Fix custom loss for MXNet backend. Fix bug in Concat layer (#110)

* Fix custom loss usage in MXNet backend. Issue - 25

* Fix CR comments

* Fix CR comments

keras/backend/mxnet_backend.py

@@ -378,6 +378,7 @@ def placeholder(shape=None, ndim=None, dtype=None, sparse=False, name=None):
         shape = tuple([0 if dim is None else dim for dim in shape])
     else:
         shape = tuple([0 for _ in range(ndim)])
+
     sym = _keras_variable(name, shape=shape, dtype=dtype)
     sym._keras_shape = tuple([d if d != 0 else None for d in shape])
     sym._mxnet_placeholder = True
@@ -1949,9 +1950,14 @@ def concatenate(tensors, axis=-1):
         A tensor.
     """
     if axis < 0:
-        axis += ndim(tensors[0])
+        rank = ndim(tensors[0])
+        if rank:
+            axis %= rank
+        else:
+            axis = 0
+
     tensors = [t.symbol for t in tensors]
-    return KerasSymbol(mx.sym.Concat(*tensors, dim=axis))
+    return KerasSymbol(mx.sym.concat(*tensors, dim=axis))
 
 
 @keras_mxnet_symbol
@@ -2809,7 +2815,7 @@ def softmax(x):
     # Returns
         A tensor.
     """
-    return KerasSymbol(mx.sym.SoftmaxActivation(data=x.symbol))
+    return KerasSymbol(mx.sym.softmax(data=x.symbol))
 
 
 @keras_mxnet_symbol

keras/engine/training_utils.py

@@ -417,6 +417,11 @@ def weighted(y_true, y_pred, weights, mask=None):
             weight_ndim = K.ndim(weights)
             score_array = K.mean(score_array,
                                  axis=list(range(weight_ndim, ndim)))
+            # If sample_weights shape is like (100, ), we convert it to (100, 1).

+            # Because, MXNet treats the shape (100, ) as (100) leading to broadcast operator
+            # 
failures in below operations.

+            if K.backend() == 'mxnet' and weight_ndim == 1:
+                weights = K.reshape(weights, shape=(weights.shape[0], 1))
             score_array *= weights
             score_array /= K.mean(K.cast(K.not_equal(weights, 0), K.floatx()))
         return K.mean(score_array)

tests/keras/layers/normalization_test.py

@@ -82,7 +82,7 @@ def test_batchnorm_correctness_2d():
 
 
 @pytest.mark.skipif((K.backend() == 'mxnet'),
-                    reason='MXNet backend does not allow predict() before compile()')
+                    reason='MXNet backend uses native BatchNorm operator. Do not do updates in the model.')
 @keras_test
 def test_batchnorm_training_argument():
     bn1 = normalization.BatchNormalization(input_shape=(10,))
@@ -160,7 +160,7 @@ def test_batchnorm_convnet_no_center_no_scale():
 
 
 @pytest.mark.skipif((K.backend() == 'mxnet'),
-                    reason='MXNet backend uses native BatchNorm operator. Do do updates in the model.')
+                    reason='MXNet backend uses native BatchNorm operator. Do not do updates in the model.')
 @keras_test
 def test_shared_batchnorm():
     '''Test that a BN layer can be shared

tests/keras/losses_test.py

@@ -139,9 +139,6 @@ def test_serializing_loss_class():
     assert deserialized.mse_fraction == 0.3
 
 
-# https://github.com/deep-learning-tools/keras/issues/25
-@pytest.mark.skipif(K.backend() == 'mxnet',
-                    reason='MXNet backend does not fully support custom loss yet.')
 def test_serializing_model_with_loss_class(tmpdir):
     model_filename = str(tmpdir / 'custom_loss.hdf')
 

tests/keras/utils/vis_utils_test.py

@@ -9,11 +9,8 @@
 from keras.layers import TimeDistributed
 from keras.models import Sequential
 from keras.utils import vis_utils
-from keras import backend as K
 
 
-@pytest.mark.skipif(K.backend() == 'mxnet',
-                    reason='MXNet backend does not support LSTM yet')
 def test_plot_model():
     model = Sequential()
     model.add(Conv2D(filters=2, kernel_size=(2, 3), input_shape=(3, 5, 5), name='conv'))

tests/test_loss_weighting.py

@@ -121,8 +121,6 @@ def test_sequential_sample_weights():
     assert(score < standard_score_sequential)
 
 
-@pytest.mark.skipif(K.backend() == 'mxnet',
-                    reason='MXNet backend does not support GRU yet.')
 @keras_test
 def test_sequential_temporal_sample_weights():
     (x_train, y_train), (x_test, y_test), (sample_weight, class_weight, test_ids) = _get_test_data()