Integrate ebarsoum/BSvsPython into master

bindings/python/cntk/cntk_py.i

@@ -22,6 +22,9 @@
 
 %rename(momentum_as_time_constant_schedule) CNTK::MomentumAsTimeConstantSchedule;
 
+%rename(l1_regularization_weight) CNTK::AdditionalLearningOptions::l1RegularizationWeight;
+%rename(l2_regularization_weight) CNTK::AdditionalLearningOptions::l2RegularizationWeight;
+
 // if we don't except RandomUniform the corresponding template functions will not be generated
 %rename("%(utitle)s", %$isfunction, notregexmatch$name="RandomUniform") "";
 %rename("%(utitle)s", %$isvariable) "";

bindings/python/cntk/tests/distributed_test.py

@@ -42,7 +42,7 @@ def run_distributed_trainer(tmpdir, quantized):
     momentum_time_constant = momentum_as_time_constant_schedule(1100)
 
     trainer = Trainer(z, ce, errs, \
-            sgd(z.parameters, 0.007, momentum_time_constant, 0.5, True),
+            momentum_sgd(z.parameters, 0.007, momentum_time_constant),
             distributed_trainer=dist_trainer)
     in1_value = [[1],[2]]
     label_value = [[0], [1]]

bindings/python/cntk/tests/trainer_test.py

@@ -23,7 +23,7 @@ def test_trainer(tmpdir):
     m_schedule = momentum_schedule(1100)
 
     trainer = Trainer(z, ce, errs, \
-            [sgd(z.parameters, 0.007, m_schedule, 0.5, True)])
+            [momentum_sgd(z.parameters, 0.007, m_schedule)])
     in1_value = [[1],[2]]
     label_value = [[0], [1]]
     arguments = {in1: in1_value, labels: label_value}
@@ -52,7 +52,7 @@ def test_output_to_retain():
     m_schedule = momentum_schedule(1100)
 
     trainer = Trainer(z, ce, errs, \
-            [sgd(z.parameters, 0.007, m_schedule, 0.5, True)])
+            [momentum_sgd(z.parameters, 0.007, m_schedule)])
     in1_value = [[1],[2]]
     label_value = [[0], [1]]
     arguments = {in1: in1_value, labels: label_value}

bindings/python/examples/Sequence2Sequence/Sequence2Sequence.py

@@ -1,4 +1,4 @@
-# Copyright (c) Microsoft. All rights reserved.
+# Copyright (c) Microsoft. All rights reserved.
 
 # Licensed under the MIT license. See LICENSE.md file in the project root
 # for full license information.
@@ -157,7 +157,10 @@ def sequence_to_sequence_translator(debug_output=False, run_test=False):
     momentum_time_constant = momentum_as_time_constant_schedule(1100)
     clipping_threshold_per_sample = 2.3
     gradient_clipping_with_truncation = True
-    learner = momentum_sgd(z.parameters, lr_per_sample, momentum_time_constant, clipping_threshold_per_sample, gradient_clipping_with_truncation)
+    learner = momentum_sgd(z.parameters, 
+                           lr_per_sample, momentum_time_constant, 
+                           gradient_clipping_threshold_per_sample=clipping_threshold_per_sample, 
+                           gradient_clipping_with_truncation=gradient_clipping_with_truncation)
     trainer = Trainer(z, ce, errs, learner)
 
     # setup data

bindings/python/examples/test/cifar_resnet_test.py

@@ -44,7 +44,7 @@ def test_cifar_resnet_error(device_id):
     reader_test  = create_reader(os.path.join(base_path, 'test_map.txt'), os.path.join(base_path, 'CIFAR-10_mean.xml'), False)
 
     test_error = train_and_evaluate(reader_train, reader_test, max_epochs=5)
-    expected_test_error = 0.384
+    expected_test_error = 0.36
 
     assert np.allclose(test_error, expected_test_error,
                        atol=TOLERANCE_ABSOLUTE)