add splice node to the python api

contrib/Python/cntk/ops/__init__.py

@@ -981,15 +981,52 @@ def slice(x, begin_index, end_index, axis=0, name=None):
     else:
         cntk_axis = abs(axis) if axis<0 else x.rank - axis
     op = Slice(x, begin_index, end_index, cntk_axis, name=name)
+    wrap_numpy_arrays(op)            
     op.rank = op._.rank
     return op
+    
+def splice(inputs, begin_index, end_index, axis=0, name=None): 
+    '''
+    Concatenate the input tensors along an axis.    
+
+    Examples:
+        >>> # create 2x2 matrix in a sequence of length 1 in a batch of one sample
+        >>> data1 = np.asarray([[[1, 2],
+        ...                     [4, 5]]])
+        >>> x = C.input_numpy(data1)
+        >>> # create 3x2 matrix in a sequence of length 1 in a batch of one sample
+        >>> data2 = np.asarray([[[10, 20],
+        ...                     [30, 40],
+        ...                     [50, 60]]])
+        >>> y = C.input_numpy(data2)
+        >>> # splice both inputs on axis=0 returns a 5x2 matrix
+        >>> C.eval(C.splice([x,y], 0))
+        [array([[[1, 2],
+                 [4, 5],
+                 [10, 20],
+                 [30, 40],
+                 [50, 60]]])]        
+
+    Args:
+        inputs (list): list of input tensors
+        axis (int): axis along which the concatenation will be performed
+
+    Returns:
+        :class:`cntk.graph.ComputationNode`
+    '''
+    from cntk.ops.cntk2 import Splice
+    #cntk uses column major, thus it will read the indices of data passed from 
+    # python in reverse
+    cntk_axis = abs(axis) if axis<0 else inputs[0].rank - axis
+    op = Splice(inputs, cntk_axis, name=name)
+    wrap_numpy_arrays(op)            
+    op.rank = op._[0].rank
+    return op
 
 ################################################################################
 # training ops
 ################################################################################
-
-# unittests might require training and testing at the same time ? which 
-# sounds more like end2end test ?
+ 
 def dropout(x, name=None):
     """
     Compute a new tensor with `dropoutRate` perecent set to zero. The values 
@@ -998,10 +1035,6 @@ def dropout(x, name=None):
 
     The output tensor has the same shape as `x`, but with `dropoutRate` of the
     elements set to zero (droped out).
-    
-    
-    Examples:
-        TBA
             
     Args:        
         x: source tensor

contrib/Python/cntk/ops/cntk2.py

@@ -19,6 +19,15 @@ class Slice(ComputationNode):
         self.inputs = ['_']
         self.params_with_defaults = ['axis']
 
+class Splice(ComputationNode):
+    def __init__(self, _, axis=1, op_name='CNTK2.Splice',
+            name=None):
+        super(Splice, self).__init__(params=['_', 'axis'], op_name=op_name, name=name)
+        self._ = _
+        self.axis = axis
+        self.inputs = ['_']
+        self.params_with_defaults = ['axis']
+
 class Ceil(ComputationNode):
     def __init__(self, _, op_name='CNTK2.Ceil', name=None):
         super(Ceil, self).__init__(params=['_'], op_name=op_name, name=name)

contrib/Python/cntk/utils/__init__.py

@@ -261,6 +261,7 @@ def wrap_numpy_arrays(node):
         for p in node.params:
             if p in node.inputs:
                 val = getattr(node, p)
+                #TODO: add support to list of numpy arrays, e.g. Splice()
                 if not (isinstance(val, ComputationNode) or isinstance(val, str)):
                     # One param needs to be an Input() node. This will be fixed in 
                     # CNTK soon, so that we can remove this workaround and evaluate a 

contrib/Python/cntk/utils/_fetch_ops.py

@@ -297,6 +297,15 @@ class Slice(ComputationNode):
         self.inputs = ['_']
         self.params_with_defaults = ['axis']
 
+class Splice(ComputationNode):
+    def __init__(self, _, axis=1, op_name='CNTK2.Splice',
+            name=None):
+        super(Splice, self).__init__(params=['_', 'axis'], op_name=op_name, name=name)
+        self._ = _
+        self.axis = axis
+        self.inputs = ['_']
+        self.params_with_defaults = ['axis']
+
 class Ceil(ComputationNode):
     def __init__(self, _, op_name='CNTK2.Ceil', name=None):
         super(Ceil, self).__init__(params=['_'], op_name=op_name, name=name)