Unpickle cudaNdarray as numpy.ndarray, if cuda is not enabled.

theano/sandbox/cuda/tests/CudaNdarray.pkl

+ctheano.sandbox.cuda.type
+CudaNdarray_unpickler
+p1
+(cnumpy.core.multiarray
+_reconstruct
+p2
+(cnumpy
+ndarray
+p3
+(I0
+tS'b'
+tRp4
+(I1
+(I1
+tcnumpy
+dtype
+p5
+(S'f4'
+I0
+I1
+tRp6
+(I3
+S'<'
+NNNI-1
+I-1
+I0
+tbI00
+S'\x00\x00(\xc2'
+tbtR.
\ No newline at end of file

theano/sandbox/cuda/tests/test_var.py

 import unittest
+import cPickle
 import numpy
 from numpy.testing.decorators import skipif
+import os.path
 
 import theano
 from theano import tensor
@@ -10,6 +12,9 @@ from theano.sandbox.cuda import CudaNdarrayType, cuda_available
 import theano.sandbox.cuda as cuda
 
 
+if cuda_available:
+    from theano.sandbox.cuda import CudaNdarray
+
 @skipif(not cuda_available, msg='Optional package cuda disabled')
 def test_float32_shared_constructor():
     npy_row = numpy.zeros((1, 10), dtype='float32')
@@ -121,3 +126,20 @@ class T_updates(unittest.TestCase):
         output_func = theano.function(inputs=[], outputs=[],
                                       updates=[(output_var, up)])
         output_func()
+
+
+def test_unpickle_cudandarray_as_numpy_ndarray():
+    # testfile created on cuda enabled machine using
+    # >>> with open('CudaNdarray.pkl', 'wb') as fp:
+    # >>> cPickle.dump(theano.sandbox.cuda.CudaNdarray(np.array([-42.0], dtype=np.float32)), fp)
+
+    testfile_dir = os.path.dirname(os.path.realpath(__file__))
+    with open(os.path.join(testfile_dir, 'CudaNdarray.pkl')) as fp:
+        mat = cPickle.load(fp)
+
+        if cuda_available:
+            assert isinstance(mat, CudaNdarray)
+        else:
+            assert isinstance(mat, numpy.ndarray)
+
+        assert mat[0] == -42.0
\ No newline at end of file

theano/sandbox/cuda/type.py

@@ -2,6 +2,7 @@
 """
 import os
 import copy_reg
+import warnings
 
 import numpy
 
@@ -487,7 +488,13 @@ theano.compile.register_deep_copy_op_c_code(
 # equal the pickled version, and the cmodule cache is not happy with
 # the situation.
 def CudaNdarray_unpickler(npa):
-    return cuda.CudaNdarray(npa)
+    if cuda:
+        return cuda.CudaNdarray(npa)
+    else:
+        # directly return numpy array
+        warnings.warn("CUDA not found. Unpickling CudaNdarray as numpy.ndarray")
+        return npa
+
 copy_reg.constructor(CudaNdarray_unpickler)