Replacing numpy.asarray by theano._asarray when a dtype conversion may be…

doc/extending/unittest.txt

@@ -389,7 +389,7 @@ Here is an example showing how to use verify_grad:
 >>> def test_flatten_outdimNone():
 >>>    a = dmatrix()
 >>>    # ...
->>>    a_val = numpy.asarray([[0,1,2],[3,4,5]], dtype='float64')
+>>>    a_val = theano._asarray([[0,1,2],[3,4,5]], dtype='float64')
 >>>    # ...
 >>>    tensor.verify_grad(Flatten(), [a_val])
 

doc/sandbox/logistic_regression_example.txt

@@ -67,7 +67,7 @@ BUT, YOU GOTTA RUN THIS CODE AND MAKE SURE IT STILL WORKS NICELY, HEY?
 
     x_data = numpy.random.randn(100, 10)
     y_data = numpy.random.randn(100, 3)
-    y_data = numpy.asarray(y_data == numpy.max(y_data, axis=1), dtype='int64')
+    y_data = theano._asarray(y_data == numpy.max(y_data, axis=1), dtype='int64')
 
     print "Model Training ..."
     for iteration in xrange(1000):

doc/sandbox/sandbox.txt

@@ -173,7 +173,10 @@ More simple numpy stuff
         ``x.reshape(x.size)``
         You can also use ``resize`` but there is not reason to ''resize''
     * How do you convert the type of a numpy array?
-        ``numpy.asarray(x, dtype = 'int32')``
+        ``theano._asarray(x, dtype = 'int32')``
+        Note that using ``numpy.asarray`` is potentially dangerous, due to
+        a problem in numpy where the type may not be properly set (see
+        numpy's Track ticket #870).
 
 
 =========================================

theano/__init__.py

@@ -53,6 +53,7 @@ from compile import \
     ProfileMode, \
     Param, shared
 
+from misc.safe_asarray import _asarray
 
 FancyModule = Module
 

theano/compile/tests/test_pfunc.py

@@ -211,7 +211,7 @@ class Test_pfunc(unittest.TestCase):
 
         z = tensor.ivector()
         c = z*y
-        f = pfunc([y], c+7, givens = {z : numpy.asarray([4,4,4], dtype='int32')})
+        f = pfunc([y], c+7, givens = {z : theano._asarray([4,4,4], dtype='int32')})
         assert numpy.all(f([1,1,1]) == [11,11,11])
         assert x.value == 0
 

theano/compile/tests/test_shared.py

@@ -106,7 +106,7 @@ class Test_SharedVariable(unittest.TestCase):
             pass
 
         # check that an assignment of a perfect value results in no copying
-        uval = numpy.asarray([5,6,7,8], dtype='float64')
+        uval = theano._asarray([5,6,7,8], dtype='float64')
         u.value = uval
         assert u.value is uval
 

theano/misc/safe_asarray.py

+"""
+Helper function to safely convert an array to a new data type.
+"""
+
+__docformat__ = "restructuredtext en"
+
+import numpy
+
+def _asarray(a, dtype=None, order=None):
+    """Convert the input to a Numpy array.
+
+    This function is almost identical to ``numpy.asarray``, but it should be
+    used instead of its numpy counterpart when a data type is provided in
+    order to perform type conversion if required.
+    The reason is that ``numpy.asarray`` may not actually update the array's
+    data type to the user-provided type. For more information see ticket
+    http://projects.scipy.org/numpy/ticket/870.
+
+    Currently, this issue has only been causing trouble when the target
+    data type is 'int32', on some computers. As a result, this is the only
+    situation where we do more than a simple call to ``numpy.asarray``. If it
+    turns out that a similar problem can occur for more data type, this
+    function should be updated accordingly.
+
+    This function's name starts with a '_' to indicate that it is meant to be
+    used internally. It is imported so as to be available directly through
+        theano._asarray
+    """
+    rval = numpy.asarray(a, dtype=dtype, order=order)
+    if dtype is numpy.int32 or dtype == 'int32':
+        # Make sure the type is properly set to the correct type.
+        return rval.view(dtype=numpy.int32)
+    else:
+        # Using ``numpy.asarray`` should work just fine.
+        return rval

theano/sandbox/cuda/basic_ops.py

@@ -57,7 +57,7 @@ class GpuFromHost(Op):
             raise TypeError(x)
         return Apply(self, [x], [CudaNdarrayType(broadcastable=x.broadcastable)()])
     def perform(self, node, (x,), (z,)):
-        z[0] = type_support_filter(numpy.asarray(x, dtype='float32'), tuple([0]*x.ndim), 0)
+        z[0] = type_support_filter(theano._asarray(x, dtype='float32'), tuple([0]*x.ndim), 0)
     def grad(self, inputs, (gz,)):
         return gz,
         #return [HostFromGpu()(gz)]

theano/sandbox/cuda/tests/test_bench_loopfusion.py

@@ -23,7 +23,7 @@ _logger = logging.getLogger('driver_kouh')
 
 def _shared_uniform(rng, low, high, size, dtype, name=None):
     return shared(
-            numpy.asarray(
+            theano._asarray(
                 rng.uniform(low=low, high=high, size=size),
                 dtype=dtype), name)
 
@@ -46,7 +46,7 @@ class Kouh2008(object):
         if len(w_list) != len(x_list):
             raise ValueError('w_list must have same len as x_list')
         output = (sum(w * tensor.pow(x, p) for (w,x) in zip(w_list, x_list)))\
-                / (numpy.asarray(eps, dtype=k.type.dtype) + k + tensor.pow(sum(tensor.pow(x, q) for x in x_list), r))
+                / (theano._asarray(eps, dtype=k.type.dtype) + k + tensor.pow(sum(tensor.pow(x, q) for x in x_list), r))
 
         assert output.type.ndim == 2
         self.__dict__.update(locals())
@@ -79,8 +79,8 @@ class Kouh2008(object):
             w_l2_sqr = sum((wi**2).sum() for wi in w_list)
 
         e_range_low, e_range_high = exponent_range
-        e_range_low = numpy.asarray(e_range_low, dtype=dtype)
-        e_range_high = numpy.asarray(e_range_high, dtype=dtype)
+        e_range_low = theano._asarray(e_range_low, dtype=dtype)
+        e_range_high = theano._asarray(e_range_high, dtype=dtype)
         e_range_mag = e_range_high - e_range_low
         if e_range_mag < 0:
             raise ValueError('exponent range must have low <= high')
@@ -93,8 +93,8 @@ class Kouh2008(object):
         p = tensor.nnet.sigmoid(p_unbounded) * e_range_mag + e_range_low
         q = tensor.nnet.sigmoid(q_unbounded) * e_range_mag + e_range_low
         r = tensor.nnet.sigmoid(r_unbounded) * \
-                numpy.asarray(1.0/e_range_low - 1.0/e_range_high, dtype=dtype) \
-                + numpy.asarray(1.0/e_range_high, dtype=dtype)
+                theano._asarray(1.0/e_range_low - 1.0/e_range_high, dtype=dtype) \
+                + theano._asarray(1.0/e_range_high, dtype=dtype)
 
         k = softsign(k_unbounded)
 
@@ -157,10 +157,10 @@ class Kouh2008(object):
 
         b_list = [shared_uniform(low=0, high=.01, size=(n_out,), name='b_%i'%i)
                 for i in xrange(n_terms)]
-        #x_list = [numpy.asarray(eps, dtype=dtype)+softplus(tensor.dot(input, f_list[i])) for i in xrange(n_terms)]
-        filter_range = numpy.asarray(filter_range, dtype=dtype)
-        half_filter_range = numpy.asarray(filter_range/2, dtype=dtype)
-        x_list = [numpy.asarray(filter_range + eps, dtype=dtype)+half_filter_range *softsign(tensor.dot(input, f_list[i]) +
+        #x_list = [theano._asarray(eps, dtype=dtype)+softplus(tensor.dot(input, f_list[i])) for i in xrange(n_terms)]
+        filter_range = theano._asarray(filter_range, dtype=dtype)
+        half_filter_range = theano._asarray(filter_range/2, dtype=dtype)
+        x_list = [theano._asarray(filter_range + eps, dtype=dtype)+half_filter_range *softsign(tensor.dot(input, f_list[i]) +
             b_list[i]) for i in xrange(n_terms)]
 
         rval = cls.new_expbounds(rng, x_list, n_out, dtype=dtype, params=f_list + b_list,
@@ -304,7 +304,7 @@ def test_bench_elemwise(n_iter=1000, **kwargs):
 
     train_nll = pfunc([x, y, s_lr], [], updates=updates)
 
-    xval = numpy.asarray(
+    xval = theano._asarray(
         rng.uniform(size=(conf.ft_batchsize, x.type.shape[1])),
         dtype=conf.dtype2,
         )

theano/sandbox/cuda/tests/test_conv_cuda_ndarray.py

@@ -59,14 +59,14 @@ def _params_allgood_header():
 def _params_allgood(ishape, kshape, mode, subsample=(1,1), img_stride=(1,1), kern_stride=(1,1), version=-1, verbose=0, random=True, print_=None, id=None, rtol=1e-5, atol = 1e-8, nb_iter=0, ones=False):
     if ones:
         assert not random
-        npy_img = numpy.asarray(numpy.ones(ishape), dtype='float32')
-        npy_kern = -numpy.asarray(numpy.ones(kshape), dtype='float32')
+        npy_img = theano._asarray(numpy.ones(ishape), dtype='float32')
+        npy_kern = -theano._asarray(numpy.ones(kshape), dtype='float32')
     elif random:
-        npy_img = numpy.asarray(numpy.random.rand(*ishape), dtype='float32')
-        npy_kern = numpy.asarray(numpy.random.rand(*kshape), dtype='float32')
+        npy_img = theano._asarray(numpy.random.rand(*ishape), dtype='float32')
+        npy_kern = theano._asarray(numpy.random.rand(*kshape), dtype='float32')
     else:
-        npy_img = numpy.asarray(numpy.arange(numpy.prod(ishape)).reshape(ishape), dtype='float32')+1
-        npy_kern = -(numpy.asarray(numpy.arange(numpy.prod(kshape)).reshape(kshape), dtype='float32')+1)
+        npy_img = theano._asarray(numpy.arange(numpy.prod(ishape)).reshape(ishape), dtype='float32')+1
+        npy_kern = -(theano._asarray(numpy.arange(numpy.prod(kshape)).reshape(kshape), dtype='float32')+1)
 
     img = cuda_ndarray.CudaNdarray(npy_img)
     kern = cuda_ndarray.CudaNdarray(npy_kern)
@@ -369,8 +369,8 @@ def _test_dummy():
     mode = 'valid'
     subsample = (1,1)
 
-    npy_img = numpy.asarray(numpy.random.rand(*ishape), dtype='float32')
-    npy_kern = numpy.asarray(numpy.random.rand(*kshape), dtype='float32')
+    npy_img = theano._asarray(numpy.random.rand(*ishape), dtype='float32')
+    npy_kern = theano._asarray(numpy.random.rand(*kshape), dtype='float32')
 
     img = cuda_ndarray.CudaNdarray(npy_img)
     kern = cuda_ndarray.CudaNdarray(npy_kern)

theano/sandbox/cuda/tests/test_cuda_ndarray.py

@@ -9,14 +9,14 @@ import numpy
 def test_host_to_device():
     print >>sys.stderr, 'starting test_host_to_dev'
     for shape in ((), (3,), (2,3), (3,4,5,6)):
-        a = numpy.asarray(numpy.random.rand(*shape), dtype='float32')
+        a = theano._asarray(numpy.random.rand(*shape), dtype='float32')
         b = cuda_ndarray.CudaNdarray(a)
         c = numpy.asarray(b)
         assert numpy.all(a == c)
 
 def test_add():
     for shape in ((), (3,), (2,3), (1,10000000),(10,1000000), (100,100000),(1000,10000),(10000,1000)):
-        a0 = numpy.asarray(numpy.random.rand(*shape), dtype='float32')
+        a0 = theano._asarray(numpy.random.rand(*shape), dtype='float32')
         a1 = a0.copy()
         b0 = cuda_ndarray.CudaNdarray(a0)
         b1 = cuda_ndarray.CudaNdarray(a1)
@@ -54,7 +54,7 @@ def test_add():
 def test_exp():
     print >>sys.stderr, 'starting test_exp'
     for shape in ((), (3,), (2,3), (1,10000000),(10,1000000), (100,100000),(1000,10000),(10000,1000)):
-        a0 = numpy.asarray(numpy.random.rand(*shape), dtype='float32')
+        a0 = theano._asarray(numpy.random.rand(*shape), dtype='float32')
         a1 = a0.copy()
         b0 = cuda_ndarray.CudaNdarray(a0)
         b1 = cuda_ndarray.CudaNdarray(a1)
@@ -75,7 +75,7 @@ def test_exp():
 def test_copy():
     print >>sys.stderr, 'starting test_copy'
     shape = (5,)
-    a = numpy.asarray(numpy.random.rand(*shape), dtype='float32')
+    a = theano._asarray(numpy.random.rand(*shape), dtype='float32')
 
     print >>sys.stderr, '.. creating device object'
     b = cuda_ndarray.CudaNdarray(a)
@@ -92,8 +92,8 @@ def test_copy():
 
 def test_dot():
     print >>sys.stderr, 'starting test_dot'
-    a0 = numpy.asarray(numpy.random.rand(4, 7), dtype='float32')
-    a1 = numpy.asarray(numpy.random.rand(7, 6), dtype='float32')
+    a0 = theano._asarray(numpy.random.rand(4, 7), dtype='float32')
+    a1 = theano._asarray(numpy.random.rand(7, 6), dtype='float32')
 
     b0 = cuda_ndarray.CudaNdarray(a0)
     b1 = cuda_ndarray.CudaNdarray(a1)
@@ -104,7 +104,7 @@ def test_dot():
 
 def test_sum():
     shape = (2,3)
-    a0 = numpy.asarray(numpy.arange(shape[0]*shape[1]).reshape(shape), dtype='float32')
+    a0 = theano._asarray(numpy.arange(shape[0]*shape[1]).reshape(shape), dtype='float32')
 
     b0 = cuda_ndarray.CudaNdarray(a0)
 
@@ -121,17 +121,17 @@ def test_sum():
     assert numpy.allclose(a0, numpy.asarray(b0.reduce_sum([0,0])))
 
     shape = (3,4,5,6,7,8)
-    a0 = numpy.asarray(numpy.arange(3*4*5*6*7*8).reshape(shape), dtype='float32')
+    a0 = theano._asarray(numpy.arange(3*4*5*6*7*8).reshape(shape), dtype='float32')
     b0 = cuda_ndarray.CudaNdarray(a0)
     assert numpy.allclose(a0.sum(axis=5).sum(axis=3).sum(axis=0), numpy.asarray(b0.reduce_sum([1,0,0,1,0,1])))
 
     shape = (16,2048)
-    a0 = numpy.asarray(numpy.arange(16*2048).reshape(shape), dtype='float32')
+    a0 = theano._asarray(numpy.arange(16*2048).reshape(shape), dtype='float32')
     b0 = cuda_ndarray.CudaNdarray(a0)
     assert numpy.allclose(a0.sum(axis=0), numpy.asarray(b0.reduce_sum([1,0])))
 
     shape = (16,10)
-    a0 = numpy.asarray(numpy.arange(160).reshape(shape), dtype='float32')
+    a0 = theano._asarray(numpy.arange(160).reshape(shape), dtype='float32')
     b0 = cuda_ndarray.CudaNdarray(a0)
     assert numpy.allclose(a0.sum(), numpy.asarray(b0.reduce_sum([1,1])))
 
@@ -147,7 +147,7 @@ def test_reshape():
 
     def subtest(shape_1, shape_2):
         #print >> sys.stderr, "INFO: shapes", shape_1, shape_2
-        a = numpy.asarray(numpy.random.rand(*shape_1), dtype='float32')
+        a = theano._asarray(numpy.random.rand(*shape_1), dtype='float32')
         b = cuda_ndarray.CudaNdarray(a)
 
         aa = a.reshape(shape_2)
@@ -178,7 +178,7 @@ def test_getshape():
              ]
 
     def subtest(shape):
-        a = numpy.asarray(numpy.random.rand(*shape_1), dtype='float32')
+        a = theano._asarray(numpy.random.rand(*shape_1), dtype='float32')
         b = cuda_ndarray.CudaNdarray(a)
         assert b.shape == a.shape
 
@@ -188,7 +188,7 @@ def test_getshape():
 
 def test_stride_manipulation():
 
-    a = numpy.asarray([[0,1,2], [3,4,5]], dtype='float32')
+    a = theano._asarray([[0,1,2], [3,4,5]], dtype='float32')
     b = cuda_ndarray.CudaNdarray(a)
     v = b.view()
     v._dev_data += 0
@@ -212,7 +212,7 @@ def test_stride_manipulation():
 
 def test_copy_subtensor0():
     sizeof_float=4
-    a = numpy.asarray(numpy.random.rand(30,20,5,5), dtype='float32')
+    a = theano._asarray(numpy.random.rand(30,20,5,5), dtype='float32')
     cuda_a = cuda_ndarray.CudaNdarray(a)
     a_view = cuda_a.view()
     a_view_strides = a_view._strides
@@ -225,7 +225,7 @@ def test_copy_subtensor0():
     assert numpy.all(a[:,:,::-1,::-1] == numpy.asarray(a_view_copy))
 
 def test_mapping_getitem_ellipsis():
-    a = numpy.asarray(numpy.random.rand(5,4,3,2), dtype='float32')
+    a = theano._asarray(numpy.random.rand(5,4,3,2), dtype='float32')
     a = cuda_ndarray.CudaNdarray(a)
 
     b = a[...]
@@ -235,7 +235,7 @@ def test_mapping_getitem_ellipsis():
 
 def test_mapping_getitem_reverse_some_dims():
     dim=(5,4,3,2)
-    a = numpy.asarray(numpy.random.rand(*dim), dtype='float32')
+    a = theano._asarray(numpy.random.rand(*dim), dtype='float32')
     _a = cuda_ndarray.CudaNdarray(a)
 
     _b = _a[:,:,::-1, ::-1]
@@ -252,7 +252,7 @@ def test_mapping_getitem_w_int():
         assert numpy.all(x == y)
 
     dim =(2,)
-    a = numpy.asarray(numpy.random.rand(*dim), dtype='float32')
+    a = theano._asarray(numpy.random.rand(*dim), dtype='float32')
     _a = cuda_ndarray.CudaNdarray(a)
     _cmp(numpy.asarray(_a[1]), a[1])
     _cmp(numpy.asarray(_a[::1]), a[::1])
@@ -260,14 +260,14 @@ def test_mapping_getitem_w_int():
     _cmp(numpy.asarray(_a[...]), a[...])
 
     dim =()
-    a = numpy.asarray(numpy.random.rand(*dim), dtype='float32')
+    a = theano._asarray(numpy.random.rand(*dim), dtype='float32')
     _a = cuda_ndarray.CudaNdarray(a)
     _cmp(numpy.asarray(_a[...]), a[...])
 
 
 
     dim =(5,4,3,2)
-    a = numpy.asarray(numpy.random.rand(*dim), dtype='float32')
+    a = theano._asarray(numpy.random.rand(*dim), dtype='float32')
     _a = cuda_ndarray.CudaNdarray(a)
 
     _cmp(numpy.asarray(_a[:,:,::-1, ::-1]), a[:,:,::-1,::-1])
@@ -280,9 +280,9 @@ def test_mapping_getitem_w_int():
     _cmp(numpy.asarray(_a[...]), a[...])
 
 def test_gemm_vector_vector():
-    a = numpy.asarray(numpy.random.rand(5,1), dtype='float32')
+    a = theano._asarray(numpy.random.rand(5,1), dtype='float32')
     _a = cuda_ndarray.CudaNdarray(a)
-    b = numpy.asarray(numpy.random.rand(1,5), dtype='float32')
+    b = theano._asarray(numpy.random.rand(1,5), dtype='float32')
     _b = cuda_ndarray.CudaNdarray(b)
     
     _c = cuda_ndarray.dot(_a,_b)

theano/sandbox/cuda/tests/test_nnet.py

@@ -49,10 +49,10 @@ def run_nnet(use_gpu, n_batch=60, n_in=1024, n_hid=2048, n_out=10, n_iter=100):
         v = tcn.shared_constructor(numpy.zeros((n_hid, n_out)), 'c')
         c = tcn.shared_constructor(numpy.zeros(n_out), 'c')
     else:
-        w = shared(numpy.asarray(0.01*(numpy.random.rand(n_in,n_hid)-0.5), dtype='float32'), 'w')
-        b = shared(numpy.asarray(numpy.zeros(n_hid), dtype='float32'), 'b')
-        v = shared(numpy.asarray(numpy.zeros((n_hid, n_out)), dtype='float32'), 'c')
-        c = shared(numpy.asarray(numpy.zeros(n_out), dtype='float32'), 'c')
+        w = shared(theano._asarray(0.01*(numpy.random.rand(n_in,n_hid)-0.5), dtype='float32'), 'w')
+        b = shared(theano._asarray(numpy.zeros(n_hid), dtype='float32'), 'b')
+        v = shared(theano._asarray(numpy.zeros((n_hid, n_out)), dtype='float32'), 'c')
+        c = shared(theano._asarray(numpy.zeros(n_out), dtype='float32'), 'c')
 
     x = tensor.fmatrix('x')
     y = tensor.fmatrix('y')
@@ -75,9 +75,9 @@ def run_nnet(use_gpu, n_batch=60, n_in=1024, n_hid=2048, n_out=10, n_iter=100):
         for i, n in enumerate(train.maker.env.toposort()):
             print i, n
 
-    xval = numpy.asarray(numpy.random.rand(n_batch, n_in), dtype='float32')
-    yval = numpy.asarray(numpy.random.rand(n_batch, n_out), dtype='float32')
-    lr = numpy.asarray(0.01, dtype='float32')
+    xval = theano._asarray(numpy.random.rand(n_batch, n_in), dtype='float32')
+    yval = theano._asarray(numpy.random.rand(n_batch, n_out), dtype='float32')
+    lr = theano._asarray(0.01, dtype='float32')
 
     t0 = time.time()
     rval = []
@@ -123,10 +123,10 @@ def run_conv_nnet1(use_gpu):
     n_hid = n_kern * logical_hid_shape[0] * logical_hid_shape[1]
     n_out = 10
 
-    w = shared_fn(numpy.asarray(0.01*(numpy.random.rand(*shape_kern)-0.5), dtype='float32'), 'w')
-    b = shared_fn(numpy.asarray(numpy.zeros((n_kern,)), dtype='float32'), 'b')
-    v = shared_fn(numpy.asarray(numpy.zeros((n_hid, n_out)), dtype='float32'), 'c')
-    c = shared_fn(numpy.asarray(numpy.zeros(n_out), dtype='float32'), 'c')
+    w = shared_fn(theano._asarray(0.01*(numpy.random.rand(*shape_kern)-0.5), dtype='float32'), 'w')
+    b = shared_fn(theano._asarray(numpy.zeros((n_kern,)), dtype='float32'), 'b')
+    v = shared_fn(theano._asarray(numpy.zeros((n_hid, n_out)), dtype='float32'), 'c')
+    c = shared_fn(theano._asarray(numpy.zeros(n_out), dtype='float32'), 'c')
 
     x = tensor.Tensor(dtype='float32', broadcastable=(0,1,0,0))('x')
     y = tensor.fmatrix('y')
@@ -152,9 +152,9 @@ def run_conv_nnet1(use_gpu):
 #    for i, n in enumerate(train.maker.env.toposort()):
 #        print i, n
 
-    xval = numpy.asarray(numpy.random.rand(*shape_img), dtype='float32')
-    yval = numpy.asarray(numpy.random.rand(n_batch, n_out), dtype='float32')
-    lr = numpy.asarray(0.01, dtype='float32')
+    xval = theano._asarray(numpy.random.rand(*shape_img), dtype='float32')
+    yval = theano._asarray(numpy.random.rand(n_batch, n_out), dtype='float32')
+    lr = theano._asarray(0.01, dtype='float32')
 
     for i in xrange(10):
         rval = train(xval, yval, lr)
@@ -204,12 +204,12 @@ def run_conv_nnet2(use_gpu): # pretend we are training LeNet for MNIST
     n_hid = n_kern1 * logical_hid_shape1[0] * logical_hid_shape1[1]
     n_out = 10
 
-    w0 = shared_fn(numpy.asarray(0.01*(numpy.random.rand(*shape_kern)-0.5), dtype='float32'), 'w0')
-    b0 = shared_fn(numpy.asarray(numpy.zeros((n_kern,)), dtype='float32'), 'b0')
-    w1 = shared_fn(numpy.asarray(0.01*(numpy.random.rand(*shape_kern1)-0.5), dtype='float32'), 'w1')
-    b1 = shared_fn(numpy.asarray(numpy.zeros((n_kern1,)), dtype='float32'), 'b1')
-    v = shared_fn(numpy.asarray(numpy.zeros((n_hid, n_out)), dtype='float32'), 'c')
-    c = shared_fn(numpy.asarray(numpy.zeros(n_out), dtype='float32'), 'c')
+    w0 = shared_fn(theano._asarray(0.01*(numpy.random.rand(*shape_kern)-0.5), dtype='float32'), 'w0')
+    b0 = shared_fn(theano._asarray(numpy.zeros((n_kern,)), dtype='float32'), 'b0')
+    w1 = shared_fn(theano._asarray(0.01*(numpy.random.rand(*shape_kern1)-0.5), dtype='float32'), 'w1')
+    b1 = shared_fn(theano._asarray(numpy.zeros((n_kern1,)), dtype='float32'), 'b1')
+    v = shared_fn(theano._asarray(numpy.zeros((n_hid, n_out)), dtype='float32'), 'c')
+    c = shared_fn(theano._asarray(numpy.zeros(n_out), dtype='float32'), 'c')
 
     x = tensor.Tensor(dtype='float32', broadcastable=(0,1,0,0))('x')
     y = tensor.fmatrix('y')
@@ -238,9 +238,9 @@ def run_conv_nnet2(use_gpu): # pretend we are training LeNet for MNIST
 #    for i, n in enumerate(train.maker.env.toposort()):
 #        print i, n
 
-    xval = numpy.asarray(numpy.random.rand(*shape_img), dtype='float32')
-    yval = numpy.asarray(numpy.random.rand(n_batch,n_out), dtype='float32')#int32 make all 0...
-    lr = numpy.asarray(0.01, dtype='float32')
+    xval = theano._asarray(numpy.random.rand(*shape_img), dtype='float32')
+    yval = theano._asarray(numpy.random.rand(n_batch,n_out), dtype='float32')#int32 make all 0...
+    lr = theano._asarray(0.01, dtype='float32')
     for i in xrange(n_train):
         rval = train(xval, yval, lr)
 
@@ -284,12 +284,12 @@ def run_conv_nnet2_classif(use_gpu, isize, ksize, n_batch, n_iter,
     n_out = 10
 
 
-    w0 = shared_fn(numpy.asarray(0.01*(numpy.random.rand(*shape_kern)-0.5), dtype='float32'), 'w0')
-    b0 = shared_fn(numpy.asarray(numpy.zeros((n_kern,)), dtype='float32'), 'b0')
-    w1 = shared_fn(numpy.asarray(0.01*(numpy.random.rand(*shape_kern1)-0.5), dtype='float32'), 'w1')
-    b1 = shared_fn(numpy.asarray(numpy.zeros((n_kern1,)), dtype='float32'), 'b1')
-    v = shared_fn(numpy.asarray(0.01*numpy.random.randn(n_hid, n_out), dtype='float32'), 'v')
-    c = shared_fn(numpy.asarray(numpy.zeros(n_out), dtype='float32'), 'c')
+    w0 = shared_fn(theano._asarray(0.01*(numpy.random.rand(*shape_kern)-0.5), dtype='float32'), 'w0')
+    b0 = shared_fn(theano._asarray(numpy.zeros((n_kern,)), dtype='float32'), 'b0')
+    w1 = shared_fn(theano._asarray(0.01*(numpy.random.rand(*shape_kern1)-0.5), dtype='float32'), 'w1')
+    b1 = shared_fn(theano._asarray(numpy.zeros((n_kern1,)), dtype='float32'), 'b1')
+    v = shared_fn(theano._asarray(0.01*numpy.random.randn(n_hid, n_out), dtype='float32'), 'v')
+    c = shared_fn(theano._asarray(numpy.zeros(n_out), dtype='float32'), 'c')
 
     print 'ALLOCATING ARCH: w0 shape', w0.value.shape
     print 'ALLOCATING ARCH: w1 shape', w1.value.shape
@@ -330,9 +330,9 @@ def run_conv_nnet2_classif(use_gpu, isize, ksize, n_batch, n_iter,
         for i, n in enumerate(train.maker.env.toposort()):
             print i, n
 
-    xval = numpy.asarray(numpy.random.rand(*shape_img), dtype='float32')
-    yval = numpy.asarray(numpy.random.rand(n_batch,n_out), dtype='float32')
-    lr = numpy.asarray(0.01, dtype='float32')
+    xval = theano._asarray(numpy.random.rand(*shape_img), dtype='float32')
+    yval = theano._asarray(numpy.random.rand(n_batch,n_out), dtype='float32')
+    lr = theano._asarray(0.01, dtype='float32')
 
     rvals=numpy.zeros(n_iter)
     t0 = time.time()

theano/sandbox/cuda/var.py

@@ -71,7 +71,7 @@ def shared_constructor(value, name, strict=False, broadcastable=None):
     if strict:
         _value = value
     else:
-        _value = numpy.asarray(value, dtype='float32')
+        _value = theano._asarray(value, dtype='float32')
 
     if not isinstance(_value, numpy.ndarray):
         raise TypeError('ndarray required')

theano/sandbox/downsample.py

@@ -8,7 +8,7 @@ DownsampleFactorMax, DownsampleAvg, DownsampleSoftmax.
 
 from theano import gof, Op, tensor, Variable, Apply
 from theano.printing import Print
-import numpy
+import numpy, theano
 import __builtin__
 
 class DownsampleFactorMaxGrad(Op):
@@ -259,7 +259,7 @@ class DownsampleFactorMax(Op):
             raise NotImplementedError('DownsampleFactorMax requires 4D input for now')
         if z[0] is None:
             z[0] = numpy.zeros(self.out_shape(x.shape, self.ds, self.ignore_border)) -float('inf')
-            z[0] = numpy.asarray(z[0], dtype=x.dtype)
+            z[0] = theano._asarray(z[0], dtype=x.dtype)
         zz=z[0]
         ds0, ds1 = self.ds
         if self.ignore_border:

theano/sandbox/scan.py

@@ -197,7 +197,7 @@ class Scan(theano.Op):
             def zero(p):
               return theano.tensor.TensorConstant(theano.tensor.TensorType(\
                       dtype=p.type.dtype, broadcastable=[]),
-                      numpy.asarray(0,dtype = p.type.dtype))
+                      theano._asarray(0,dtype = p.type.dtype))
 
             return [gmap.get(p, zero(p)) for p in inputs]
 

theano/sandbox/theano_object.py

@@ -211,7 +211,7 @@ class TheanoObject(object):
         v = tensor.lscalar(name)
         v._theanoclass_container = \
                 theano.gof.Container(v, 
-                        storage = [numpy.asarray(ival, dtype='int64')],
+                        storage = [theano._asarray(ival, dtype='int64')],
                         readonly=False)
         assert not hasattr(v, 'set')
         assert not hasattr(v, 'get')

theano/scalar/basic.py

@@ -2,7 +2,7 @@ import operator
 import math
 from copy import copy
 
-import numpy
+import numpy, theano
 
 from theano import gof
 from theano.gof import Op, utils, Variable, Constant, Type, Apply, Env
@@ -33,7 +33,7 @@ def as_scalar(x, name = None):
 def constant(x):
     if isinstance(x, float):
         for dtype in ['float32', 'float64']:
-            x_ = numpy.asarray(x, dtype=dtype)
+            x_ = theano._asarray(x, dtype=dtype)
             if numpy.all(x == x_):
                 break
             x_ = None
@@ -41,7 +41,7 @@ def constant(x):
         return ScalarConstant(Scalar(str(x_.dtype)), x)
     if isinstance(x, int):
         for dtype in ['int8', 'int16', 'int32', 'int64']:
-            x_ = numpy.asarray(x, dtype=dtype)
+            x_ = theano._asarray(x, dtype=dtype)
             if numpy.all(x == x_):
                 break
             x_ = None
@@ -1090,7 +1090,7 @@ floor = Floor(same_out_nocomplex, name = 'ceil')
 
 class IRound(UnaryScalarOp):
     def impl(self, x):
-        return numpy.asarray(numpy.round(x), dtype = 'int64')
+        return theano._asarray(numpy.round(x), dtype = 'int64')
     def c_code(self, node, name, (x, ), (z, ), sub):
         return "%(z)s = round(%(x)s);" % locals()
 iround = IRound(int_out_nocomplex)

theano/sparse/basic.py

@@ -7,7 +7,7 @@ To read about different sparse formats, see U{http://www-users.cs.umn.edu/~saad/
 """
 
 import sys, operator
-import numpy
+import numpy, theano
 from scipy import sparse
 import scipy.sparse
 from theano.printing import Print
@@ -279,9 +279,9 @@ class CSMProperties(gof.Op):
           out[0][0] = csm.data[self.kmap]
         #backport
         #out[0][0] = csm.data if self.kmap is None else csm.data[self.kmap]
-        out[1][0] = numpy.asarray(csm.indices, dtype='int32')
-        out[2][0] = numpy.asarray(csm.indptr, dtype='int32')
-        out[3][0] = numpy.asarray(csm.shape, dtype='int32')
+        out[1][0] = theano._asarray(csm.indices, dtype='int32')
+        out[2][0] = theano._asarray(csm.indptr, dtype='int32')
+        out[3][0] = theano._asarray(csm.shape, dtype='int32')
 
     # TODO FIX THIS
     def grad(self, (csm,), g):
@@ -344,28 +344,12 @@ class CSM(gof.Op):
         """
         data = tensor.as_tensor_variable(data)
 
-        # Note that we use `view(numpy.int32)` in addition to providing the
-        # 'int32' dtype to `numpy.asarray`. This is because on some computers
-        # (e.g. a Windows 32 bits machine), we can have the following assert
-        # fail:
-        #   x = numpy.array([0], dtype=numpy.intc)
-        #   y = numpy.asarray(x, dtype=numpy.int32)
-        #   assert y.dtype.num == numpy.dtype(numpy.int32).num
-        # while the assert does *not* fail when replacing the second line by:
-        #   y = numpy.asarray(x, dtype='int32').view(numpy.int32)
-        # This is a known defect in Numpy. For more information see ticket
-        # http://projects.scipy.org/numpy/ticket/870
-
-        # Note also that it is important to keep "dtype='int32'" when calling
-        # `numpy.asarray`. This is because `view` is only some kind of cast to
-        # the exact data type we want to use. If a conversion is required (e.g.
-        # from int64 to int32), it must be done in the call to `numpy.asarray`.
         if not isinstance(indices, tensor.TensorVariable):
-            indices = numpy.asarray(indices, dtype='int32').view(numpy.int32)
+            indices = theano._asarray(indices, dtype='int32')
         if not isinstance(indptr, tensor.TensorVariable):
-            indptr = numpy.asarray(indptr, dtype='int32').view(numpy.int32)
+            indptr = theano._asarray(indptr, dtype='int32')
         if not isinstance(shape, tensor.TensorVariable):
-            shape = numpy.asarray(shape, dtype='int32').view(numpy.int32)
+            shape = theano._asarray(shape, dtype='int32')
         indices = tensor.as_tensor_variable(indices)
         indptr = tensor.as_tensor_variable(indptr)
         shape = tensor.as_tensor_variable(shape)

theano/tensor/basic.py

@@ -9,7 +9,7 @@ import traceback #for overriding Op.__call__
 if sys.version_info >= (2,5):
   import functools
 
-import numpy
+import numpy, theano
 from copy import copy
 
 from theano import gof
@@ -151,7 +151,7 @@ class NumpyAutocaster(object):
         self.dtypes = tuple(dtypes)
     def __call__(self, x):
         for dtype in self.dtypes:
-            x_ = numpy.asarray(x, dtype=dtype)
+            x_ = theano._asarray(x, dtype=dtype)
             if numpy.all(x == x_):
                 break
         # returns either an exact x_==x, or the last casted x_
@@ -163,7 +163,7 @@ autocast_float = NumpyAutocaster(('float32', 'float64'))
 # Note: it's a bit weird for a compiler to automatically downcast literals like this, and it might
 # have implications for efficiency when mixing types.  For example when you add 1.0 +
 # dmatrix(), the 1.0 could be converted to float32, and require upcasting for the + operation
-# at every position in the dmatrix.  using numpy.asarray(1.0, dtype='float64') will circumvent
+# at every position in the dmatrix.  using theano._asarray(1.0, dtype='float64') will circumvent
 # this autocasting, and in future, our ops might be smarter about factoring out upcasts.   The
 # advantage of this mechanism is to combine it with floatX so that 1.0 + xmatrix() will always
 # have the same type as the xmatrix().
@@ -197,7 +197,7 @@ def constant_or_value(x, rtype, name=None, ndim=None, dtype=None):
     """
     if dtype is not None:
         # in this case, the semantics are that the caller is forcing the dtype
-        x_ = numpy.asarray(x, dtype=dtype)
+        x_ = theano._asarray(x, dtype=dtype)
     else:
         # in this case, this function should infer the dtype according to the autocasting
         # rules.  See autocasting above.
@@ -378,7 +378,7 @@ class TensorType(Type):
                             self, self.shape, data.shape))
             return data
         else:
-            data = numpy.asarray(data, dtype = self.dtype)
+            data = theano._asarray(data, dtype = self.dtype)
         if not self.ndim == data.ndim:
             raise TypeError("Wrong number of dimensions: expected %s, got %s with shape %s." % (self.ndim, data.ndim, data.shape), data)
         if any(b and d != 1 for d, b in zip(data.shape, self.broadcastable)):
@@ -1258,7 +1258,7 @@ class Shape(Op):
         x = as_tensor_variable(x)
         return Apply(self, [x], [lvector()])
     def perform(self, node, (x, ), (out, )):
-        out[0] = numpy.asarray(x.shape, dtype = 'int64')
+        out[0] = theano._asarray(x.shape, dtype = 'int64')
     def grad(self, (x,), (gz,)):
         return [None]
 _shape = Shape()
@@ -1300,9 +1300,7 @@ class MaxAndArgmax(Op):
         return Apply(self, inputs, outputs)
     def perform(self, node, (x, axis), (max, max_idx)):
         max[0] = numpy.asarray(numpy.max(x, axis))
-        # Note: using 'view' is important until Numpy's ticket 870 is resolved.
-        max_idx[0] = numpy.asarray(numpy.argmax(x, axis), dtype='int32').view(
-            numpy.int32)
+        max_idx[0] = theano._asarray(numpy.argmax(x, axis), dtype='int32')
     def grad(self, (x, axis), (g_max, g_max_idx)):
         # @warning: This only works if axis is 0, else the max is
         # broadcasted wrong in the call to eq.
@@ -2498,7 +2496,7 @@ class Join(Op):
 
     def perform(self, node, axis_and_tensors, (out, )):
         axis, tensors = axis_and_tensors[0], axis_and_tensors[1:]
-        out[0] = numpy.asarray(numpy.concatenate(tensors, axis = axis),
+        out[0] = theano._asarray(numpy.concatenate(tensors, axis = axis),
                 dtype=node.outputs[0].type.dtype)
 
     def grad(self, axis_and_tensors, (gz,)):
@@ -3422,7 +3420,7 @@ def grad(cost, wrt, g_cost=None, consider_constant=[], warn_type=False):
     def zero(p):
         return TensorConstant(
                 TensorType(dtype = p.type.dtype, broadcastable = []),
-                numpy.asarray(0, dtype=p.type.dtype))
+                theano._asarray(0, dtype=p.type.dtype))
 
     #try:
         #it = iter(wrt)

theano/tensor/elemwise.py

 import sys
 import elemwise_cgen as cgen
 
-import numpy
+import numpy, theano
 from theano import gof
 from theano.gof import Op, Apply
 from theano import scalar
@@ -823,7 +823,7 @@ class CAReduce(Op):
         if to_reduce:
             for dimension in to_reduce:
                 variable = self.ufunc.reduce(variable, dimension)
-            output[0] = numpy.asarray(variable, dtype = node.outputs[0].type.dtype)
+            output[0] = theano._asarray(variable, dtype = node.outputs[0].type.dtype)
         else:
             output[0] = numpy.copy(variable)
 

theano/tensor/nnet.py

@@ -35,7 +35,7 @@ class ScalarSigmoid(scalar.UnaryScalarOp):
         if node.inputs[0].type == scalar.float32:
             # These constants were obtained by looking at the output of python commands like:
             #  for i in xrange(750):
-            #      print i, repr( numpy.asarray(1.0, dtype=dt) / (numpy.asarray(1.0, dtype=dt) + numpy.exp(-numpy.asarray([i,-i], dtype=dt))))
+            #      print i, repr( theano._asarray(1.0, dtype=dt) / (theano._asarray(1.0, dtype=dt) + numpy.exp(-theano._asarray([i,-i], dtype=dt))))
             # the boundary checks prevent us from generating inf
             return """%(z)s = %(x)s < -88.0f ? 0.0 : %(x)s > 15.0f ? 1.0f : 1.0f /(1.0f + exp(-%(x)s));""" % locals()
         elif node.inputs[0].type == scalar.float64:
@@ -70,7 +70,7 @@ class ScalarSoftplus(scalar.UnaryScalarOp):
         if node.inputs[0].type == scalar.float32:
             # These constants were obtained by looking at the output of python commands like:
             #  for i in xrange(750):
-            #      print i, repr( numpy.log1p(numpy.exp(numpy.asarray([i,-i], dtype=dt))))
+            #      print i, repr( numpy.log1p(numpy.exp(theano._asarray([i,-i], dtype=dt))))
             # the boundary checks prevent us from generating inf
             return """%(z)s = %(x)s < -103.0f ? 0.0 : %(x)s > 14.0f ? %(x)s : log1p(exp(%(x)s));""" % locals()
         elif node.inputs[0].type == scalar.float64:

theano/tensor/opt.py

@@ -14,7 +14,7 @@ from elemwise import Elemwise, DimShuffle
 from theano import scalar
 import basic as T
 import inplace as I
-import numpy
+import numpy, theano
 import numpy as N #guys... please don't do this in the library :(
 import operator
 import itertools
@@ -874,7 +874,7 @@ def mul_calculate(num, denum, aslist=False, out_type=None):
         #first = num[0] if num else denum[0]
         one = N.asarray(first).dtype.type(1)
     else:
-        one = N.asarray(1, dtype=out_type.dtype)
+        one = theano._asarray(1, dtype=out_type.dtype)
     v = reduce(N.multiply, num, one) / reduce(N.multiply, denum, one)
     if aslist:
         if N.all(v == 1):
@@ -977,7 +977,7 @@ def local_mul_zero(node):
             #print 'MUL by value', value, node.inputs
             if N.all(value == 0):
                 #print '... returning zeros'
-                return _fill_chain(N.asarray(0, dtype=otype.dtype), node.inputs)
+                return _fill_chain(theano._asarray(0, dtype=otype.dtype), node.inputs)
 register_canonicalize(local_mul_zero)
 
 @gof.local_optimizer([T.true_div])
@@ -1162,8 +1162,8 @@ def add_calculate(num, denum, aslist = False, out_type=None):
     if out_type is None:
       zero = 0.0
     else:
-      zero = N.asarray(0, dtype=out_type.dtype)
-    #zero = 0.0 if out_type is None else N.asarray(0, dtype=out_type.dtype)
+      zero = theano._asarray(0, dtype=out_type.dtype)
+    #zero = 0.0 if out_type is None else theano._asarray(0, dtype=out_type.dtype)
     v = reduce(N.add, num, zero) - reduce(N.add, denum, zero)
     if aslist:
         if N.all(v == 0):

theano/tensor/raw_random.py

@@ -6,7 +6,7 @@ import numpy
 
 #local imports
 import basic as tensor
-import opt
+import opt, theano
 from theano import gof
 from theano.compile import optdb
 
@@ -166,7 +166,7 @@ class RandomFunction(gof.Op):
         rval = self.fn(r, *(args + [tuple(shape)]))
         if not isinstance(rval, numpy.ndarray) \
                or str(rval.dtype) != node.outputs[1].type.dtype:
-            out[0] = numpy.asarray(rval, dtype = node.outputs[1].type.dtype)
+            out[0] = theano._asarray(rval, dtype = node.outputs[1].type.dtype)
         else:
             out[0] = rval
         if len(rval.shape) != self.outtype.ndim:
@@ -182,7 +182,7 @@ def _infer_ndim(ndim, shape):
     """
 
     if isinstance(shape, (tuple, list)):
-        v_shape = tensor.TensorConstant(type=tensor.lvector, data=numpy.asarray(shape, dtype='int64'))
+        v_shape = tensor.TensorConstant(type=tensor.lvector, data=theano._asarray(shape, dtype='int64'))
     else:
         v_shape = tensor.as_tensor_variable(shape)
 

theano/tensor/tests/test_basic.py

@@ -1063,8 +1063,8 @@ class test_bitwise(unittest.TestCase):
     def test_or(self):
         x, y = bvector(), bvector()
         fn = inplace_func([x,y], x|y)
-        l = numpy.asarray([0,0,1,1], dtype = 'int8')
-        r = numpy.asarray([0,1,0,1], dtype = 'int8')
+        l = theano._asarray([0,0,1,1], dtype = 'int8')
+        r = theano._asarray([0,1,0,1], dtype = 'int8')
         v = fn(l, r)
         self.failUnless(numpy.all(v == (operator.or_(l, r))), (l, r, v))
 
@@ -1074,8 +1074,8 @@ class test_bitwise(unittest.TestCase):
         ix = x
         ix = inplace.xor_inplace(ix, y)
         gn = inplace_func([x,y], ix)
-        l = numpy.asarray([0,0,1,1], dtype = 'int8')
-        r = numpy.asarray([0,1,0,1], dtype = 'int8')
+        l = theano._asarray([0,0,1,1], dtype = 'int8')
+        r = theano._asarray([0,1,0,1], dtype = 'int8')
         v = fn(l, r)
         self.failUnless(numpy.all(v == (operator.xor(l, r))), (l, r, v))
         v = gn(l, r)
@@ -1085,16 +1085,16 @@ class test_bitwise(unittest.TestCase):
     def test_and(self):
         x, y = bvector(), bvector()
         fn = inplace_func([x,y], x&y)
-        l = numpy.asarray([0,0,1,1], dtype = 'int8')
-        r = numpy.asarray([0,1,0,1], dtype = 'int8')
+        l = theano._asarray([0,0,1,1], dtype = 'int8')
+        r = theano._asarray([0,1,0,1], dtype = 'int8')
         v = fn(l, r)
         self.failUnless(numpy.all(v == (operator.and_(l, r))), (l, r, v))
 
     def test_inv(self):
         x, y = bvector(), bvector()
         fn = inplace_func([x,y], ~x)
-        l = numpy.asarray([0,0,1,1], dtype = 'int8')
-        r = numpy.asarray([0,1,0,1], dtype = 'int8')
+        l = theano._asarray([0,0,1,1], dtype = 'int8')
+        r = theano._asarray([0,1,0,1], dtype = 'int8')
         v = fn(l, r)
         self.failUnless(numpy.all(v == (~l)), (l, r, v))
 
@@ -1723,9 +1723,9 @@ def test_reshape():
     assert numpy.all(a_val == a_val_copy)
 
     #test that it works with inplace operations
-    a_val = numpy.asarray([0,1,2,3,4,5], dtype='float64')
-    a_val_copy = numpy.asarray([0,1,2,3,4,5], dtype='float64')
-    b_val = numpy.asarray([[0,1,2],[3,4,5]], dtype='float64')
+    a_val = theano._asarray([0,1,2,3,4,5], dtype='float64')
+    a_val_copy = theano._asarray([0,1,2,3,4,5], dtype='float64')
+    b_val = theano._asarray([[0,1,2],[3,4,5]], dtype='float64')
 
     f_sub = inplace_func([a,b], c-b)
     assert numpy.all(f_sub(a_val, b_val) == 0.0)
@@ -1733,7 +1733,7 @@ def test_reshape():
 
     # verify gradient
     def just_vals(v):
-        return Reshape(2)(v, numpy.asarray([2,3], dtype='int32'))
+        return Reshape(2)(v, theano._asarray([2,3], dtype='int32'))
     utt.verify_grad(just_vals, [a_val])
 
 
@@ -1744,8 +1744,8 @@ def test_flatten_outdimNone():
     a = dmatrix()
     c = flatten(a)
     f = inplace_func([a], c)
-    a_val = numpy.asarray([[0,1,2],[3,4,5]], dtype='float64')
-    c_val = numpy.asarray([0,1,2,3,4,5], dtype='float64')
+    a_val = theano._asarray([[0,1,2],[3,4,5]], dtype='float64')
+    c_val = theano._asarray([0,1,2,3,4,5], dtype='float64')
     assert numpy.all(f(a_val)==c_val)
     f = inplace_func([a], c)
     assert numpy.all(f(a_val)==c_val)
@@ -1756,8 +1756,8 @@ def test_flatten_scalar():
     a = dscalar()
     c = flatten(a)
     f = inplace_func([a], c)
-    a_val = numpy.asarray(3.0, dtype='float64')
-    c_val = numpy.asarray([3.0], dtype='float64')
+    a_val = theano._asarray(3.0, dtype='float64')
+    c_val = theano._asarray([3.0], dtype='float64')
     assert numpy.all(f(a_val)==c_val)
     f = inplace_func([a], c)
     assert numpy.all(f(a_val)==c_val)
@@ -1768,8 +1768,8 @@ def test_flatten_outdim1():
     a = dmatrix()
     c = flatten(a, 1)
     f = inplace_func([a], c)
-    a_val = numpy.asarray([[0,1,2],[3,4,5]], dtype='float64')
-    c_val = numpy.asarray([0,1,2,3,4,5], dtype='float64')
+    a_val = theano._asarray([[0,1,2],[3,4,5]], dtype='float64')
+    c_val = theano._asarray([0,1,2,3,4,5], dtype='float64')
     assert numpy.all(f(a_val)==c_val)
     f = inplace_func([a], c)
     assert numpy.all(f(a_val)==c_val)
@@ -1780,7 +1780,7 @@ def test_flatten_outdim2():
     a = dmatrix()
     c = flatten(a, 2)
     f = inplace_func([a], c)
-    a_val = numpy.asarray([[0,1,2],[3,4,5]], dtype='float64')
+    a_val = theano._asarray([[0,1,2],[3,4,5]], dtype='float64')
     assert numpy.all(f(a_val)==a_val)
     f = inplace_func([a], c)
     assert numpy.all(f(a_val)==a_val)
@@ -1791,8 +1791,8 @@ def test_flatten_outdim2_of_3():
     a = TensorType('float64', (False, False, False))()
     c = flatten(a, 2)
     f = inplace_func([a], c)
-    a_val = numpy.asarray([[[0,1],[2,3]], [[4,5],[6,7]]], dtype='float64')
-    c_val = numpy.asarray([[0,1,2,3], [4,5,6,7]], dtype='float64')
+    a_val = theano._asarray([[[0,1],[2,3]], [[4,5],[6,7]]], dtype='float64')
+    c_val = theano._asarray([[0,1,2,3], [4,5,6,7]], dtype='float64')
     assert numpy.all(f(a_val)==c_val)
     f = inplace_func([a], c)
     assert numpy.all(f(a_val)==c_val)
@@ -2288,8 +2288,8 @@ def test_autocast():
         ac.__enter__()
         assert (dvector()+ 1.1).dtype == 'float64'
         assert (fvector()+ 1.1).dtype == 'float32'
-        assert (fvector()+ numpy.asarray(1.1,dtype='float64')).dtype == 'float64'
-        assert (fvector()+ numpy.asarray(1.1,dtype='float32')).dtype == 'float32'
+        assert (fvector()+ theano._asarray(1.1,dtype='float64')).dtype == 'float64'
+        assert (fvector()+ theano._asarray(1.1,dtype='float32')).dtype == 'float32'
 
         assert (dvector()+ 1).dtype == 'float64'
         assert (fvector()+ 1).dtype == 'float32'
@@ -2303,8 +2303,8 @@ def test_autocast():
         assert (dvector()+ 1.1).dtype == 'float64'
         assert (fvector()+ 1.1).dtype == 'float64'
         assert (fvector()+ 1.0).dtype == 'float64'
-        assert (fvector()+ numpy.asarray(1.1,dtype='float64')).dtype == 'float64'
-        assert (fvector()+ numpy.asarray(1.1,dtype='float32')).dtype == 'float32'
+        assert (fvector()+ theano._asarray(1.1,dtype='float64')).dtype == 'float64'
+        assert (fvector()+ theano._asarray(1.1,dtype='float32')).dtype == 'float32'
 
         assert (dvector()+ 1).dtype == 'float64'
         assert (fvector()+ 1).dtype == 'float32'

theano/tensor/tests/test_blas.py

@@ -2,7 +2,7 @@ import traceback
 import theano.tensor as T
 from theano.gof import Env
 from theano.printing import pp
-import numpy
+import numpy, theano
 from theano.tensor.blas import *
 from theano.tensor.blas import _dot22, res_is_a, _as_scalar, _is_real_matrix
 from unittest import TestCase
@@ -152,7 +152,7 @@ class t_gemm(TestCase):
         C = self.rand(4,5)[:,:4]
 
         def t(z,x,y,a=1.0, b=0.0,l='c|py',dt='float64'):
-            z,a,x,y,b = [numpy.asarray(p,dtype=dt) for p in z,a,x,y,b]
+            z,a,x,y,b = [theano._asarray(p,dtype=dt) for p in z,a,x,y,b]
             z_orig = z.copy()
             z_after = self._gemm(z, a, x, y, b)
 

theano/tensor/tests/test_casting.py

@@ -13,7 +13,7 @@ class test_casting(unittest.TestCase):
                 x = type_fn()
                 f = function([x], op_fn(x))
 
-                xval = numpy.asarray(numpy.random.rand(10)*10, dtype=type_fn.dtype)
+                xval = theano._asarray(numpy.random.rand(10)*10, dtype=type_fn.dtype)
                 yval = f(xval)
                 assert str(yval.dtype) == op_fn.scalar_op.output_types_preference.spec[0].dtype
 

theano/tensor/tests/test_opt.py

@@ -73,7 +73,7 @@ def test_add_canonizer_problem0():
 
     n_segments = 10
     label = lscalar('label')
-    segment_labels = label + numpy.asarray([0] * n_segments, dtype='int64')
+    segment_labels = label + theano._asarray([0] * n_segments, dtype='int64')
 
     r = segment_labels * 5
     f = function([label], r)
@@ -149,14 +149,14 @@ class test_canonize(unittest.TestCase):
         dx, dy, dz = dmatrices('xyz')
         fv = fvector('r').dimshuffle('x',0)
         dv = dvector('s').dimshuffle('x',0)
-        fxv = numpy.asarray(numpy.random.rand(*shp),dtype='float32')
-        fyv = numpy.asarray(numpy.random.rand(*shp),dtype='float32')
-        fzv = numpy.asarray(numpy.random.rand(*shp),dtype='float32')
-        fvv = numpy.asarray(numpy.random.rand(shp[0]),dtype='float32').reshape(1,shp[0])
-        dxv = numpy.asarray(numpy.random.rand(*shp),dtype='float64')
-        dyv = numpy.asarray(numpy.random.rand(*shp),dtype='float64')
-        dzv = numpy.asarray(numpy.random.rand(*shp),dtype='float64')
-        dvv = numpy.asarray(numpy.random.rand(shp[0]),dtype='float64').reshape(1,shp[0])
+        fxv = theano._asarray(numpy.random.rand(*shp),dtype='float32')
+        fyv = theano._asarray(numpy.random.rand(*shp),dtype='float32')
+        fzv = theano._asarray(numpy.random.rand(*shp),dtype='float32')
+        fvv = theano._asarray(numpy.random.rand(shp[0]),dtype='float32').reshape(1,shp[0])
+        dxv = theano._asarray(numpy.random.rand(*shp),dtype='float64')
+        dyv = theano._asarray(numpy.random.rand(*shp),dtype='float64')
+        dzv = theano._asarray(numpy.random.rand(*shp),dtype='float64')
+        dvv = theano._asarray(numpy.random.rand(shp[0]),dtype='float64').reshape(1,shp[0])
         cases = [
             (fx+fy,(fx,fy),(fxv,fyv),1,'float32'),
             (fx*fy,(fx,fy),(fxv,fyv),1,'float32'),
@@ -229,14 +229,14 @@ class test_canonize(unittest.TestCase):
         dx, dy, dz = dmatrices('xyz')
         fv = fvector('r').dimshuffle('x',0)
         dv = dvector('s').dimshuffle('x',0)
-        fxv = numpy.asarray(numpy.random.rand(*shp),dtype='float32')
-        fyv = numpy.asarray(numpy.random.rand(*shp),dtype='float32')
-        fzv = numpy.asarray(numpy.random.rand(*shp),dtype='float32')
-        fvv = numpy.asarray(numpy.random.rand(shp[0]),dtype='float32').reshape(1,shp[0])
-        dxv = numpy.asarray(numpy.random.rand(*shp),dtype='float64')
-        dyv = numpy.asarray(numpy.random.rand(*shp),dtype='float64')
-        dzv = numpy.asarray(numpy.random.rand(*shp),dtype='float64')
-        dvv = numpy.asarray(numpy.random.rand(shp[0]),dtype='float64').reshape(1,shp[0])
+        fxv = theano._asarray(numpy.random.rand(*shp),dtype='float32')
+        fyv = theano._asarray(numpy.random.rand(*shp),dtype='float32')
+        fzv = theano._asarray(numpy.random.rand(*shp),dtype='float32')
+        fvv = theano._asarray(numpy.random.rand(shp[0]),dtype='float32').reshape(1,shp[0])
+        dxv = theano._asarray(numpy.random.rand(*shp),dtype='float64')
+        dyv = theano._asarray(numpy.random.rand(*shp),dtype='float64')
+        dzv = theano._asarray(numpy.random.rand(*shp),dtype='float64')
+        dvv = theano._asarray(numpy.random.rand(shp[0]),dtype='float64').reshape(1,shp[0])
         cases = [
             (fx+fy,(fx,fy),(fxv,fyv),1,'float32'),
             (fx*fy,(fx,fy),(fxv,fyv),1,'float32'),
@@ -312,16 +312,16 @@ class test_canonize(unittest.TestCase):
         dx, dy, dz, dw = dmatrices('xyzw')
         fv = fvector('r').dimshuffle('x',0)
         dv = dvector('s').dimshuffle('x',0)
-        fxv = numpy.asarray(numpy.random.rand(*shp),dtype='float32')
-        fyv = numpy.asarray(numpy.random.rand(*shp),dtype='float32')
-        fzv = numpy.asarray(numpy.random.rand(*shp),dtype='float32')
-        fwv = numpy.asarray(numpy.random.rand(*shp),dtype='float32')
-        fvv = numpy.asarray(numpy.random.rand(shp[0]),dtype='float32').reshape(1,shp[0])
-        dxv = numpy.asarray(numpy.random.rand(*shp),dtype='float64')
-        dyv = numpy.asarray(numpy.random.rand(*shp),dtype='float64')
-        dzv = numpy.asarray(numpy.random.rand(*shp),dtype='float64')
-        dwv = numpy.asarray(numpy.random.rand(*shp),dtype='float64')
-        dvv = numpy.asarray(numpy.random.rand(shp[0]),dtype='float64').reshape(1,shp[0])
+        fxv = theano._asarray(numpy.random.rand(*shp),dtype='float32')
+        fyv = theano._asarray(numpy.random.rand(*shp),dtype='float32')
+        fzv = theano._asarray(numpy.random.rand(*shp),dtype='float32')
+        fwv = theano._asarray(numpy.random.rand(*shp),dtype='float32')
+        fvv = theano._asarray(numpy.random.rand(shp[0]),dtype='float32').reshape(1,shp[0])
+        dxv = theano._asarray(numpy.random.rand(*shp),dtype='float64')
+        dyv = theano._asarray(numpy.random.rand(*shp),dtype='float64')
+        dzv = theano._asarray(numpy.random.rand(*shp),dtype='float64')
+        dwv = theano._asarray(numpy.random.rand(*shp),dtype='float64')
+        dvv = theano._asarray(numpy.random.rand(shp[0]),dtype='float64').reshape(1,shp[0])
 
         #We must be sure that the Canonizer is working, but that we don't have other
         # optimisation that could hide bug in the Canonizer as local_elemwise_fusion
@@ -463,13 +463,13 @@ class test_canonize(unittest.TestCase):
         shp=(4,4)
         fx, fy, fz = fmatrices('xyz')
         dx, dy, dz = dmatrices('xyz')
-        fxv = numpy.asarray(numpy.random.rand(*shp),dtype='float32')
-        fyv = numpy.asarray(numpy.random.rand(*shp),dtype='float32')
-        fzv = numpy.asarray(numpy.random.rand(*shp),dtype='float32')
-        dxv = numpy.asarray(numpy.random.rand(*shp),dtype='float32')
-        dyv = numpy.asarray(numpy.random.rand(*shp),dtype='float32')
-        dzv = numpy.asarray(numpy.random.rand(*shp),dtype='float32')
-        fvv = numpy.asarray(numpy.random.rand(shp[0]),dtype='float32').reshape(1,shp[0])
+        fxv = theano._asarray(numpy.random.rand(*shp),dtype='float32')
+        fyv = theano._asarray(numpy.random.rand(*shp),dtype='float32')
+        fzv = theano._asarray(numpy.random.rand(*shp),dtype='float32')
+        dxv = theano._asarray(numpy.random.rand(*shp),dtype='float32')
+        dyv = theano._asarray(numpy.random.rand(*shp),dtype='float32')
+        dzv = theano._asarray(numpy.random.rand(*shp),dtype='float32')
+        fvv = theano._asarray(numpy.random.rand(shp[0]),dtype='float32').reshape(1,shp[0])
         #We must be sure that the Canonizer is working, but that we don't have other
         # optimisation that could hide bug in the Canonizer as local_elemwise_fusion
         mode=compile.mode.predefined_modes[compile.mode.default_mode]
@@ -589,7 +589,7 @@ class test_fusion(unittest.TestCase):
         """
         #TODO: disable the canonizer?
         def my_init(shp, dtype='float64', num=0):
-            #ret = numpy.asarray(numpy.random.rand(*shp),dtype=dtype)
+            #ret = theano._asarray(numpy.random.rand(*shp),dtype=dtype)
             ret = numpy.zeros(shp, dtype=dtype)+num
             return ret
         fw, fx, fy, fz = fmatrices('wxyz')
@@ -601,15 +601,15 @@ class test_fusion(unittest.TestCase):
         fxv = my_init(shp,'float32',2)
         fyv = my_init(shp,'float32',3)
         fzv = my_init(shp,'float32',4)
-        fvv = numpy.asarray(numpy.random.rand(shp[0]),dtype='float32').reshape(1,shp[0])
+        fvv = theano._asarray(numpy.random.rand(shp[0]),dtype='float32').reshape(1,shp[0])
         dwv = my_init(shp,'float64',5)
-        ixv = numpy.asarray(my_init(shp,num=60),dtype='int32')
-        iyv = numpy.asarray(my_init(shp,num=70),dtype='int32')
-        izv = numpy.asarray(my_init(shp,num=70),dtype='int32')
+        ixv = theano._asarray(my_init(shp,num=60),dtype='int32')
+        iyv = theano._asarray(my_init(shp,num=70),dtype='int32')
+        izv = theano._asarray(my_init(shp,num=70),dtype='int32')
 #        dxv = my_init(shp,'float64',6)
 #        dyv = my_init(shp,'float64',7)
 #        dzv = my_init(shp,'float64',8)
-#        dvv = numpy.asarray(numpy.random.rand(shp[0]),dtype='float64').reshape(1,shp[0])
+#        dvv = theano._asarray(numpy.random.rand(shp[0]),dtype='float64').reshape(1,shp[0])
         fwx=fw+fx
         cases = [
             (fx+fy+fz,(fx,fy,fz),(fxv,fyv,fzv),1,fxv+fyv+fzv,'float32'),#1