reader supports multiple inputs

LanguageBindings/Python/cntk/context.py

@@ -119,7 +119,7 @@ def add_macro(self, path):
         '''
         self.macros.append(path)
 
-    def _generate_train_config(self, input_map):
+    def _generate_train_config(self, reader):
         '''
         Generates the configuration file for the train action.
         '''
@@ -130,7 +130,7 @@ def _generate_train_config(self, input_map):
             'DevideId': self.device_id,
             'ModelDescription': self.to_description(),
             'ModelPath': model_filename,
-            'Reader': self._generate_reader_config(input_map),
+            'Reader': reader_config,
             'SGD': self.optimizer.generate_config(),
         }
         return tmpl % tmpl_dict
@@ -283,7 +283,7 @@ def _call_cntk(self, config_file_name, config_content):
         subprocess.check_call(
             [CNTK_EXECUTABLE_PATH, "configFile=%s" % filename])
 
-    def train(self, input_map):
+    def train(self, reader):
         '''
         Run the train action locally.
         :param input_map: mapping of input node to (reader, (start_dim, num_dim))

LanguageBindings/Python/cntk/reader.py

@@ -23,71 +23,82 @@ def __ne__(self, x): return x is not self
 class UCIFastReader(AbstractReader):
 
     """This is the reader class
-
     :param filename: data file path
-    :param features_dim: number of label columns
-    :param features_start: the index of the first label column    
+    :param label_node_name: the name of the labels node in the network
     :param labels_dim: number of label columns
     :param labels_start: the index of the first label column
     :param num_of_classes: the number of classes
     :param label_mapping_file: 
         the mapping file path, it can be simply with all the possible classes, one per line
-    :param custom_delimiter: the default is space and tab, you can specify other delimiters to be used
+    :param custom_delimiter: the default is space and tab, you can specify other delimiters to be used        
+    :param inputs_def: is a list of tuples (input_name, input_start, input_dim)
+        input_name: the name of the input in the network definition
+        input_start: the start column   
+        input_dim: the number of columns
     """
 
     def __init__(self, filename,
-                 features_dim=None,
-                 features_start=None,
+                 label_node_name="labels",
                  labels_dim=None,
                  labels_start=None,
-                 num_of_classes=None,
-                 label_mapping_file=None,
-                 custom_delimiter=None):
+                 num_of_classes=None,                 
+                 label_mapping_file=None,                 
+                 custom_delimiter=None,
+                 inputs_def=None):
         """ Reader constructor    
         """
         self["ReaderType"] = self.__class__.__name__
         self["FileName"] = filename
         self["LabelsDim"] = labels_dim
-        self["LabelsStart"] = labels_start
-        self["FeaturesDim"] = features_dim
-        self["FeaturesStart"] = features_start
+        self["LabelsStart"] = labels_start            
         self["NumOfClasses"] = num_of_classes
         self["LabelMappingFile"] = label_mapping_file
         self["CustomDelimiter"] = custom_delimiter
+        self.inputs_def = inputs_def or []
 
+    def add_input(self, input_name, input_start, input_dim):
+        """Add an input to the reader
+        :param input_name: the name of the input in the network definition
+        :param input_start: the start column   
+        :param input_dim: the number of columns
+        """
+        if (not (input_name and input_start and input_dim)):
+            raise ValueError("one of the parameters of add_input is None or empty string") 
+
+        self.inputs_def.append((input_name, input_start, input_dim))
+
     def generate_config(self):
         """Generate the reader configuration block
         """
         template = '''
-        		readerType = "%(ReaderType)s"
-        		file = "%(FileName)s"
-        		randomize = "none"
-        		verbosity = 1          
-               '''
+            readerType = "%(ReaderType)s"
+            file = "%(FileName)s"
+            randomize = "none"
+            verbosity = 1          
+            '''
 
         if self['CustomDelimiter'] is not None:
             template += '''
-               customDelimiter=%(CustomDelimiter)s
-               '''
-        # TODO: generalize the reader to take n features sequences and m label
-        # sequences
-        if self['FeaturesStart'] is not None:
-            template += '''
-
-        		v2=[
-        			start = "%(FeaturesStart)s"
-        			dim = "%(FeaturesDim)s"		          
-        		]'''
-
+      customDelimiter=%(CustomDelimiter)s
+       '''
+
         if self['LabelsStart'] is not None:
             template += '''
+            labels=[
+                start = "%(LabelsStart)s"
+                dim = "%(LabelsDim)s"
+                labelDim="%(NumOfClasses)s"        			
+                labelMappingFile="%(LabelMappingFile)s" 
+            ]'''
+
+        if self.inputs_def is not None:
+            for (name, start, dim) in self.inputs_def:                
+                template += '''
+            {0}=[
+                start = {1}
+                dim = {2}		          
+            ]'''.format(name, start, dim)
 
-        		v0=[
-        			start = "%(LabelsStart)s"
-        			dim = "%(LabelsDim)s"		          
-                    labelDim="%(NumOfClasses)s"        			
-                    labelMappingFile="%(LabelMappingFile)s" 
-        		]'''
 
         return template % self