Refactor arg checking and processing.

theano/tensor/raw_random.py

@@ -301,14 +301,12 @@ def _infer_ndim_bcast(ndim, shape, *args):
     else:
         args_ndim = 0
 
-    # there is a convention that -1 means the corresponding shape of a
-    # potentially-broadcasted symbolic arg
-    if (isinstance(shape, (tuple, list))
-            and numpy.all(numpy.asarray(shape) >= 0)):
-        bcast = [(s == 1) for s in shape]
-        v_shape = tensor.TensorConstant(type=tensor.lvector,
-                                        data=theano._asarray(shape,
-                                                             dtype='int64'))
+    if isinstance(shape, (tuple, list)):
+        # there is a convention that -1 means the corresponding shape of a
+        # potentially-broadcasted symbolic arg
+        #
+        # This case combines together symbolic and non-symbolic shape
+        # information
         shape_ndim = len(shape)
         if ndim is None:
             ndim = shape_ndim
@@ -317,18 +315,7 @@ def _infer_ndim_bcast(ndim, shape, *args):
                 raise ValueError('ndim should be equal to len(shape), but\n',
                             'ndim = %s, len(shape) = %s, shape = %s'
                             % (ndim, shape_ndim, shape))
-    elif isinstance(shape, (tuple, list)):
-        # there is a convention that -1 means the corresponding shape of a
-        # potentially-broadcasted symbolic arg
-        #
-        # This case combines together symbolic and non-symbolic shape
-        # information
-        if ndim is None:
-            ndim = args_ndim
-        else:
-            ndim = max(args_ndim, ndim)
-        ndim = max(args_ndim, len(shape))
-        shape = [-1] * (ndim - len(shape)) + list(shape)
+
         bcast = []
         pre_v_shape = []
         for i, s in enumerate(shape):