add config generation for parallel sgd

contrib/Python/cntk/sgd.py

@@ -265,10 +265,10 @@ class SGDParams:
         self.parallel_training_ = None
         
     def _set_global_parallel_params(self, 
-                                    parallalization_method = 'none', 
-                                    parallelization_start_epoch = 0,
-                                    distributed_mb_reading = False,
-                                    sync_perf_stats = 0):
+                                    parallalization_method = None, 
+                                    parallelization_start_epoch = None,
+                                    distributed_mb_reading = None,
+                                    sync_perf_stats = None):
         self.parallel_training = {
             'parallelizationMethod':parallalization_method,
             'parallelizationStartEpoch':parallelization_start_epoch,
@@ -276,12 +276,12 @@ class SGDParams:
             'syncPerfStats':sync_perf_stats}
             
     def set_parallel_to_data_parallel(self, 
-                                      parallelization_start_epoch = 0,
-                                      distributed_mb_reading = False,
-                                      sync_perf_stats = 0,
-                                      gradient_bits = 8,
-                                      use_zero_threshold_for_1bit = True,
-                                      use_buffered_async_gradient_aggregation = False):
+                                      parallelization_start_epoch = None,
+                                      distributed_mb_reading = None,
+                                      sync_perf_stats = None,
+                                      gradient_bits = None,
+                                      use_zero_threshold_for_1bit = None,
+                                      use_buffered_async_gradient_aggregation = None):
 
         self._set_global_parallel_params('DataParallelSGD',
                                          parallelization_start_epoch,
@@ -294,10 +294,10 @@ class SGDParams:
             'useBufferedAsyncGradientAggregation':use_buffered_async_gradient_aggregation}
             
     def set_parallel_to_model_average(self, 
-                                      parallelization_start_epoch = 0,
-                                      distributed_mb_reading = False,
-                                      sync_perf_stats = 0,
-                                      sync_period = 40000,
+                                      parallelization_start_epoch = None,
+                                      distributed_mb_reading = None,
+                                      sync_perf_stats = None,
+                                      sync_period = None,
                                       sync_frequency_in_frames = None):
         
         self._set_global_parallel_params('ModelAveragingSGD',
@@ -310,13 +310,13 @@ class SGDParams:
             'syncFrequencyInFrames':sync_frequency_in_frames}
                                               
     def set_parallel_to_block_momentum(self, 
-                                      parallelization_start_epoch = 0,
-                                      distributed_mb_reading = False,
-                                      sync_perf_stats = 0,
-                                      sync_period = 120000,
-                                      reset_sgd_momentum = True,
-                                      use_nesterov_momentum = True,
-                                      block_learning_rate = 1.0,
+                                      parallelization_start_epoch = None,
+                                      distributed_mb_reading = None,
+                                      sync_perf_stats = None,
+                                      sync_period = None,
+                                      reset_sgd_momentum = None,
+                                      use_nesterov_momentum = None,
+                                      block_learning_rate = None,
                                       block_momentum_per_sync = None,
                                       block_momentum_as_time_constant = None):
         
@@ -333,6 +333,20 @@ class SGDParams:
             'blockMomentumPerSync':block_momentum_per_sync,
             'blockMomentumAsTimeConstant':block_momentum_as_time_constant}
 
+
+    def _generate_parallel_training_config(self):
+        config = ['ParallelTrain=[']        
+        for k,v in self.parallel_training.items():
+            if v:
+                config.append('\t{0} = {1}'.format(k, v))    
+        
+        config.append('\t{0} = ['.format(self.parallel_training['parallelizationMethod']))    
+        for k,v in self.parallel_training_subblock.items():            
+            if v:
+                config.append('\t\t{0} = {1}'.format(k, v))    
+        config.append['\t]']                
+        config.append[']']
+        
     def _to_config_description(self):
         """Generate the SGDParams configuration block
         """