Merge pull request deepchem#1118 from abster12/master

Adding Hinge Loss Layer

deepchem/models/tensorgraph/layers.py

@@ -1542,6 +1542,12 @@ def create_tensor(self, in_layers=None, set_tensors=True, **kwargs):
 
 
 class SparseSoftMaxCrossEntropy(Layer):
+  """Computes Sparse softmax cross entropy between logits and labels.
+  labels: Tensor of shape [d_0,d_1,....,d_{r-1}](where r is rank of logits) and must be of dtype int32 or int64.
+  logits: Unscaled log probabilities of shape [d_0,....d{r-1},num_classes] and of dtype float32 or float64.
+  Note: the rank of the logits should be 1 greater than that of labels.
+  The output will be a tensor of same shape as labels and of same type as logits with the loss.
+  """
 
   def __init__(self, in_layers=None, **kwargs):
     super(SparseSoftMaxCrossEntropy, self).__init__(in_layers, **kwargs)
@@ -4309,3 +4315,30 @@ def batch_mat_mult(self, A, B):
     result = tf.matmul(A_reshape, B)
     result = tf.reshape(result, tf.stack([A_shape[0], A_shape[1], axis_2]))
     return result
+
+
+class Hingeloss(Layer):
+  """This layer computes the hinge loss on inputs:[labels,logits] 
+  labels: The values of this tensor is expected to be 1.0 or 0.0. The shape should be the same as logits.
+  logits: Holds the log probabilities for labels, a float tensor.
+  The output is a weighted loss tensor of same shape as labels.
+  """
+
+  def __init__(self, in_layers=None, **kwargs):
+    super(Hingeloss, self).__init__(in_layers, **kwargs)
+    try:
+      self._shape = self.in_layers[1].shape
+    except:
+      pass
+
+  def create_tensor(self, in_layers=None, set_tensors=True, **kwargs):
+    inputs = self._get_input_tensors(in_layers)
+    if len(inputs) != 2:
+      raise ValueError()
+    labels, logits = inputs[0], inputs[1]
+    reduction = tf.losses.Reduction
+    out_tensor = tf.losses.hinge_loss(
+        labels=labels, logits=logits, reduction=reduction.NONE)
+    if set_tensors:
+      self.out_tensor = out_tensor
+    return out_tensor

deepchem/models/tensorgraph/tests/test_layers.py

@@ -25,6 +25,7 @@
 from deepchem.models.tensorgraph.layers import Gather
 from deepchem.models.tensorgraph.layers import GraphConv
 from deepchem.models.tensorgraph.layers import GraphGather
+from deepchem.models.tensorgraph.layers import Hingeloss
 from deepchem.models.tensorgraph.layers import Input
 from deepchem.models.tensorgraph.layers import InputFifoQueue
 from deepchem.models.tensorgraph.layers import InteratomicL2Distances
@@ -45,6 +46,7 @@
 from deepchem.models.tensorgraph.layers import SigmoidCrossEntropy
 from deepchem.models.tensorgraph.layers import SoftMax
 from deepchem.models.tensorgraph.layers import SoftMaxCrossEntropy
+from deepchem.models.tensorgraph.layers import SparseSoftMaxCrossEntropy
 from deepchem.models.tensorgraph.layers import StopGradient
 from deepchem.models.tensorgraph.layers import TensorWrapper
 from deepchem.models.tensorgraph.layers import TimeSeriesDense
@@ -381,6 +383,18 @@ def test_softmax_cross_entropy(self):
       out_tensor = out_tensor.eval()
       assert out_tensor.shape == (batch_size,)
 
+  def test_sparse_softmax_cross_entropy(self):
+    batch_size = 10
+    n_features = 5
+    logit_tensor = np.random.rand(batch_size, n_features)
+    label_tensor = np.random.rand(batch_size)
+    with self.test_session() as sess:
+      logit_tensor = tf.convert_to_tensor(logit_tensor, dtype=tf.float32)
+      label_tensor = tf.convert_to_tensor(label_tensor, dtype=tf.int32)
+      out_tensor = SparseSoftMaxCrossEntropy()(label_tensor, logit_tensor)
+      out_tensor = out_tensor.eval()
+      assert out_tensor.shape == (batch_size,)
+
   def test_reduce_mean(self):
     """Test that ReduceMean can be invoked."""
     batch_size = 10
@@ -875,3 +889,16 @@ def test_IRV(self):
       assert out_tensor.shape == (batch_size, n_tasks)
       irv_reg = IRVRegularize(irv_layer, 1.)()
       assert irv_reg.eval() >= 0
+
+  def test_hingeloss(self):
+
+    labels = 1
+    logits = 1
+    logits_tensor = np.random.rand(logits)
+    labels_tensor = np.random.rand(labels)
+    with self.test_session() as sess:
+      logits_tensor = tf.convert_to_tensor(logits_tensor, dtype=tf.float32)
+      labels_tensor = tf.convert_to_tensor(labels_tensor, dtype=tf.float32)
+      out_tensor = Hingeloss()(labels_tensor, logits_tensor)
+      out_tensor = out_tensor.eval()
+      assert out_tensor.shape == (labels,)

deepchem/models/tensorgraph/tests/test_layers_pickle.py

@@ -10,7 +10,7 @@
   SoftMaxCrossEntropy, ReduceMean, ToFloat, ReduceSquareDifference, Conv2D, MaxPool2D, ReduceSum, GraphConv, GraphPool, \
   GraphGather, BatchNorm, WeightedError, ReLU, \
   Conv3D, MaxPool3D, Conv2DTranspose, Conv3DTranspose, \
-  LSTMStep, AttnLSTMEmbedding, IterRefLSTMEmbedding, GraphEmbedPoolLayer, GraphCNN, Cast
+  LSTMStep, AttnLSTMEmbedding, IterRefLSTMEmbedding, GraphEmbedPoolLayer, GraphCNN, Cast,Hingeloss,SparseSoftMaxCrossEntropy
 from deepchem.models.tensorgraph.symmetry_functions import AtomicDifferentiatedDense
 from deepchem.models.tensorgraph.IRV import IRVLayer, IRVRegularize, Slice
 
@@ -269,6 +269,17 @@ def test_SoftmaxCrossEntropy_pickle():
   tg.save()
 
 
+def test_SparseSoftmaxCrossEntropy_pickle():
+  tg = TensorGraph()
+  logits = Feature(shape=(tg.batch_size, 5))
+  labels = Feature(shape=(tg.batch_size,), dtype=tf.int32)
+  layer = SparseSoftMaxCrossEntropy(in_layers=[labels, logits])
+  tg.add_output(layer)
+  tg.set_loss(layer)
+  tg.build()
+  tg.save()
+
+
 def test_SigmoidCrossEntropy_pickle():
   tg = TensorGraph()
   feature = Feature(shape=(tg.batch_size, 1))
@@ -682,3 +693,13 @@ def test_Slice_pickle():
   tg.set_loss(out)
   tg.build()
   tg.save()
+
+
+def test_hingeloss_pickle():
+  tg = TensorGraph()
+  feature = Feature(shape=(1, None))
+  layer = Hingeloss(in_layers=[feature, feature])
+  tg.add_output(layer)
+  tg.set_loss(layer)
+  tg.build()
+  tg.save()