Fix bug reported by Olivier introduced by my last push. I was calling

tensor.as_tensor_variable on inputs of x, but I should do that only if they where numpy.ndarrays (maybe other types) .. because they could be random states case in which tensor.as_tensor_variable fails.

Fix bug reported by Olivier introduced by my last push. I was calling
2ba9c67b · Razvan Pascanu · 8872aff8 · 2ba9c67b
--- a/theano/scan_module/scan_utils.py
+++ b/theano/scan_module/scan_utils.py
@@ -49,6 +49,7 @@ def info(*msg):
 def safe_new(x):
+    if isinstance(x, numpy.ndarray):
        x = tensor.as_tensor_variable(x)
    if cuda.cuda_available and isinstance(x.type, cuda.CudaNdarrayType):
        return tensor.TensorType(
@@ -58,6 +59,7 @@ def safe_new(x):
        return x.type()
 def safe_to_cpu(x):
+    if isinstance(x, numpy.ndarray):
        x = tensor.as_tensor_variable(x)
    if cuda.cuda_available and isinstance(x.type, cuda.CudaNdarrayType):
        return cuda.basic_ops.host_from_gpu(x)