNormalization( axis=self.axis, momentum=self.momentum, epsilon=self.epsilon) def create_tensor(self, in_layers=None, set_tensors=True, **kwargs): inputs = self._get_input_tensors(in_layers) parent_tensor = inputs[0] out_tensor = tf.layers.batch_normalization(parent_tensor) if tfe.in_eager_mode(): if not self._built: self._layer = self._build_layer() layer = self._layer else: layer = self._build_layer() out_tensor = layer(parent_tensor) if set_tensors: self.out_tensor = out_tensor if tfe.in_eager_mode() and not self._built: self._built = True self.variables = self._layer.variables return out_tensor Loading @@ -3155,6 +3177,9 @@ class BatchNormalization(Layer): beta_init='zero', gamma_init='one', **kwargs): warnings.warn( 'BatchNormalization is deprecated and will be removed in a future release. Use BatchNorm instead.', DeprecationWarning) self.beta_init = initializations.get(beta_init) self.gamma_init = initializations.get(gamma_init) self.epsilon = epsilon Loading Loading @@ -3242,13 +3267,17 @@ class VinaFreeEnergy(Layer): self.stop = stop super(VinaFreeEnergy, self).__init__(**kwargs) def _build_layers(self): weighted_combo = WeightedLinearCombo() w = create_variable(tf.random_normal((1,), stddev=self.stddev)) return (weighted_combo, w) def cutoff(self, d, x): out_tensor = tf.where(d < 8, x, tf.zeros_like(x)) return out_tensor def nonlinearity(self, c): def nonlinearity(self, c, w): """Computes non-linearity used in Vina.""" w = tf.Variable(tf.random_normal((1,), stddev=self.stddev)) out_tensor = c / (1 + w * self.Nrot) return w, out_tensor Loading Loading @@ -3298,6 +3327,14 @@ class VinaFreeEnergy(Layer): X = inputs[0] Z = inputs[1] if tfe.in_eager_mode(): if not self._built: self._weighted_combo, self._w = self._build_layers() weighted_combo = self._weighted_combo w = self._w else: weighted_combo, w = self._build_layers() # TODO(rbharath): This layer shouldn't be neighbor-listing. Make # neighbors lists an argument instead of a part of this layer. nbr_list = NeighborList(self.N_atoms, self.M_nbrs, self.ndim, Loading @@ -3314,20 +3351,22 @@ class VinaFreeEnergy(Layer): gauss_2 = self.gaussian_second(dists) # Shape (N, M) weighted_combo = WeightedLinearCombo() interactions = weighted_combo(repulsion, hydrophobic, hbond, gauss_1, gauss_2) # Shape (N, M) thresholded = self.cutoff(dists, interactions) weight, free_energies = self.nonlinearity(thresholded) weight, free_energies = self.nonlinearity(thresholded, w) free_energy = ReduceSum()(free_energies) out_tensor = free_energy if set_tensors: self._record_variable_scope(self.name) self.out_tensor = out_tensor if tfe.in_eager_mode() and not self._built: self.variables = weighted_combo.variables + [w] self._built = True return out_tensor Loading @@ -3342,14 +3381,23 @@ class WeightedLinearCombo(Layer): except: pass def _create_variables(self, inputs): return [ create_variable(tf.random_normal([1], stddev=self.std)) for i in range(len(inputs)) ] def create_tensor(self, in_layers=None, set_tensors=True, **kwargs): inputs = self._get_input_tensors(in_layers, True) weights = [] if tfe.in_eager_mode(): if not self._built: self.variables = self._create_variables(inputs) self._built = True weights = self.variables else: weights = self._create_variables(inputs) out_tensor = None for in_tensor in inputs: w = tf.Variable(tf.random_normal([ 1, ], stddev=self.std)) for in_tensor, w in zip(inputs, weights): if out_tensor is None: out_tensor = w * in_tensor else: Loading Loading @@ -3659,7 +3707,8 @@ class Dropout(Layer): def create_tensor(self, in_layers=None, set_tensors=True, **kwargs): inputs = self._get_input_tensors(in_layers) parent_tensor = inputs[0] keep_prob = 1.0 - self.dropout_prob * kwargs['training'] training = kwargs['training'] if 'training' in kwargs else 1.0 keep_prob = 1.0 - self.dropout_prob * training out_tensor = tf.nn.dropout(parent_tensor, keep_prob) if set_tensors: self.out_tensor = out_tensor Loading Loading @@ -3763,11 +3812,18 @@ class AtomicConvolution(Layer): R = self.distance_matrix(D) sym = [] rsf_zeros = tf.zeros((B, N, M)) for param in self.radial_params: for i, param in enumerate(self.radial_params): if tfe.in_eager_mode(): if not self._built: self.variables += self._create_radial_variables(*param) param_variables = self.variables[3 * i:3 * i + 3] else: param_variables = self._create_radial_variables(*param) # We apply the radial pooling filter before atom type conv # to reduce computation param_variables, rsf = self.radial_symmetry_function(R, *param) rsf = self.radial_symmetry_function(R, *param_variables) if not self.atom_types: cond = tf.not_equal(Nbrs_Z, 0.0) Loading @@ -3784,8 +3840,17 @@ class AtomicConvolution(Layer): if set_tensors: self._record_variable_scope(self.name) self.out_tensor = out_tensor if tfe.in_eager_mode() and not self._built: self._built = True return out_tensor def _create_radial_variables(self, rc, rs, e): with tf.name_scope(None, "NbrRadialSymmetryFunction", [rc, rs, e]): rc = create_variable(rc) rs = create_variable(rs) e = create_variable(e) return (rc, rs, e) def radial_symmetry_function(self, R, rc, rs, e): """Calculates radial symmetry function. Loading @@ -3809,13 +3874,9 @@ class AtomicConvolution(Layer): """ with tf.name_scope(None, "NbrRadialSymmetryFunction", [rc, rs, e]): rc = tf.Variable(rc) rs = tf.Variable(rs) e = tf.Variable(e) K = self.gaussian_distance_matrix(R, rs, e) FC = self.radial_cutoff(R, rc) return [rc, rs, e], tf.multiply(K, FC) return tf.multiply(K, FC) def radial_cutoff(self, R, rc): """Calculates radial cutoff matrix. Loading Loading @@ -4021,7 +4082,17 @@ class AlphaShareLayer(Layer): subspaces = tf.reshape(tf.stack(subspaces), [n_alphas, -1]) # create the alpha learnable parameters alphas = tf.Variable(tf.random_normal([n_alphas, n_alphas]), name='alphas') if tfe.in_eager_mode(): if not self._built: self.variables = [ create_variable( tf.random_normal([n_alphas, n_alphas]), name='alphas') ] self._built = True alphas = self.variables[0] else: alphas = create_variable( tf.random_normal([n_alphas, n_alphas]), name='alphas') subspaces = tf.matmul(alphas, subspaces) Loading Loading @@ -4113,7 +4184,15 @@ class BetaShare(Layer): n_betas = len(inputs) subspaces = tf.reshape(tf.stack(subspaces), [n_betas, -1]) betas = tf.Variable(tf.random_normal([1, n_betas]), name='betas') if tfe.in_eager_mode(): if not self._built: self.variables = [ create_variable(tf.random_normal([1, n_betas]), name='betas') ] self._built = True betas = self.variables[0] else: betas = create_variable(tf.random_normal([1, n_betas]), name='betas') out_tensor = tf.matmul(betas, subspaces) self.betas = betas self.out_tensor = tf.reshape(out_tensor, [-1, original_cols]) Loading @@ -4134,7 +4213,7 @@ class ANIFeat(Layer): """ def __init__(self, in_layers, in_layers=None, max_atoms=23, radial_cutoff=4.6, angular_cutoff=3.1, Loading Loading @@ -4405,17 +4484,25 @@ class GraphEmbedPoolLayer(Layer): self.out_tensors = [result, result_A] return result, result_A def _create_variables(self, no_features, no_filters, name): W = create_variable( tf.truncated_normal( [no_features, no_filters], stddev=1.0 / math.sqrt(no_features)), name='%s_weights' % name, dtype=tf.float32) b = create_variable( tf.constant(0.1), name='%s_bias' % self.name, dtype=tf.float32) return [W, b] def embedding_factors(self, V, no_filters, name="default"): no_features = V.get_shape()[-1].value W = tf.get_variable( '%s_weights' % name, [no_features, no_filters], initializer=tf.truncated_normal_initializer( stddev=1.0 / math.sqrt(no_features)), dtype=tf.float32) b = tf.get_variable( '%s_bias' % self.name, [no_filters], initializer=tf.constant_initializer(0.1), dtype=tf.float32) if tfe.in_eager_mode(): if not self._built: self.variables = self._create_variables(no_features, no_filters, name) self._built = True W, b = self.variables else: W, b = self._create_variables(no_features, no_filters, name) V_reshape = tf.reshape(V, (-1, no_features)) s = tf.slice(tf.shape(V), [0], [len(V.get_shape()) - 1]) s = tf.concat([s, tf.stack([no_filters])], 0) Loading Loading @@ -4487,6 +4574,23 @@ class GraphCNN(Layer): self.num_filters = num_filters super(GraphCNN, self).__init__(**kwargs) def _create_variables(self, no_features, no_A): W = create_variable( 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 = create_variable( 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 = create_variable( tf.constant(0.1), name='%s_bias' % self.name, dtype=tf.float32) return [W, W_I, b] def create_tensor(self, in_layers=None, set_tensors=True, **kwargs): inputs = self._get_input_tensors(in_layers) if len(inputs) == 3: Loading @@ -4495,21 +4599,13 @@ class GraphCNN(Layer): V, A = inputs no_A = A.get_shape()[2].value no_features = V.get_shape()[2].value W = tf.get_variable( '%s_weights' % self.name, [no_features * no_A, self.num_filters], initializer=tf.truncated_normal_initializer( stddev=math.sqrt(1.0 / (no_features * (no_A + 1) * 1.0))), dtype=tf.float32) W_I = tf.get_variable( '%s_weights_I' % self.name, [no_features, self.num_filters], initializer=tf.truncated_normal_initializer( stddev=math.sqrt(1.0 / (no_features * (no_A + 1) * 1.0))), dtype=tf.float32) b = tf.get_variable( '%s_bias' % self.name, [self.num_filters], initializer=tf.constant_initializer(0.1), dtype=tf.float32) if tfe.in_eager_mode(): if not self._built: self.variables = self._create_variables(no_features, no_A) self._built = True W, W_I, b = self.variables else: W, W_I, b = self._create_variables(no_features, no_A) n = self.graphConvolution(V, A) A_shape = tf.shape(A) Loading
deepchem/models/tensorgraph/models/graph_models.py +3 −3 Original line number Diff line number Diff line Loading @@ -14,7 +14,7 @@ from deepchem.models.tensorgraph.graph_layers import WeaveGather, \ from deepchem.models.tensorgraph.graph_layers import WeaveLayerFactory from deepchem.models.tensorgraph.layers import Dense, SoftMax, \ SoftMaxCrossEntropy, GraphConv, BatchNorm, \ GraphPool, GraphGather, WeightedError, Dropout, BatchNormalization, Stack, Flatten, GraphCNN, GraphCNNPool GraphPool, GraphGather, WeightedError, Dropout, BatchNorm, Stack, Flatten, GraphCNN, GraphCNNPool from deepchem.models.tensorgraph.layers import L2Loss, Label, Weights, Feature from deepchem.models.tensorgraph.tensor_graph import TensorGraph from deepchem.trans import undo_transforms Loading Loading @@ -86,7 +86,7 @@ class WeaveTensorGraph(TensorGraph): out_channels=self.n_graph_feat, activation_fn=tf.nn.tanh, in_layers=weave_layer2A) batch_norm1 = BatchNormalization(epsilon=1e-5, mode=1, in_layers=[dense1]) batch_norm1 = BatchNorm(epsilon=1e-5, in_layers=[dense1]) weave_gather = WeaveGather( self.batch_size, n_input=self.n_graph_feat, Loading
deepchem/models/tensorgraph/models/text_cnn.py +1 −1 Original line number Diff line number Diff line Loading @@ -9,7 +9,7 @@ import copy from deepchem.metrics import to_one_hot, from_one_hot from deepchem.models.tensorgraph.layers import Dense, Concat, SoftMax, \ SoftMaxCrossEntropy, BatchNorm, WeightedError, Dropout, BatchNormalization, \ SoftMaxCrossEntropy, BatchNorm, WeightedError, Dropout, \ Conv1D, ReduceMax, Squeeze, Stack, Highway from deepchem.models.tensorgraph.graph_layers import DTNNEmbedding Loading
deepchem/models/tensorgraph/tests/test_layers_eager.py +249 −1 Original line number Diff line number Diff line Loading @@ -28,6 +28,7 @@ class TestLayersEager(test_util.TensorFlowTestCase): result = layer(input) self.assertEqual(result.shape[0], batch_size) self.assertEqual(result.shape[2], filters) assert len(layer.variables) == 2 # Creating a second layer should produce different results, since it has # different random weights. Loading Loading @@ -137,7 +138,7 @@ class TestLayersEager(test_util.TensorFlowTestCase): mean = np.random.rand(5, 3).astype(np.float32) std = np.random.rand(5, 3).astype(np.float32) layer = layers.CombineMeanStd(training_only=True, noise_epsilon=0.01) result1 = layer(mean, std) result1 = layer(mean, std, training=False) assert np.array_equal(result1, mean) # No noise in test mode result2 = layer(mean, std, training=True) assert not np.array_equal(result2, mean) Loading Loading @@ -730,3 +731,250 @@ class TestLayersEager(test_util.TensorFlowTestCase): assert test_out.shape == (n_test, n_feat) assert support_out.shape == (n_support, n_feat) assert len(layer.variables) == 8 def test_batch_norm(self): """Test invoking BatchNorm in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): batch_size = 10 n_features = 5 input = np.random.rand(batch_size, n_features).astype(np.float32) layer = layers.BatchNorm() result = layer(input) assert result.shape == (batch_size, n_features) assert len(layer.variables) == 4 def test_weighted_error(self): """Test invoking WeightedError in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): input1 = np.random.rand(5, 10).astype(np.float32) input2 = np.random.rand(5, 10).astype(np.float32) result = layers.WeightedError()(input1, input2) expected = np.sum(input1 * input2) assert np.allclose(result, expected) def test_vina_free_energy(self): """Test invoking VinaFreeEnergy in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): n_atoms = 5 m_nbrs = 1 ndim = 3 nbr_cutoff = 1 start = 0 stop = 4 X = np.random.rand(n_atoms, ndim).astype(np.float32) Z = np.random.randint(0, 2, (n_atoms)).astype(np.float32) layer = layers.VinaFreeEnergy(n_atoms, m_nbrs, ndim, nbr_cutoff, start, stop) result = layer(X, Z) assert len(layer.variables) == 6 assert result.shape == tuple() # Creating a second layer should produce different results, since it has # different random weights. layer2 = layers.VinaFreeEnergy(n_atoms, m_nbrs, ndim, nbr_cutoff, start, stop) result2 = layer2(X, Z) assert not np.allclose(result, result2) # But evaluating the first layer again should produce the same result as before. result3 = layer(X, Z) assert np.allclose(result, result3) def test_weighted_linear_combo(self): """Test invoking WeightedLinearCombo in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): input1 = np.random.rand(5, 10).astype(np.float32) input2 = np.random.rand(5, 10).astype(np.float32) layer = layers.WeightedLinearCombo() result = layer(input1, input2) assert len(layer.variables) == 2 expected = input1 * layer.variables[0] + input2 * layer.variables[1] assert np.allclose(result, expected) def test_neighbor_list(self): """Test invoking NeighborList in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): N_atoms = 5 start = 0 stop = 12 nbr_cutoff = 3 ndim = 3 M_nbrs = 2 coords = start + np.random.rand(N_atoms, ndim) * (stop - start) coords = tf.to_float(tf.stack(coords)) layer = layers.NeighborList(N_atoms, M_nbrs, ndim, nbr_cutoff, start, stop) result = layer(coords) assert result.shape == (N_atoms, M_nbrs) def test_dropout(self): """Test invoking Dropout in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): rate = 0.5 input = np.random.rand(5, 10).astype(np.float32) layer = layers.Dropout(rate) result1 = layer(input, training=False) assert np.allclose(result1, input) result2 = layer(input, training=True) assert not np.allclose(result2, input) nonzero = result2.numpy() != 0 assert np.allclose(result2.numpy()[nonzero], input[nonzero] / rate) def test_atomic_convolution(self): """Test invoking AtomicConvolution in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): batch_size = 4 max_atoms = 5 max_neighbors = 2 dimensions = 3 params = [[5.0, 2.0, 0.5], [10.0, 2.0, 0.5]] input1 = np.random.rand(batch_size, max_atoms, dimensions).astype( np.float32) input2 = np.random.randint( max_atoms, size=(batch_size, max_atoms, max_neighbors)) input3 = np.random.randint( 1, 10, size=(batch_size, max_atoms, max_neighbors)) layer = layers.AtomicConvolution(radial_params=params) result = layer(input1, input2, input3) assert result.shape == (batch_size, max_atoms, len(params)) assert len(layer.variables) == 3 * len(params) def test_alpha_share_layer(self): """Test invoking AlphaShareLayer in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): batch_size = 10 length = 6 input1 = np.random.rand(batch_size, length).astype(np.float32) input2 = np.random.rand(batch_size, length).astype(np.float32) layer = layers.AlphaShareLayer() result = layer(input1, input2) assert input1.shape == result[0].shape assert input2.shape == result[1].shape # Creating a second layer should produce different results, since it has # different random weights. layer2 = layers.AlphaShareLayer() result2 = layer2(input1, input2) assert not np.allclose(result[0], result2[0]) assert not np.allclose(result[1], result2[1]) # But evaluating the first layer again should produce the same result as before. result3 = layer(input1, input2) assert np.allclose(result[0], result3[0]) assert np.allclose(result[1], result3[1]) def test_sluice_loss(self): """Test invoking SluiceLoss in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): input1 = np.ones((3, 4)).astype(np.float32) input2 = np.ones((2, 2)).astype(np.float32) result = layers.SluiceLoss()(input1, input2) assert np.allclose(result, 40.0) def test_beta_share(self): """Test invoking BetaShare in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): batch_size = 10 length = 6 input1 = np.random.rand(batch_size, length).astype(np.float32) input2 = np.random.rand(batch_size, length).astype(np.float32) layer = layers.BetaShare() result = layer(input1, input2) assert input1.shape == result.shape assert input2.shape == result.shape # Creating a second layer should produce different results, since it has # different random weights. layer2 = layers.BetaShare() result2 = layer2(input1, input2) assert not np.allclose(result, result2) # But evaluating the first layer again should produce the same result as before. result3 = layer(input1, input2) assert np.allclose(result, result3) def test_ani_feat(self): """Test invoking ANIFeat in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): batch_size = 10 max_atoms = 5 input = np.random.rand(batch_size, max_atoms, 4).astype(np.float32) layer = layers.ANIFeat(max_atoms=max_atoms) result = layer(input) # TODO What should the output shape be? It's not documented, and there # are no other test cases for it. def test_graph_embed_pool_layer(self): """Test invoking GraphEmbedPoolLayer in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): V = np.random.uniform(size=(10, 100, 50)).astype(np.float32) adjs = np.random.uniform(size=(10, 100, 5, 100)).astype(np.float32) layer = layers.GraphEmbedPoolLayer(num_vertices=6) result = layer(V, adjs) assert result[0].shape == (10, 6, 50) assert result[1].shape == (10, 6, 5, 6) # Creating a second layer should produce different results, since it has # different random weights. layer2 = layers.GraphEmbedPoolLayer(num_vertices=6) result2 = layer2(V, adjs) assert not np.allclose(result[0], result2[0]) assert not np.allclose(result[1], result2[1]) # But evaluating the first layer again should produce the same result as before. result3 = layer(V, adjs) assert np.allclose(result[0], result3[0]) assert np.allclose(result[1], result3[1]) def test_graph_cnn(self): """Test invoking GraphCNN in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): V = np.random.uniform(size=(10, 100, 50)).astype(np.float32) adjs = np.random.uniform(size=(10, 100, 5, 100)).astype(np.float32) layer = layers.GraphCNN(num_filters=6) result = layer(V, adjs) assert result.shape == (10, 100, 6) # Creating a second layer should produce different results, since it has # different random weights. layer2 = layers.GraphCNN(num_filters=6) result2 = layer2(V, adjs) assert not np.allclose(result, result2) # But evaluating the first layer again should produce the same result as before. result3 = layer(V, adjs) assert np.allclose(result, result3) def test_hinge_loss(self): """Test invoking HingeLoss in eager mode.""" with context.eager_mode(): with tfe.IsolateTest(): n_labels = 1 n_logits = 1 logits = np.random.rand(n_logits).astype(np.float32) labels = np.random.rand(n_labels).astype(np.float32) result = layers.Hingeloss()(labels, logits) assert result.shape == (n_labels,)
