Skip to content

P
pytensor
提交 824cb369 authored 作者: ballasn's avatar ballasn 提交者： GitHub
浏览文件
操作

Merge pull request #5458 from bscellier/import_numpy_gpuarray

Import numpy gpuarray
上级 2bb065fc 5159a6b3
全部展开 隐藏空白字符变更
内嵌 并排
正在显示 包含 361 行增加 和 364 行删除
theano/gpuarray/basic_ops.py
浏览文件 @ 824cb369
...@@ -2,7 +2,7 @@ from __future__ import absolute_import, print_function, division ...@@ -2,7 +2,7 @@ from __future__ import absolute_import, print_function, division
import os import os
import copy import copy
import re import re
import numpy import numpy as np
from theano import Op, Apply, Type, Variable from theano import Op, Apply, Type, Variable
from theano import tensor, config from theano import tensor, config
...@@ -206,7 +206,7 @@ class Kernel(object): ...@@ -206,7 +206,7 @@ class Kernel(object):
def get_flags(*types): def get_flags(*types):
def get_dtype(t): def get_dtype(t):
if isinstance(t, string_types): if isinstance(t, string_types):
return numpy.dtype(t) return np.dtype(t)
elif isinstance(t, Type): elif isinstance(t, Type):
return t.dtype return t.dtype
elif isinstance(t, Variable): elif isinstance(t, Variable):
...@@ -215,13 +215,13 @@ class Kernel(object): ...@@ -215,13 +215,13 @@ class Kernel(object):
raise TypeError("can't get a dtype from %s" % (type(t),)) raise TypeError("can't get a dtype from %s" % (type(t),))
dtypes = [get_dtype(t) for t in types] dtypes = [get_dtype(t) for t in types]
flags = dict(cluda=True) flags = dict(cluda=True)
if any(d == numpy.float64 for d in dtypes): if any(d == np.float64 for d in dtypes):
flags['have_double'] = True flags['have_double'] = True
if any(d.itemsize < 4 for d in dtypes): if any(d.itemsize < 4 for d in dtypes):
flags['have_small'] = True flags['have_small'] = True
if any(d.kind == 'c' for d in dtypes): if any(d.kind == 'c' for d in dtypes):
flags['have_complex'] = True flags['have_complex'] = True
if any(d == numpy.float16 for d in dtypes): if any(d == np.float16 for d in dtypes):
flags['have_half'] = True flags['have_half'] = True
return flags return flags
...@@ -275,8 +275,8 @@ def get_ctype(dtype): ...@@ -275,8 +275,8 @@ def get_ctype(dtype):
elif dtype == gpuarray.SSIZE: elif dtype == gpuarray.SSIZE:
return "ssize_t" return "ssize_t"
else: else:
if not isinstance(dtype, numpy.dtype): if not isinstance(dtype, np.dtype):
dtype = numpy.dtype(dtype) dtype = np.dtype(dtype)
return 'npy_' + dtype.name return 'npy_' + dtype.name
...@@ -311,7 +311,7 @@ class GpuKernelBase(object): ...@@ -311,7 +311,7 @@ class GpuKernelBase(object):
except MethodNotDefined: except MethodNotDefined:
o = [] o = []
# We rely on the input types for the directory to gpuarray includes # We rely on the input types for the directory to gpuarray includes
return o + [numpy.get_include()] return o + [np.get_include()]
def _generate_kernel_bin(self, k, ctx): def _generate_kernel_bin(self, k, ctx):
gk = gpuarray.GpuKernel(k.code, k.name, k.params, context=ctx, gk = gpuarray.GpuKernel(k.code, k.name, k.params, context=ctx,
...@@ -466,7 +466,7 @@ def get_dtype(s): ...@@ -466,7 +466,7 @@ def get_dtype(s):
if s == 'ssize': if s == 'ssize':
return gpuarray.SSIZE return gpuarray.SSIZE
else: else:
return numpy.dtype(s) return np.dtype(s)
class CGpuKernelBase(COp, GpuKernelBase): class CGpuKernelBase(COp, GpuKernelBase):
...@@ -565,7 +565,7 @@ class HostFromGpu(Op): ...@@ -565,7 +565,7 @@ class HostFromGpu(Op):
def perform(self, node, inp, out): def perform(self, node, inp, out):
x, = inp x, = inp
z, = out z, = out
z[0] = numpy.asarray(x) z[0] = np.asarray(x)
def c_code(self, node, name, inputs, outputs, sub): def c_code(self, node, name, inputs, outputs, sub):
return """ return """
...@@ -1285,7 +1285,7 @@ class GpuJoin(HideC, Join): ...@@ -1285,7 +1285,7 @@ class GpuJoin(HideC, Join):
if axis < 0: if axis < 0:
axis += axis_and_tensors[1].ndim axis += axis_and_tensors[1].ndim
# we check these tensors for being empty. # we check these tensors for being empty.
if (view != -1) and numpy.all( if (view != -1) and np.all(
[tensor.shape[axis] == 0 for tensor in [tensor.shape[axis] == 0 for tensor in
tensors[0:view] + tensors[view + 1:]]): tensors[0:view] + tensors[view + 1:]]):
out[0] = tensors[view] out[0] = tensors[view]
......
theano/gpuarray/blocksparse.py
浏览文件 @ 824cb369
...@@ -2,7 +2,7 @@ from __future__ import absolute_import, print_function, division ...@@ -2,7 +2,7 @@ from __future__ import absolute_import, print_function, division
import logging import logging
import os import os
import numpy import numpy as np
from theano import Apply, tensor from theano import Apply, tensor
from theano.gof import COp from theano.gof import COp
from theano.tensor import discrete_dtypes, as_tensor_variable from theano.tensor import discrete_dtypes, as_tensor_variable
...@@ -121,7 +121,7 @@ class GpuSparseBlockOuter(COp): ...@@ -121,7 +121,7 @@ class GpuSparseBlockOuter(COp):
def make_node(self, o, x, y, xIdx, yIdx, alpha=None): def make_node(self, o, x, y, xIdx, yIdx, alpha=None):
ctx = infer_context_name(o, x, y) ctx = infer_context_name(o, x, y)
one = tensor.constant(numpy.asarray(1.0, dtype='float32')) one = tensor.constant(np.asarray(1.0, dtype='float32'))
o = as_gpuarray_variable(o, ctx) o = as_gpuarray_variable(o, ctx)
x = as_gpuarray_variable(x, ctx) x = as_gpuarray_variable(x, ctx)
y = as_gpuarray_variable(y, ctx) y = as_gpuarray_variable(y, ctx)
......
theano/gpuarray/dnn.py
浏览文件 @ 824cb369
...@@ -4,7 +4,7 @@ import os ...@@ -4,7 +4,7 @@ import os
import sys import sys
import warnings import warnings
import numpy import numpy as np
from six import integer_types from six import integer_types
import theano import theano
...@@ -457,8 +457,8 @@ gpu_dnn_conv_desc.cache = {} ...@@ -457,8 +457,8 @@ gpu_dnn_conv_desc.cache = {}
# scalar constants # scalar constants
_zero = constant(numpy.asarray(0.0, dtype='float64')) _zero = constant(np.asarray(0.0, dtype='float64'))
_one = constant(numpy.asarray(1.0, dtype='float64')) _one = constant(np.asarray(1.0, dtype='float64'))
def ensure_dt(val, default, name, dtype): def ensure_dt(val, default, name, dtype):
...@@ -2482,8 +2482,8 @@ class RNNBlock(object): ...@@ -2482,8 +2482,8 @@ class RNNBlock(object):
bytesize = _get_param_size(self.desc, input_size, self.dtype, bytesize = _get_param_size(self.desc, input_size, self.dtype,
self.context_name) self.context_name)
bytesize = int(bytesize) bytesize = int(bytesize)
assert bytesize % numpy.dtype(self.dtype).itemsize == 0 assert bytesize % np.dtype(self.dtype).itemsize == 0
return bytesize // numpy.dtype(self.dtype).itemsize return bytesize // np.dtype(self.dtype).itemsize
def split_params(self, w, layer, input_size): def split_params(self, w, layer, input_size):
if not isinstance(w, GpuArraySharedVariable): if not isinstance(w, GpuArraySharedVariable):
......
theano/gpuarray/elemwise.py
浏览文件 @ 824cb369
from __future__ import absolute_import, print_function, division from __future__ import absolute_import, print_function, division
import copy import copy
import numpy import numpy as np
import theano import theano
from theano import Apply, scalar, config, Op from theano import Apply, scalar, config, Op
...@@ -27,7 +27,7 @@ from .fp16_help import load_w, write_w ...@@ -27,7 +27,7 @@ from .fp16_help import load_w, write_w
def make_argument(v, name): def make_argument(v, name):
return ArrayArg(numpy.dtype(v.type.dtype), name) return ArrayArg(np.dtype(v.type.dtype), name)
def as_C_string_const(s): def as_C_string_const(s):
......
theano/gpuarray/linalg.py
浏览文件 @ 824cb369
...@@ -7,7 +7,7 @@ import warnings ...@@ -7,7 +7,7 @@ import warnings
from theano import Op from theano import Op
from theano.gpuarray import basic_ops, GpuArrayType from theano.gpuarray import basic_ops, GpuArrayType
import numpy import numpy as np
from numpy.linalg.linalg import LinAlgError from numpy.linalg.linalg import LinAlgError
try: try:
...@@ -107,7 +107,7 @@ class GpuCusolverSolve(Op): ...@@ -107,7 +107,7 @@ class GpuCusolverSolve(Op):
ctx.cusolver_handle = cusolver.cusolverDnCreate() ctx.cusolver_handle = cusolver.cusolverDnCreate()
def check_dev_info(self, dev_info): def check_dev_info(self, dev_info):
val = numpy.asarray(dev_info)[0] val = np.asarray(dev_info)[0]
if val > 0: if val > 0:
raise LinAlgError('A is singular') raise LinAlgError('A is singular')
......
theano/gpuarray/neighbours.py
浏览文件 @ 824cb369
from __future__ import absolute_import, print_function, division from __future__ import absolute_import, print_function, division
import numpy import numpy as np
from theano import Op, Apply, config from theano import Op, Apply, config
from theano.tensor.nnet.neighbours import Images2Neibs from theano.tensor.nnet.neighbours import Images2Neibs
...@@ -256,8 +256,8 @@ class GpuImages2Neibs(GpuKernelBase, Images2Neibs, Op): ...@@ -256,8 +256,8 @@ class GpuImages2Neibs(GpuKernelBase, Images2Neibs, Op):
dtype_neib_shape = node.inputs[1].dtype dtype_neib_shape = node.inputs[1].dtype
dtype_neib_step = node.inputs[2].dtype dtype_neib_step = node.inputs[2].dtype
dtype_z = node.outputs[0].dtype dtype_z = node.outputs[0].dtype
itemsize_ten4 = numpy.dtype(dtype_ten4).itemsize itemsize_ten4 = np.dtype(dtype_ten4).itemsize
itemsize_z = numpy.dtype(dtype_z).itemsize itemsize_z = np.dtype(dtype_z).itemsize
typecode_z = pygpu.gpuarray.dtype_to_typecode(node.outputs[0].dtype) typecode_z = pygpu.gpuarray.dtype_to_typecode(node.outputs[0].dtype)
ten4, neib_shape, neib_step = inp ten4, neib_shape, neib_step = inp
z, = out z, = out
......
theano/gpuarray/nnet.py
浏览文件 @ 824cb369
from __future__ import absolute_import, print_function, division from __future__ import absolute_import, print_function, division
import os import os
import numpy import numpy as np
from theano import Op, Apply, config from theano import Op, Apply, config
from six import StringIO from six import StringIO
...@@ -195,13 +195,13 @@ class GpuCrossentropySoftmaxArgmax1HotWithBias(GpuKernelBase, Op): ...@@ -195,13 +195,13 @@ class GpuCrossentropySoftmaxArgmax1HotWithBias(GpuKernelBase, Op):
def c_code(self, node, nodename, inp, out, sub): def c_code(self, node, nodename, inp, out, sub):
if node.inputs[0].type.context.kind != b'cuda': if node.inputs[0].type.context.kind != b'cuda':
raise NotImplementedError('cuda only') raise NotImplementedError('cuda only')
itemsize_x = numpy.dtype(node.inputs[0].dtype).itemsize itemsize_x = np.dtype(node.inputs[0].dtype).itemsize
worksize_x = numpy.dtype(work_dtype(node.inputs[0].dtype)).itemsize worksize_x = np.dtype(work_dtype(node.inputs[0].dtype)).itemsize
itemsize_b = numpy.dtype(node.inputs[1].dtype).itemsize itemsize_b = np.dtype(node.inputs[1].dtype).itemsize
itemsize_y_idx = numpy.dtype(node.inputs[2].dtype).itemsize itemsize_y_idx = np.dtype(node.inputs[2].dtype).itemsize
itemsize_nll = numpy.dtype(node.outputs[0].dtype).itemsize itemsize_nll = np.dtype(node.outputs[0].dtype).itemsize
itemsize_sm = numpy.dtype(node.outputs[1].dtype).itemsize itemsize_sm = np.dtype(node.outputs[1].dtype).itemsize
itemsize_am = numpy.dtype(node.outputs[2].dtype).itemsize itemsize_am = np.dtype(node.outputs[2].dtype).itemsize
x, b, y_idx = inp x, b, y_idx = inp
nll, sm, am = out nll, sm, am = out
fail = sub['fail'] fail = sub['fail']
...@@ -307,15 +307,15 @@ class GpuCrossentropySoftmax1HotWithBiasDx(GpuKernelBase, Op): ...@@ -307,15 +307,15 @@ class GpuCrossentropySoftmax1HotWithBiasDx(GpuKernelBase, Op):
if node.inputs[0].type.context.kind != b'cuda': if node.inputs[0].type.context.kind != b'cuda':
raise NotImplementedError("cuda only") raise NotImplementedError("cuda only")
typecode_dx = pygpu.gpuarray.dtype_to_typecode(node.outputs[0].dtype) typecode_dx = pygpu.gpuarray.dtype_to_typecode(node.outputs[0].dtype)
itemsize_dnll = numpy.dtype(node.inputs[0].dtype).itemsize itemsize_dnll = np.dtype(node.inputs[0].dtype).itemsize
itemsize_sm = numpy.dtype(node.inputs[1].dtype).itemsize itemsize_sm = np.dtype(node.inputs[1].dtype).itemsize
itemsize_y_idx = numpy.dtype(node.inputs[2].dtype).itemsize itemsize_y_idx = np.dtype(node.inputs[2].dtype).itemsize
itemsize_dx = numpy.dtype(node.outputs[0].dtype).itemsize itemsize_dx = np.dtype(node.outputs[0].dtype).itemsize
dtype_dnll = node.inputs[0].dtype dtype_dnll = node.inputs[0].dtype
dtype_sm = node.inputs[1].dtype dtype_sm = node.inputs[1].dtype
dtype_y_idx = node.inputs[2].dtype dtype_y_idx = node.inputs[2].dtype
dtype_dx = node.outputs[0].dtype dtype_dx = node.outputs[0].dtype
type_intp = gpuarray.dtype_to_ctype(numpy.intp) type_intp = gpuarray.dtype_to_ctype(np.intp)
dnll, sm, y_idx = inp dnll, sm, y_idx = inp
dx, = out dx, = out
fail = sub['fail'] fail = sub['fail']
...@@ -519,8 +519,8 @@ class GpuSoftmax(GpuKernelBase, Op): ...@@ -519,8 +519,8 @@ class GpuSoftmax(GpuKernelBase, Op):
dtype_x = node.inputs[0].dtype dtype_x = node.inputs[0].dtype
work_x = work_dtype(dtype_x) work_x = work_dtype(dtype_x)
dtype_z = node.outputs[0].dtype dtype_z = node.outputs[0].dtype
itemsize_x = numpy.dtype(dtype_x).itemsize itemsize_x = np.dtype(dtype_x).itemsize
itemsize_z = numpy.dtype(dtype_z).itemsize itemsize_z = np.dtype(dtype_z).itemsize
typecode = pygpu.gpuarray.dtype_to_typecode(node.outputs[0].dtype) typecode = pygpu.gpuarray.dtype_to_typecode(node.outputs[0].dtype)
x, = inp x, = inp
z, = out z, = out
...@@ -719,9 +719,9 @@ class GpuSoftmaxWithBias(GpuKernelBase, Op): ...@@ -719,9 +719,9 @@ class GpuSoftmaxWithBias(GpuKernelBase, Op):
dtype_b = node.inputs[1].dtype dtype_b = node.inputs[1].dtype
dtype_z = node.outputs[0].dtype dtype_z = node.outputs[0].dtype
work_x = work_dtype(dtype_x) work_x = work_dtype(dtype_x)
itemsize_x = numpy.dtype(dtype_x).itemsize itemsize_x = np.dtype(dtype_x).itemsize
itemsize_b = numpy.dtype(dtype_b).itemsize itemsize_b = np.dtype(dtype_b).itemsize
itemsize_z = numpy.dtype(dtype_z).itemsize itemsize_z = np.dtype(dtype_z).itemsize
typecode = pygpu.gpuarray.dtype_to_typecode(node.outputs[0].dtype) typecode = pygpu.gpuarray.dtype_to_typecode(node.outputs[0].dtype)
x, b = inp x, b = inp
z, = out z, = out
......
theano/gpuarray/opt.py
浏览文件 @ 824cb369
from __future__ import absolute_import, print_function, division from __future__ import absolute_import, print_function, division
import copy import copy
import numpy import numpy as np
import logging import logging
import pdb import pdb
import time import time
...@@ -622,7 +622,7 @@ def local_gpualloc_memset_0(node): ...@@ -622,7 +622,7 @@ def local_gpualloc_memset_0(node):
inp = node.inputs[0] inp = node.inputs[0]
if (isinstance(inp, GpuArrayConstant) and if (isinstance(inp, GpuArrayConstant) and
inp.data.size == 1 and inp.data.size == 1 and
(numpy.asarray(inp.data) == 0).all()): (np.asarray(inp.data) == 0).all()):
new_op = gpu_alloc(node.op.context_name, memset_0=True) new_op = gpu_alloc(node.op.context_name, memset_0=True)
return [new_op(*node.inputs)] return [new_op(*node.inputs)]
...@@ -632,7 +632,7 @@ def local_gpualloc_memset_0(node): ...@@ -632,7 +632,7 @@ def local_gpualloc_memset_0(node):
def local_gpua_alloc_empty_to_zeros(node): def local_gpua_alloc_empty_to_zeros(node):
if isinstance(node.op, GpuAllocEmpty): if isinstance(node.op, GpuAllocEmpty):
context_name = infer_context_name(*node.inputs) context_name = infer_context_name(*node.inputs)
z = numpy.asarray(0, dtype=node.outputs[0].dtype) z = np.asarray(0, dtype=node.outputs[0].dtype)
return [gpu_alloc(context_name)(as_gpuarray_variable(z, context_name), return [gpu_alloc(context_name)(as_gpuarray_variable(z, context_name),
*node.inputs)] *node.inputs)]
optdb.register('local_gpua_alloc_empty_to_zeros', optdb.register('local_gpua_alloc_empty_to_zeros',
...@@ -830,7 +830,7 @@ def local_gpua_shape_graph(op, context_name, inputs, outputs): ...@@ -830,7 +830,7 @@ def local_gpua_shape_graph(op, context_name, inputs, outputs):
def gpu_print_wrapper(op, cnda): def gpu_print_wrapper(op, cnda):
op.old_op.global_fn(op.old_op, numpy.asarray(cnda)) op.old_op.global_fn(op.old_op, np.asarray(cnda))
@register_opt('fast_compile') @register_opt('fast_compile')
......
theano/gpuarray/opt_util.py
浏览文件 @ 824cb369
from __future__ import absolute_import, print_function, division from __future__ import absolute_import, print_function, division
from functools import wraps from functools import wraps
import numpy import numpy as np
from theano import tensor, scalar as scal, Constant from theano import tensor, scalar as scal, Constant
from theano.gof import local_optimizer from theano.gof import local_optimizer
...@@ -11,7 +11,7 @@ from theano.tensor import (DimShuffle, get_scalar_constant_value, ...@@ -11,7 +11,7 @@ from theano.tensor import (DimShuffle, get_scalar_constant_value,
from .basic_ops import GpuFromHost, HostFromGpu, GpuAllocEmpty, GpuReshape, gpu_alloc_empty from .basic_ops import GpuFromHost, HostFromGpu, GpuAllocEmpty, GpuReshape, gpu_alloc_empty
from .elemwise import GpuDimShuffle, GpuElemwise from .elemwise import GpuDimShuffle, GpuElemwise
_one = scal.constant(numpy.asarray(1.0, dtype='float32')) _one = scal.constant(np.asarray(1.0, dtype='float32'))
def grab_cpu_scalar(v, nd): def grab_cpu_scalar(v, nd):
......
theano/gpuarray/subtensor.py
浏览文件 @ 824cb369
...@@ -2,7 +2,7 @@ from __future__ import absolute_import, print_function, division ...@@ -2,7 +2,7 @@ from __future__ import absolute_import, print_function, division
import os import os
import numpy import numpy as np
from six import integer_types from six import integer_types
from six.moves import StringIO from six.moves import StringIO
...@@ -128,7 +128,7 @@ class GpuSubtensor(HideC, Subtensor): ...@@ -128,7 +128,7 @@ class GpuSubtensor(HideC, Subtensor):
def fix_idx(idx): def fix_idx(idx):
if idx is None: if idx is None:
return "0", 1 return "0", 1
elif isinstance(idx, (numpy.integer, integer_types)): elif isinstance(idx, (np.integer, integer_types)):
return str(idx), 0 return str(idx), 0
elif isinstance(idx, gof.Type): elif isinstance(idx, gof.Type):
return indices.pop(0), 0 return indices.pop(0), 0
...@@ -155,7 +155,7 @@ class GpuSubtensor(HideC, Subtensor): ...@@ -155,7 +155,7 @@ class GpuSubtensor(HideC, Subtensor):
else: else:
if isinstance(idx, gof.Type): if isinstance(idx, gof.Type):
start = indices.pop(0) start = indices.pop(0)
elif isinstance(idx, (numpy.integer, integer_types)): elif isinstance(idx, (np.integer, integer_types)):
start = idx start = idx
else: else:
assert 0, idx assert 0, idx
...@@ -511,7 +511,7 @@ class GpuAdvancedSubtensor(HideC, tensor.AdvancedSubtensor): ...@@ -511,7 +511,7 @@ class GpuAdvancedSubtensor(HideC, tensor.AdvancedSubtensor):
# if there are more than one (narray > 1) it should be ignored. # if there are more than one (narray > 1) it should be ignored.
ap = 0 ap = 0
for k, i in enumerate(list(nidx)): for k, i in enumerate(list(nidx)):
if (isinstance(i, numpy.ndarray) and if (isinstance(i, np.ndarray) and
i.ndim != 0): i.ndim != 0):
transp.remove(k) transp.remove(k)
transp.insert(p, k) transp.insert(p, k)
...@@ -545,7 +545,7 @@ class GpuAdvancedSubtensor(HideC, tensor.AdvancedSubtensor): ...@@ -545,7 +545,7 @@ class GpuAdvancedSubtensor(HideC, tensor.AdvancedSubtensor):
x = x.__getitem__(idx_) x = x.__getitem__(idx_)
# flatten the array-indexed dimensions # flatten the array-indexed dimensions
shape = ((numpy.prod(x.shape[0: p]),) + shape = ((np.prod(x.shape[0: p]),) +
x.shape[p:]) x.shape[p:])
input_flat = x.reshape(shape) input_flat = x.reshape(shape)
...@@ -644,7 +644,7 @@ class GpuAdvancedIncSubtensor1(Op): ...@@ -644,7 +644,7 @@ class GpuAdvancedIncSubtensor1(Op):
# content to index x and y (This is because we serve as # content to index x and y (This is because we serve as
# fallback for _dev20). # fallback for _dev20).
if isinstance(idx, gpuarray.GpuArray): if isinstance(idx, gpuarray.GpuArray):
idx = numpy.asarray(idx) idx = np.asarray(idx)
# If `y` has as many dimensions as `x`, then we want to iterate # If `y` has as many dimensions as `x`, then we want to iterate
# jointly on `x` and `y`. Otherwise, it means `y` should be # jointly on `x` and `y`. Otherwise, it means `y` should be
...@@ -877,10 +877,10 @@ if (GpuArray_vector_add_fast(%(out)s, %(y)s, %(ind)s, %(set_instead_of_inc)s)) { ...@@ -877,10 +877,10 @@ if (GpuArray_vector_add_fast(%(out)s, %(y)s, %(ind)s, %(set_instead_of_inc)s)) {
dtype_y = node.inputs[1].dtype dtype_y = node.inputs[1].dtype
dtype_ind = node.inputs[2].dtype dtype_ind = node.inputs[2].dtype
dtype_out = node.outputs[0].dtype dtype_out = node.outputs[0].dtype
itemsize_x = numpy.dtype(dtype_x).itemsize itemsize_x = np.dtype(dtype_x).itemsize
itemsize_y = numpy.dtype(dtype_y).itemsize itemsize_y = np.dtype(dtype_y).itemsize
itemsize_ind = numpy.dtype(dtype_ind).itemsize itemsize_ind = np.dtype(dtype_ind).itemsize
itemsize_out = numpy.dtype(dtype_out).itemsize itemsize_out = np.dtype(dtype_out).itemsize
flags = Kernel.get_flags(dtype_x, dtype_y, dtype_ind) flags = Kernel.get_flags(dtype_x, dtype_y, dtype_ind)
type_x = gpuarray.dtype_to_ctype(dtype_x) type_x = gpuarray.dtype_to_ctype(dtype_x)
type_y = gpuarray.dtype_to_ctype(dtype_y) type_y = gpuarray.dtype_to_ctype(dtype_y)
...@@ -1007,10 +1007,10 @@ __device__ ga_half atomicExch(ga_half *addr, ga_half val) { ...@@ -1007,10 +1007,10 @@ __device__ ga_half atomicExch(ga_half *addr, ga_half val) {
dtype_y = node.inputs[1].dtype dtype_y = node.inputs[1].dtype
dtype_ind = node.inputs[2].dtype dtype_ind = node.inputs[2].dtype
dtype_out = node.outputs[0].dtype dtype_out = node.outputs[0].dtype
itemsize_x = numpy.dtype(dtype_x).itemsize itemsize_x = np.dtype(dtype_x).itemsize
itemsize_y = numpy.dtype(dtype_y).itemsize itemsize_y = np.dtype(dtype_y).itemsize
itemsize_ind = numpy.dtype(dtype_ind).itemsize itemsize_ind = np.dtype(dtype_ind).itemsize
itemsize_out = numpy.dtype(dtype_out).itemsize itemsize_out = np.dtype(dtype_out).itemsize
k_var = "k_vector_add_fast_" + nodename k_var = "k_vector_add_fast_" + nodename
return super(GpuAdvancedIncSubtensor1_dev20, self).c_support_code_struct(node, nodename) + """ return super(GpuAdvancedIncSubtensor1_dev20, self).c_support_code_struct(node, nodename) + """
...@@ -1112,7 +1112,7 @@ class GpuDiagonal(Subtensor): ...@@ -1112,7 +1112,7 @@ class GpuDiagonal(Subtensor):
if x.size == 0: if x.size == 0:
out_shape = [d for i, d in enumerate(x.shape) out_shape = [d for i, d in enumerate(x.shape)
if i not in (self.axis1, self.axis2)] if i not in (self.axis1, self.axis2)]
diag_size = numpy.min((x.shape[self.axis1], x.shape[self.axis2])) diag_size = np.min((x.shape[self.axis1], x.shape[self.axis2]))
out_shape.append(diag_size) out_shape.append(diag_size)
z[0] = node.outputs[0].type.value_zeros(tuple(out_shape)) z[0] = node.outputs[0].type.value_zeros(tuple(out_shape))
return return
...@@ -1128,15 +1128,15 @@ class GpuDiagonal(Subtensor): ...@@ -1128,15 +1128,15 @@ class GpuDiagonal(Subtensor):
if x.shape[stride_axis] < x.shape[slice_axis]: if x.shape[stride_axis] < x.shape[slice_axis]:
# in the bigger triangle # in the bigger triangle
numstride = small_axis - numpy.max(( numstride = small_axis - np.max((
0, small_axis + numpy.abs(self.offset) - large_axis)) 0, small_axis + np.abs(self.offset) - large_axis))
else: else:
# in the smaller triangle # in the smaller triangle
numstride = small_axis - numpy.abs(self.offset) numstride = small_axis - np.abs(self.offset)
slicer = [numpy.s_[:], ] * x.ndim slicer = [np.s_[:], ] * x.ndim
slicer[stride_axis] = numpy.s_[:numstride] slicer[stride_axis] = np.s_[:numstride]
slicer[slice_axis] = numpy.abs(self.offset) slicer[slice_axis] = np.abs(self.offset)
slicer = tuple(slicer) slicer = tuple(slicer)
# step 2) Swap stride_axis to the last dim because we want the dim on # step 2) Swap stride_axis to the last dim because we want the dim on
......
theano/gpuarray/tests/rnn_support.py
浏览文件 @ 824cb369
...@@ -2,7 +2,7 @@ from __future__ import absolute_import, print_function, division ...@@ -2,7 +2,7 @@ from __future__ import absolute_import, print_function, division
import theano import theano
import theano.tensor as T import theano.tensor as T
import numpy import numpy as np
class Model(object): class Model(object):
...@@ -27,9 +27,9 @@ class Model(object): ...@@ -27,9 +27,9 @@ class Model(object):
def uniform(stdev, size): def uniform(stdev, size):
"""uniform distribution with the given stdev and size""" """uniform distribution with the given stdev and size"""
return numpy.random.uniform( return np.random.uniform(
low=-stdev * numpy.sqrt(3), low=-stdev * np.sqrt(3),
high=stdev * numpy.sqrt(3), high=stdev * np.sqrt(3),
size=size size=size
).astype(theano.config.floatX) ).astype(theano.config.floatX)
...@@ -37,7 +37,7 @@ def uniform(stdev, size): ...@@ -37,7 +37,7 @@ def uniform(stdev, size):
def linear_transform_weights(input_dim, output_dim, def linear_transform_weights(input_dim, output_dim,
param_list=None, name=""): param_list=None, name=""):
"theano shared variable given input and output dimension" "theano shared variable given input and output dimension"
weight_inialization = uniform(numpy.sqrt(2.0 / input_dim), weight_inialization = uniform(np.sqrt(2.0 / input_dim),
(input_dim, output_dim)) (input_dim, output_dim))
W = theano.shared(weight_inialization, name=name) W = theano.shared(weight_inialization, name=name)
...@@ -49,7 +49,7 @@ def linear_transform_weights(input_dim, output_dim, ...@@ -49,7 +49,7 @@ def linear_transform_weights(input_dim, output_dim,
def bias_weights(length, param_list=None, name=""): def bias_weights(length, param_list=None, name=""):
"theano shared variable for bias unit, given length" "theano shared variable for bias unit, given length"
bias_initialization = numpy.zeros(length).astype(theano.config.floatX) bias_initialization = np.zeros(length).astype(theano.config.floatX)
bias = theano.shared( bias = theano.shared(
bias_initialization, bias_initialization,
......
theano/gpuarray/tests/test_abstractconv.py
浏览文件 @ 824cb369
...@@ -3,7 +3,7 @@ from __future__ import absolute_import, print_function, division ...@@ -3,7 +3,7 @@ from __future__ import absolute_import, print_function, division
from nose.plugins.skip import SkipTest from nose.plugins.skip import SkipTest
from nose.tools import assert_raises from nose.tools import assert_raises
import numpy import numpy as np
from theano.tensor.nnet.tests import test_abstract_conv from theano.tensor.nnet.tests import test_abstract_conv
from ..type import GpuArrayType, gpuarray_shared_constructor, get_context from ..type import GpuArrayType, gpuarray_shared_constructor, get_context
...@@ -250,7 +250,7 @@ class TestDnnConvTypes(test_abstract_conv.TestConvTypes): ...@@ -250,7 +250,7 @@ class TestDnnConvTypes(test_abstract_conv.TestConvTypes):
self.filters = gpu_ftensor4() self.filters = gpu_ftensor4()
self.topgrad = gpu_ftensor4() self.topgrad = gpu_ftensor4()
self.constant_tensor = gpuarray.array( self.constant_tensor = gpuarray.array(
numpy.zeros((3, 5, 7, 11), dtype='float32'), np.zeros((3, 5, 7, 11), dtype='float32'),
context=get_context(test_ctx_name)) context=get_context(test_ctx_name))
......
theano/gpuarray/tests/test_basic_ops.py
浏览文件 @ 824cb369
...@@ -4,7 +4,7 @@ from theano.compat import izip ...@@ -4,7 +4,7 @@ from theano.compat import izip
from six import iteritems from six import iteritems
import numpy import numpy as np
import theano import theano
import theano.tensor as T import theano.tensor as T
from theano.tensor import TensorType from theano.tensor import TensorType
...@@ -29,7 +29,7 @@ from .config import mode_with_gpu, mode_without_gpu, test_ctx_name ...@@ -29,7 +29,7 @@ from .config import mode_with_gpu, mode_without_gpu, test_ctx_name
from pygpu import gpuarray from pygpu import gpuarray
utt.seed_rng() utt.seed_rng()
rng = numpy.random.RandomState(seed=utt.fetch_seed()) rng = np.random.RandomState(seed=utt.fetch_seed())
def inplace_func(inputs, outputs, mode=None, allow_input_downcast=False, def inplace_func(inputs, outputs, mode=None, allow_input_downcast=False,
...@@ -92,8 +92,8 @@ def makeTester(name, op, gpu_op, cases, checks=None, mode_gpu=mode_with_gpu, ...@@ -92,8 +92,8 @@ def makeTester(name, op, gpu_op, cases, checks=None, mode_gpu=mode_with_gpu,
for testname, inputs in iteritems(cases): for testname, inputs in iteritems(cases):
for _ in range(len(inputs)): for _ in range(len(inputs)):
if type(inputs[_]) is float: if type(inputs[_]) is float:
inputs[_] = numpy.asarray(inputs[_], inputs[_] = np.asarray(inputs[_],
dtype=theano.config.floatX) dtype=theano.config.floatX)
self.run_case(testname, inputs) self.run_case(testname, inputs)
def run_case(self, testname, inputs): def run_case(self, testname, inputs):
...@@ -177,7 +177,7 @@ def test_transfer_cpu_gpu(): ...@@ -177,7 +177,7 @@ def test_transfer_cpu_gpu():
a = T.fmatrix('a') a = T.fmatrix('a')
g = GpuArrayType(dtype='float32', broadcastable=(False, False))('g') g = GpuArrayType(dtype='float32', broadcastable=(False, False))('g')
av = numpy.asarray(rng.rand(5, 4), dtype='float32') av = np.asarray(rng.rand(5, 4), dtype='float32')
gv = gpuarray.array(av, context=get_context(test_ctx_name)) gv = gpuarray.array(av, context=get_context(test_ctx_name))
f = theano.function([a], GpuFromHost(test_ctx_name)(a)) f = theano.function([a], GpuFromHost(test_ctx_name)(a))
...@@ -186,14 +186,14 @@ def test_transfer_cpu_gpu(): ...@@ -186,14 +186,14 @@ def test_transfer_cpu_gpu():
f = theano.function([g], host_from_gpu(g)) f = theano.function([g], host_from_gpu(g))
fv = f(gv) fv = f(gv)
assert numpy.all(fv == av) assert np.all(fv == av)
def test_transfer_gpu_gpu(): def test_transfer_gpu_gpu():
g = GpuArrayType(dtype='float32', broadcastable=(False, False), g = GpuArrayType(dtype='float32', broadcastable=(False, False),
context_name=test_ctx_name)() context_name=test_ctx_name)()
av = numpy.asarray(rng.rand(5, 4), dtype='float32') av = np.asarray(rng.rand(5, 4), dtype='float32')
gv = gpuarray.array(av, context=get_context(test_ctx_name)) gv = gpuarray.array(av, context=get_context(test_ctx_name))
mode = mode_with_gpu.excluding('cut_gpua_host_transfers', 'local_cut_gpua_host_gpua') mode = mode_with_gpu.excluding('cut_gpua_host_transfers', 'local_cut_gpua_host_gpua')
f = theano.function([g], GpuToGpu(test_ctx_name)(g), mode=mode) f = theano.function([g], GpuToGpu(test_ctx_name)(g), mode=mode)
...@@ -211,7 +211,7 @@ def test_transfer_strided(): ...@@ -211,7 +211,7 @@ def test_transfer_strided():
a = T.fmatrix('a') a = T.fmatrix('a')
g = GpuArrayType(dtype='float32', broadcastable=(False, False))('g') g = GpuArrayType(dtype='float32', broadcastable=(False, False))('g')
av = numpy.asarray(rng.rand(5, 8), dtype='float32') av = np.asarray(rng.rand(5, 8), dtype='float32')
gv = gpuarray.array(av, context=get_context(test_ctx_name)) gv = gpuarray.array(av, context=get_context(test_ctx_name))
av = av[:, ::2] av = av[:, ::2]
...@@ -223,7 +223,7 @@ def test_transfer_strided(): ...@@ -223,7 +223,7 @@ def test_transfer_strided():
f = theano.function([g], host_from_gpu(g)) f = theano.function([g], host_from_gpu(g))
fv = f(gv) fv = f(gv)
assert numpy.all(fv == av) assert np.all(fv == av)
def gpu_alloc_expected(x, *shp): def gpu_alloc_expected(x, *shp):
...@@ -237,16 +237,16 @@ GpuAllocTester = makeTester( ...@@ -237,16 +237,16 @@ GpuAllocTester = makeTester(
op=alloc, op=alloc,
gpu_op=GpuAlloc(test_ctx_name), gpu_op=GpuAlloc(test_ctx_name),
cases=dict( cases=dict(
correct01=(rand(), numpy.int32(7)), correct01=(rand(), np.int32(7)),
# just gives a DeepCopyOp with possibly wrong results on the CPU # just gives a DeepCopyOp with possibly wrong results on the CPU
# correct01_bcast=(rand(1), numpy.int32(7)), # correct01_bcast=(rand(1), np.int32(7)),
correct02=(rand(), numpy.int32(4), numpy.int32(7)), correct02=(rand(), np.int32(4), np.int32(7)),
correct12=(rand(7), numpy.int32(4), numpy.int32(7)), correct12=(rand(7), np.int32(4), np.int32(7)),
correct13=(rand(7), numpy.int32(2), numpy.int32(4), correct13=(rand(7), np.int32(2), np.int32(4),
numpy.int32(7)), np.int32(7)),
correct23=(rand(4, 7), numpy.int32(2), numpy.int32(4), correct23=(rand(4, 7), np.int32(2), np.int32(4),
numpy.int32(7)), np.int32(7)),
bad_shape12=(rand(7), numpy.int32(7), numpy.int32(5)), bad_shape12=(rand(7), np.int32(7), np.int32(5)),
) )
) )
...@@ -282,7 +282,7 @@ def test_shape(): ...@@ -282,7 +282,7 @@ def test_shape():
v = gpuarray.zeros((3, 4, 5), dtype='float32', context=get_context(test_ctx_name)) v = gpuarray.zeros((3, 4, 5), dtype='float32', context=get_context(test_ctx_name))
f = theano.function([x], x.shape) f = theano.function([x], x.shape)
topo = f.maker.fgraph.toposort() topo = f.maker.fgraph.toposort()
assert numpy.all(f(v) == (3, 4, 5)) assert np.all(f(v) == (3, 4, 5))
if theano.config.mode != 'FAST_COMPILE': if theano.config.mode != 'FAST_COMPILE':
assert len(topo) == 4 assert len(topo) == 4
assert isinstance(topo[0].op, T.opt.Shape_i) assert isinstance(topo[0].op, T.opt.Shape_i)
...@@ -292,7 +292,7 @@ def test_shape(): ...@@ -292,7 +292,7 @@ def test_shape():
mode = mode_with_gpu.excluding("local_shape_to_shape_i") mode = mode_with_gpu.excluding("local_shape_to_shape_i")
f = theano.function([x], x.shape, mode=mode) f = theano.function([x], x.shape, mode=mode)
topo = f.maker.fgraph.toposort() topo = f.maker.fgraph.toposort()
assert numpy.all(f(v) == (3, 4, 5)) assert np.all(f(v) == (3, 4, 5))
assert len(topo) == 1 assert len(topo) == 1
assert isinstance(topo[0].op, T.Shape) assert isinstance(topo[0].op, T.Shape)
...@@ -300,7 +300,7 @@ def test_shape(): ...@@ -300,7 +300,7 @@ def test_shape():
def test_gpu_contiguous(): def test_gpu_contiguous():
a = T.fmatrix('a') a = T.fmatrix('a')
i = T.iscalar('i') i = T.iscalar('i')
a_val = numpy.asarray(numpy.random.rand(4, 5), dtype='float32') a_val = np.asarray(np.random.rand(4, 5), dtype='float32')
# The reshape is needed otherwise we make the subtensor on the CPU # The reshape is needed otherwise we make the subtensor on the CPU
# to transfer less data. # to transfer less data.
f = theano.function([a, i], gpu_contiguous(a.reshape((5, 4))[::i]), f = theano.function([a, i], gpu_contiguous(a.reshape((5, 4))[::i]),
...@@ -353,22 +353,22 @@ class G_Join_and_Split(test_basic.T_Join_and_Split): ...@@ -353,22 +353,22 @@ class G_Join_and_Split(test_basic.T_Join_and_Split):
self.shared = gpuarray_shared_constructor self.shared = gpuarray_shared_constructor
def test_gpusplit_opt(self): def test_gpusplit_opt(self):
rng = numpy.random.RandomState(seed=utt.fetch_seed()) rng = np.random.RandomState(seed=utt.fetch_seed())
m = self.shared(rng.rand(4, 6).astype(self.floatX)) m = self.shared(rng.rand(4, 6).astype(self.floatX))
o = T.Split(2)(m, 0, [2, 2]) o = T.Split(2)(m, 0, [2, 2])
f = theano.function([], o, mode=self.mode) f = theano.function([], o, mode=self.mode)
assert any([isinstance(node.op, self.split_op_class) assert any([isinstance(node.op, self.split_op_class)
for node in f.maker.fgraph.toposort()]) for node in f.maker.fgraph.toposort()])
o1, o2 = f() o1, o2 = f()
assert numpy.allclose(o1, m.get_value(borrow=True)[:2]) assert np.allclose(o1, m.get_value(borrow=True)[:2])
assert numpy.allclose(o2, m.get_value(borrow=True)[2:]) assert np.allclose(o2, m.get_value(borrow=True)[2:])
def test_gpujoin_gpualloc(): def test_gpujoin_gpualloc():
a = T.fmatrix('a') a = T.fmatrix('a')
a_val = numpy.asarray(numpy.random.rand(4, 5), dtype='float32') a_val = np.asarray(np.random.rand(4, 5), dtype='float32')
b = T.fmatrix('b') b = T.fmatrix('b')
b_val = numpy.asarray(numpy.random.rand(3, 5), dtype='float32') b_val = np.asarray(np.random.rand(3, 5), dtype='float32')
f = theano.function([a, b], T.join(0, T.zeros_like(a), T.ones_like(b)) + 4, f = theano.function([a, b], T.join(0, T.zeros_like(a), T.ones_like(b)) + 4,
mode=mode_without_gpu) mode=mode_without_gpu)
...@@ -387,7 +387,7 @@ def test_gpujoin_gpualloc(): ...@@ -387,7 +387,7 @@ def test_gpujoin_gpualloc():
for node in f_gpu2.maker.fgraph.toposort()]) == 2 for node in f_gpu2.maker.fgraph.toposort()]) == 2
assert sum([node.op == gpu_join assert sum([node.op == gpu_join
for node in f_gpu2.maker.fgraph.toposort()]) == 1 for node in f_gpu2.maker.fgraph.toposort()]) == 1
assert numpy.allclose(f(a_val, b_val), f_gpu2(a_val, b_val)) assert np.allclose(f(a_val, b_val), f_gpu2(a_val, b_val))
def test_gpueye(): def test_gpueye():
...@@ -401,14 +401,14 @@ def test_gpueye(): ...@@ -401,14 +401,14 @@ def test_gpueye():
M = N M = N
N_symb = T.iscalar() N_symb = T.iscalar()
M_symb = T.iscalar() M_symb = T.iscalar()
k_symb = numpy.asarray(0) k_symb = np.asarray(0)
out = T.eye(N_symb, M_symb, k_symb, dtype=dtype) out = T.eye(N_symb, M_symb, k_symb, dtype=dtype)
f = theano.function([N_symb, M_symb], f = theano.function([N_symb, M_symb],
T.stack(out), T.stack(out),
mode=mode_with_gpu) mode=mode_with_gpu)
result = numpy.asarray(f(N, M)) result = np.asarray(f(N, M))
assert numpy.allclose(result, numpy.eye(N, M_, dtype=dtype)) assert np.allclose(result, np.eye(N, M_, dtype=dtype))
assert result.dtype == numpy.dtype(dtype) assert result.dtype == np.dtype(dtype)
assert any([isinstance(node.op, GpuEye) assert any([isinstance(node.op, GpuEye)
for node in f.maker.fgraph.toposort()]) for node in f.maker.fgraph.toposort()])
...@@ -429,8 +429,8 @@ def test_hostfromgpu_shape_i(): ...@@ -429,8 +429,8 @@ def test_hostfromgpu_shape_i():
'specialize') 'specialize')
a = T.fmatrix('a') a = T.fmatrix('a')
ca = theano.gpuarray.type.GpuArrayType('float32', (False, False))() ca = theano.gpuarray.type.GpuArrayType('float32', (False, False))()
av = numpy.asarray(numpy.random.rand(5, 4), dtype='float32') av = np.asarray(np.random.rand(5, 4), dtype='float32')
cv = gpuarray.asarray(numpy.random.rand(5, 4), cv = gpuarray.asarray(np.random.rand(5, 4),
dtype='float32', dtype='float32',
context=get_context(test_ctx_name)) context=get_context(test_ctx_name))
...@@ -464,7 +464,7 @@ def test_Gpujoin_inplace(): ...@@ -464,7 +464,7 @@ def test_Gpujoin_inplace():
non-empty element. non-empty element.
""" """
s = T.lscalar() s = T.lscalar()
data = numpy.array([3, 4, 5], dtype=theano.config.floatX) data = np.array([3, 4, 5], dtype=theano.config.floatX)
x = gpuarray_shared_constructor(data, borrow=True) x = gpuarray_shared_constructor(data, borrow=True)
z = T.zeros((s,)) z = T.zeros((s,))
...@@ -473,4 +473,4 @@ def test_Gpujoin_inplace(): ...@@ -473,4 +473,4 @@ def test_Gpujoin_inplace():
f = theano.function([s], theano.Out(c, borrow=True)) f = theano.function([s], theano.Out(c, borrow=True))
assert x.get_value(borrow=True, return_internal_type=True) is f(0) assert x.get_value(borrow=True, return_internal_type=True) is f(0)
assert numpy.allclose(f(0), [3, 4, 5]) assert np.allclose(f(0), [3, 4, 5])
theano/gpuarray/tests/test_blas.py
浏览文件 @ 824cb369
...@@ -2,7 +2,7 @@ from __future__ import absolute_import, print_function, division ...@@ -2,7 +2,7 @@ from __future__ import absolute_import, print_function, division
from unittest import TestCase from unittest import TestCase
from nose.plugins.skip import SkipTest from nose.plugins.skip import SkipTest
import itertools import itertools
import numpy import numpy as np
import theano import theano
from theano import tensor from theano import tensor
...@@ -138,8 +138,8 @@ def test_gemv_zeros(): ...@@ -138,8 +138,8 @@ def test_gemv_zeros():
# Apply to an empty matrix shape (5,0) and an empty vector shape (0,) # Apply to an empty matrix shape (5,0) and an empty vector shape (0,)
dim = 1000 dim = 1000
A = numpy.zeros((dim, 0), dtype=theano.config.floatX) A = np.zeros((dim, 0), dtype=theano.config.floatX)
b = numpy.zeros((0,), dtype=theano.config.floatX) b = np.zeros((0,), dtype=theano.config.floatX)
tmp = f(A, b) tmp = f(A, b)
assert numpy.allclose(tmp, assert np.allclose(tmp,
numpy.zeros((dim,))) np.zeros((dim,)))
theano/gpuarray/tests/test_blocksparse.py
浏览文件 @ 824cb369
from __future__ import absolute_import, print_function, division from __future__ import absolute_import, print_function, division
import numpy import numpy as np
import theano import theano
from theano import tensor from theano import tensor
...@@ -39,7 +39,7 @@ class BlockSparse_Gemv_and_Outer(test_blocksparse.BlockSparse_Gemv_and_Outer): ...@@ -39,7 +39,7 @@ class BlockSparse_Gemv_and_Outer(test_blocksparse.BlockSparse_Gemv_and_Outer):
o = gpu_sparse_block_gemv(b.take(oIdx, axis=0), W, h, iIdx, oIdx) o = gpu_sparse_block_gemv(b.take(oIdx, axis=0), W, h, iIdx, oIdx)
gW = theano.grad(o.sum(), W) gW = theano.grad(o.sum(), W)
lr = numpy.asarray(0.05, dtype='float32') lr = np.asarray(0.05, dtype='float32')
upd = W - lr * gW upd = W - lr * gW
......
theano/gpuarray/tests/test_cgpukernelbase.py
浏览文件 @ 824cb369
from __future__ import division, absolute_import, print_function from __future__ import division, absolute_import, print_function
import numpy import numpy as np
from six.moves import xrange from six.moves import xrange
import theano import theano
...@@ -69,4 +69,4 @@ def test_cgpukernelbase(): ...@@ -69,4 +69,4 @@ def test_cgpukernelbase():
r = f() r = f()
assert (numpy.asarray(r) == numpy.eye(4, 5, dtype='int32')).all() assert (np.asarray(r) == np.eye(4, 5, dtype='int32')).all()
theano/gpuarray/tests/test_elemwise.py
浏览文件 @ 824cb369
from __future__ import absolute_import, print_function, division from __future__ import absolute_import, print_function, division
import numpy import numpy as np
import scipy.special import scipy.special
import theano import theano
...@@ -45,8 +45,8 @@ def test_elemwise_pow(): ...@@ -45,8 +45,8 @@ def test_elemwise_pow():
output = base ** exp output = base ** exp
f = theano.function([base, exp], output) f = theano.function([base, exp], output)
base_val = numpy.random.randint(0, 5, size=10).astype(dtype_base) base_val = np.random.randint(0, 5, size=10).astype(dtype_base)
exp_val = numpy.random.randint(0, 3, size=10).astype(dtype_exp) exp_val = np.random.randint(0, 3, size=10).astype(dtype_exp)
# Call the function to make sure the output is valid # Call the function to make sure the output is valid
out = f(base_val, exp_val) out = f(base_val, exp_val)
...@@ -68,7 +68,7 @@ class TestMathErrorFunctions(TestCase): ...@@ -68,7 +68,7 @@ class TestMathErrorFunctions(TestCase):
# to have the GPU ops run on large data. # to have the GPU ops run on large data.
default_array = [x / 10.0 for x in range(-50, 50)] * 1000 default_array = [x / 10.0 for x in range(-50, 50)] * 1000
for dtype in self.dtypes: for dtype in self.dtypes:
numpy_array = numpy.asarray(default_array, dtype=dtype) numpy_array = np.asarray(default_array, dtype=dtype)
self.default_arrays[dtype] = numpy_array self.default_arrays[dtype] = numpy_array
self.expected_erfinv_outputs[dtype] = scipy.special.erfinv(numpy_array) self.expected_erfinv_outputs[dtype] = scipy.special.erfinv(numpy_array)
self.expected_erfcinv_outputs[dtype] = scipy.special.erfcinv(numpy_array) self.expected_erfcinv_outputs[dtype] = scipy.special.erfcinv(numpy_array)
...@@ -127,7 +127,7 @@ class test_float16(): ...@@ -127,7 +127,7 @@ class test_float16():
o = (cz - cz**2 + o = (cz - cz**2 +
tensor.cast(x, 'int16') + tensor.cast(x, 'float32') + tensor.cast(x, 'int16') + tensor.cast(x, 'float32') +
tensor.cast(w, 'float16') - tensor.cast(w, 'float16') -
tensor.constant(numpy.float16(1.0))) tensor.constant(np.float16(1.0)))
theano.function([w, x, y], o, mode=mode_with_gpu) theano.function([w, x, y], o, mode=mode_with_gpu)
...@@ -154,9 +154,9 @@ class test_float16(): ...@@ -154,9 +154,9 @@ class test_float16():
i8.astype('float32')], i8.astype('float32')],
mode=mode_with_gpu) mode=mode_with_gpu)
d1 = (numpy.random.rand(4) * 10).astype('float16') d1 = (np.random.rand(4) * 10).astype('float16')
d2 = (numpy.random.rand(5) * 10).astype('float32') d2 = (np.random.rand(5) * 10).astype('float32')
d3 = (numpy.random.rand(6) * 10).astype('int8') d3 = (np.random.rand(6) * 10).astype('int8')
res = f(d1, d2, d3) res = f(d1, d2, d3)
for i, out in enumerate(f.outputs): for i, out in enumerate(f.outputs):
...@@ -337,9 +337,7 @@ class T_gpureduce_dtype(test_elemwise.T_reduce_dtype): ...@@ -337,9 +337,7 @@ class T_gpureduce_dtype(test_elemwise.T_reduce_dtype):
def speed_reduce10(): def speed_reduce10():
import numpy data = np.random.rand(1000, 1000).astype("float32")
import theano
data = numpy.random.rand(1000, 1000).astype("float32")
m = theano.tensor.fmatrix() m = theano.tensor.fmatrix()
f = theano.function([m], [m.sum(axis=0), m.T.sum(axis=0)], f = theano.function([m], [m.sum(axis=0), m.T.sum(axis=0)],
mode=mode_with_gpu) mode=mode_with_gpu)
......
theano/gpuarray/tests/test_fft.py
浏览文件 @ 824cb369
...@@ -7,7 +7,6 @@ import theano.tensor as T ...@@ -7,7 +7,6 @@ import theano.tensor as T
from theano.tests import unittest_tools as utt from theano.tests import unittest_tools as utt
import theano.gpuarray.fft import theano.gpuarray.fft
import numpy.fft
from .config import mode_with_gpu from .config import mode_with_gpu
...@@ -37,7 +36,7 @@ class TestFFT(unittest.TestCase): ...@@ -37,7 +36,7 @@ class TestFFT(unittest.TestCase):
res_rfft_comp = (np.asarray(res_rfft[:, :, 0]) + res_rfft_comp = (np.asarray(res_rfft[:, :, 0]) +
1j * np.asarray(res_rfft[:, :, 1])) 1j * np.asarray(res_rfft[:, :, 1]))
rfft_ref = numpy.fft.rfft(inputs_val, axis=1) rfft_ref = np.fft.rfft(inputs_val, axis=1)
utt.assert_allclose(rfft_ref, res_rfft_comp) utt.assert_allclose(rfft_ref, res_rfft_comp)
...@@ -72,7 +71,7 @@ class TestFFT(unittest.TestCase): ...@@ -72,7 +71,7 @@ class TestFFT(unittest.TestCase):
res_rfft_comp = (np.asarray(res_rfft[:, :, :, 0]) + res_rfft_comp = (np.asarray(res_rfft[:, :, :, 0]) +
1j * np.asarray(res_rfft[:, :, :, 1])) 1j * np.asarray(res_rfft[:, :, :, 1]))
rfft_ref = numpy.fft.rfftn(inputs_val, axes=(1, 2)) rfft_ref = np.fft.rfftn(inputs_val, axes=(1, 2))
utt.assert_allclose(rfft_ref, res_rfft_comp, atol=1e-4, rtol=1e-4) utt.assert_allclose(rfft_ref, res_rfft_comp, atol=1e-4, rtol=1e-4)
...@@ -91,7 +90,7 @@ class TestFFT(unittest.TestCase): ...@@ -91,7 +90,7 @@ class TestFFT(unittest.TestCase):
utt.assert_allclose(inputs_val, np.asarray(res_ifft)) utt.assert_allclose(inputs_val, np.asarray(res_ifft))
inputs_val = numpy.random.random((1, N, N, 2)).astype('float32') inputs_val = np.random.random((1, N, N, 2)).astype('float32')
inputs = theano.shared(inputs_val) inputs = theano.shared(inputs_val)
irfft = theano.gpuarray.fft.cuirfft(inputs) irfft = theano.gpuarray.fft.cuirfft(inputs)
...@@ -123,7 +122,7 @@ class TestFFT(unittest.TestCase): ...@@ -123,7 +122,7 @@ class TestFFT(unittest.TestCase):
res_rfft_comp = (np.asarray(res_rfft[:, :, :, 0]) + res_rfft_comp = (np.asarray(res_rfft[:, :, :, 0]) +
1j * np.asarray(res_rfft[:, :, :, 1])) 1j * np.asarray(res_rfft[:, :, :, 1]))
rfft_ref = numpy.fft.rfftn(inputs_val, axes=(1, 2)) rfft_ref = np.fft.rfftn(inputs_val, axes=(1, 2))
utt.assert_allclose(rfft_ref / N, res_rfft_comp, atol=1e-4, rtol=1e-4) utt.assert_allclose(rfft_ref / N, res_rfft_comp, atol=1e-4, rtol=1e-4)
...@@ -146,7 +145,7 @@ class TestFFT(unittest.TestCase): ...@@ -146,7 +145,7 @@ class TestFFT(unittest.TestCase):
f_irfft = theano.function([], irfft, mode=mode_with_gpu) f_irfft = theano.function([], irfft, mode=mode_with_gpu)
res_irfft = f_irfft() res_irfft = f_irfft()
irfft_ref = numpy.fft.irfftn(inputs_ref, axes=(1, 2)) irfft_ref = np.fft.irfftn(inputs_ref, axes=(1, 2))
utt.assert_allclose(irfft_ref * N, res_irfft, atol=1e-4, rtol=1e-4) utt.assert_allclose(irfft_ref * N, res_irfft, atol=1e-4, rtol=1e-4)
...@@ -195,7 +194,7 @@ class TestFFT(unittest.TestCase): ...@@ -195,7 +194,7 @@ class TestFFT(unittest.TestCase):
res_rfft_comp = (np.asarray(res_rfft[:, :, :, 0]) + res_rfft_comp = (np.asarray(res_rfft[:, :, :, 0]) +
1j * np.asarray(res_rfft[:, :, :, 1])) 1j * np.asarray(res_rfft[:, :, :, 1]))
rfft_ref = numpy.fft.rfftn(inputs_val, s=(M, M), axes=(1, 2)) rfft_ref = np.fft.rfftn(inputs_val, s=(M, M), axes=(1, 2))
utt.assert_allclose(rfft_ref, res_rfft_comp, atol=1e-4, rtol=1e-4) utt.assert_allclose(rfft_ref, res_rfft_comp, atol=1e-4, rtol=1e-4)
...@@ -214,7 +213,7 @@ class TestFFT(unittest.TestCase): ...@@ -214,7 +213,7 @@ class TestFFT(unittest.TestCase):
res_irfft = f_irfft() res_irfft = f_irfft()
inputs_ref = inputs_val[:, :, :, 0] + 1j * inputs_val[:, :, :, 1] inputs_ref = inputs_val[:, :, :, 0] + 1j * inputs_val[:, :, :, 1]
irfft_ref = numpy.fft.irfftn(inputs_ref, s=(M, M), axes=(1, 2)) * M irfft_ref = np.fft.irfftn(inputs_ref, s=(M, M), axes=(1, 2)) * M
utt.assert_allclose(irfft_ref, res_irfft, atol=1e-4, rtol=1e-4) utt.assert_allclose(irfft_ref, res_irfft, atol=1e-4, rtol=1e-4)
...@@ -243,12 +242,12 @@ class TestFFT(unittest.TestCase): ...@@ -243,12 +242,12 @@ class TestFFT(unittest.TestCase):
utt.verify_grad(f_irfft, [inputs_val], eps=eps) utt.verify_grad(f_irfft, [inputs_val], eps=eps)
def test_params(self): def test_params(self):
inputs_val = numpy.random.random((1, N)).astype('float32') inputs_val = np.random.random((1, N)).astype('float32')
inputs = theano.shared(inputs_val) inputs = theano.shared(inputs_val)
self.assertRaises(ValueError, theano.gpuarray.fft.curfft, inputs, norm=123) self.assertRaises(ValueError, theano.gpuarray.fft.curfft, inputs, norm=123)
inputs_val = numpy.random.random((1, N // 2 + 1, 2)).astype('float32') inputs_val = np.random.random((1, N // 2 + 1, 2)).astype('float32')
inputs = theano.shared(inputs_val) inputs = theano.shared(inputs_val)
self.assertRaises(ValueError, theano.gpuarray.fft.cuirfft, inputs, norm=123) self.assertRaises(ValueError, theano.gpuarray.fft.cuirfft, inputs, norm=123)
......
theano/gpuarray/tests/test_gemmcorr.py
浏览文件 @ 824cb369
from __future__ import absolute_import, print_function, division from __future__ import absolute_import, print_function, division
import unittest import unittest
import numpy import numpy as np
import theano import theano
from theano import config from theano import config
...@@ -23,8 +23,8 @@ class TestCorrMM(unittest.TestCase): ...@@ -23,8 +23,8 @@ class TestCorrMM(unittest.TestCase):
inputs_shape = [inputs_shape[i] for i in (0, 3, 1, 2)] inputs_shape = [inputs_shape[i] for i in (0, 3, 1, 2)]
filters_shape = [filters_shape[i] for i in (0, 3, 1, 2)] filters_shape = [filters_shape[i] for i in (0, 3, 1, 2)]
inputs_val = numpy.random.random(inputs_shape).astype(config.floatX) inputs_val = np.random.random(inputs_shape).astype(config.floatX)
filters_val = numpy.random.random(filters_shape).astype(config.floatX) filters_val = np.random.random(filters_shape).astype(config.floatX)
inputs = gpuarray_shared_constructor(inputs_val) inputs = gpuarray_shared_constructor(inputs_val)
filters = gpuarray_shared_constructor(filters_val) filters = gpuarray_shared_constructor(filters_val)
...@@ -122,11 +122,11 @@ class TestCorrMM(unittest.TestCase): ...@@ -122,11 +122,11 @@ class TestCorrMM(unittest.TestCase):
filters_shape = [filters_shape[i] for i in (0, 3, 1, 2)] filters_shape = [filters_shape[i] for i in (0, 3, 1, 2)]
dCdH_shape = [dCdH_shape[i] for i in (0, 3, 1, 2)] dCdH_shape = [dCdH_shape[i] for i in (0, 3, 1, 2)]
inputs_val = numpy.random.random(inputs_shape).astype(config.floatX) inputs_val = np.random.random(inputs_shape).astype(config.floatX)
dCdH_val = numpy.random.random(dCdH_shape).astype(config.floatX) dCdH_val = np.random.random(dCdH_shape).astype(config.floatX)
inputs = gpuarray_shared_constructor(inputs_val) inputs = gpuarray_shared_constructor(inputs_val)
dCdH = gpuarray_shared_constructor(dCdH_val) dCdH = gpuarray_shared_constructor(dCdH_val)
shape = gpuarray_shared_constructor(numpy.array(filters_shape[2:])) shape = gpuarray_shared_constructor(np.array(filters_shape[2:]))
if (subsample == (1, 1)): if (subsample == (1, 1)):
conv_ref = CorrMM_gradWeights(subsample=subsample)( conv_ref = CorrMM_gradWeights(subsample=subsample)(
...@@ -169,14 +169,14 @@ class TestCorrMM(unittest.TestCase): ...@@ -169,14 +169,14 @@ class TestCorrMM(unittest.TestCase):
inputs_shape = [inputs_shape[i] for i in (0, 3, 1, 2)] inputs_shape = [inputs_shape[i] for i in (0, 3, 1, 2)]
filters_shape = [filters_shape[i] for i in (0, 3, 1, 2)] filters_shape = [filters_shape[i] for i in (0, 3, 1, 2)]
inputs_val = numpy.random.random(inputs_shape).astype(config.floatX) inputs_val = np.random.random(inputs_shape).astype(config.floatX)
filters_val = numpy.random.random(filters_shape).astype(config.floatX) filters_val = np.random.random(filters_shape).astype(config.floatX)
inputs = gpuarray_shared_constructor(inputs_val) inputs = gpuarray_shared_constructor(inputs_val)
filters = gpuarray_shared_constructor(filters_val) filters = gpuarray_shared_constructor(filters_val)
bottom_height = (inputs_shape[2] - 1) * subsample[0] + filters_shape[2] bottom_height = (inputs_shape[2] - 1) * subsample[0] + filters_shape[2]
bottom_width = (inputs_shape[3] - 1) * subsample[1] + filters_shape[3] bottom_width = (inputs_shape[3] - 1) * subsample[1] + filters_shape[3]
bottom_shape = gpuarray_shared_constructor(numpy.array([bottom_height, bottom_width])) bottom_shape = gpuarray_shared_constructor(np.array([bottom_height, bottom_width]))
if (subsample == (1, 1)): if (subsample == (1, 1)):
conv_ref = CorrMM_gradInputs(subsample=subsample)( conv_ref = CorrMM_gradInputs(subsample=subsample)(
......
theano/gpuarray/tests/test_gemmcorr3d.py
浏览文件 @ 824cb369
from __future__ import absolute_import, print_function, division from __future__ import absolute_import, print_function, division
import unittest import unittest
import numpy import numpy as np
import theano import theano
from theano import config from theano import config
...@@ -23,8 +23,8 @@ class TestCorr3dMM(unittest.TestCase): ...@@ -23,8 +23,8 @@ class TestCorr3dMM(unittest.TestCase):
inputs_shape = [inputs_shape[i] for i in (0, 4, 1, 2, 3)] inputs_shape = [inputs_shape[i] for i in (0, 4, 1, 2, 3)]
filters_shape = [filters_shape[i] for i in (0, 4, 1, 2, 3)] filters_shape = [filters_shape[i] for i in (0, 4, 1, 2, 3)]
inputs_val = numpy.random.random(inputs_shape).astype(config.floatX) inputs_val = np.random.random(inputs_shape).astype(config.floatX)
filters_val = numpy.random.random(filters_shape).astype(config.floatX) filters_val = np.random.random(filters_shape).astype(config.floatX)
inputs = gpuarray_shared_constructor(inputs_val) inputs = gpuarray_shared_constructor(inputs_val)
filters = gpuarray_shared_constructor(filters_val) filters = gpuarray_shared_constructor(filters_val)
...@@ -121,11 +121,11 @@ class TestCorr3dMM(unittest.TestCase): ...@@ -121,11 +121,11 @@ class TestCorr3dMM(unittest.TestCase):
filters_shape = [filters_shape[i] for i in (0, 4, 1, 2, 3)] filters_shape = [filters_shape[i] for i in (0, 4, 1, 2, 3)]
dCdH_shape = [dCdH_shape[i] for i in (0, 4, 1, 2, 3)] dCdH_shape = [dCdH_shape[i] for i in (0, 4, 1, 2, 3)]
inputs_val = numpy.random.random(inputs_shape).astype(config.floatX) inputs_val = np.random.random(inputs_shape).astype(config.floatX)
dCdH_val = numpy.random.random(dCdH_shape).astype(config.floatX) dCdH_val = np.random.random(dCdH_shape).astype(config.floatX)
inputs = gpuarray_shared_constructor(inputs_val) inputs = gpuarray_shared_constructor(inputs_val)
dCdH = gpuarray_shared_constructor(dCdH_val) dCdH = gpuarray_shared_constructor(dCdH_val)
shape = gpuarray_shared_constructor(numpy.array(filters_shape[2:])) shape = gpuarray_shared_constructor(np.array(filters_shape[2:]))
if (subsample == (1, 1, 1)): if (subsample == (1, 1, 1)):
conv_ref = Corr3dMM_gradWeights(subsample=subsample)( conv_ref = Corr3dMM_gradWeights(subsample=subsample)(
...@@ -168,15 +168,15 @@ class TestCorr3dMM(unittest.TestCase): ...@@ -168,15 +168,15 @@ class TestCorr3dMM(unittest.TestCase):
inputs_shape = [inputs_shape[i] for i in (0, 4, 1, 2, 3)] inputs_shape = [inputs_shape[i] for i in (0, 4, 1, 2, 3)]
filters_shape = [filters_shape[i] for i in (0, 4, 1, 2, 3)] filters_shape = [filters_shape[i] for i in (0, 4, 1, 2, 3)]
inputs_val = numpy.random.random(inputs_shape).astype(config.floatX) inputs_val = np.random.random(inputs_shape).astype(config.floatX)
filters_val = numpy.random.random(filters_shape).astype(config.floatX) filters_val = np.random.random(filters_shape).astype(config.floatX)
inputs = gpuarray_shared_constructor(inputs_val) inputs = gpuarray_shared_constructor(inputs_val)
filters = gpuarray_shared_constructor(filters_val) filters = gpuarray_shared_constructor(filters_val)
bottom_height = (inputs_shape[2] - 1) * subsample[0] + filters_shape[2] bottom_height = (inputs_shape[2] - 1) * subsample[0] + filters_shape[2]
bottom_width = (inputs_shape[3] - 1) * subsample[1] + filters_shape[3] bottom_width = (inputs_shape[3] - 1) * subsample[1] + filters_shape[3]
bottom_depth = (inputs_shape[4] - 1) * subsample[2] + filters_shape[4] bottom_depth = (inputs_shape[4] - 1) * subsample[2] + filters_shape[4]
bottom_shape = gpuarray_shared_constructor(numpy.array([bottom_height, bottom_width, bottom_depth])) bottom_shape = gpuarray_shared_constructor(np.array([bottom_height, bottom_width, bottom_depth]))
if (subsample == (1, 1, 1)): if (subsample == (1, 1, 1)):
conv_ref = Corr3dMM_gradInputs(subsample=subsample)( conv_ref = Corr3dMM_gradInputs(subsample=subsample)(
......
theano/gpuarray/tests/test_linalg.py
浏览文件 @ 824cb369
from __future__ import absolute_import, division, print_function from __future__ import absolute_import, division, print_function
import unittest import unittest
import numpy import numpy as np
import theano import theano
from theano.tests import unittest_tools as utt from theano.tests import unittest_tools as utt
...@@ -19,8 +19,8 @@ if not cusolver_available: ...@@ -19,8 +19,8 @@ if not cusolver_available:
class TestCusolver(unittest.TestCase): class TestCusolver(unittest.TestCase):
def run_gpu_solve(self, A_val, x_val, A_struct=None): def run_gpu_solve(self, A_val, x_val, A_struct=None):
b_val = numpy.dot(A_val, x_val) b_val = np.dot(A_val, x_val)
b_val_trans = numpy.dot(A_val.T, x_val) b_val_trans = np.dot(A_val.T, x_val)
A = theano.tensor.matrix("A", dtype="float32") A = theano.tensor.matrix("A", dtype="float32")
b = theano.tensor.matrix("b", dtype="float32") b = theano.tensor.matrix("b", dtype="float32")
...@@ -35,59 +35,59 @@ class TestCusolver(unittest.TestCase): ...@@ -35,59 +35,59 @@ class TestCusolver(unittest.TestCase):
fn = theano.function([A, b, b_trans], [solver, solver_trans], mode=mode_with_gpu) fn = theano.function([A, b, b_trans], [solver, solver_trans], mode=mode_with_gpu)
res = fn(A_val, b_val, b_val_trans) res = fn(A_val, b_val, b_val_trans)
x_res = numpy.array(res[0]) x_res = np.array(res[0])
x_res_trans = numpy.array(res[1]) x_res_trans = np.array(res[1])
utt.assert_allclose(x_val, x_res) utt.assert_allclose(x_val, x_res)
utt.assert_allclose(x_val, x_res_trans) utt.assert_allclose(x_val, x_res_trans)
def test_diag_solve(self): def test_diag_solve(self):
numpy.random.seed(1) np.random.seed(1)
A_val = numpy.asarray([[2, 0, 0], [0, 1, 0], [0, 0, 1]], A_val = np.asarray([[2, 0, 0], [0, 1, 0], [0, 0, 1]],
dtype="float32") dtype="float32")
x_val = numpy.random.uniform(-0.4, 0.4, (A_val.shape[1], x_val = np.random.uniform(-0.4, 0.4, (A_val.shape[1],
1)).astype("float32") 1)).astype("float32")
self.run_gpu_solve(A_val, x_val) self.run_gpu_solve(A_val, x_val)
def test_bshape_solve(self): def test_bshape_solve(self):
""" """
Test when shape of b (k, m) is such as m > k Test when shape of b (k, m) is such as m > k
""" """
numpy.random.seed(1) np.random.seed(1)
A_val = numpy.asarray([[2, 0, 0], [0, 1, 0], [0, 0, 1]], A_val = np.asarray([[2, 0, 0], [0, 1, 0], [0, 0, 1]],
dtype="float32") dtype="float32")
x_val = numpy.random.uniform(-0.4, 0.4, (A_val.shape[1], x_val = np.random.uniform(-0.4, 0.4, (A_val.shape[1],
A_val.shape[1] + 1)).astype("float32") A_val.shape[1] + 1)).astype("float32")
self.run_gpu_solve(A_val, x_val) self.run_gpu_solve(A_val, x_val)
def test_sym_solve(self): def test_sym_solve(self):
numpy.random.seed(1) np.random.seed(1)
A_val = numpy.random.uniform(-0.4, 0.4, (5, 5)).astype("float32") A_val = np.random.uniform(-0.4, 0.4, (5, 5)).astype("float32")
A_sym = numpy.dot(A_val, A_val.T) A_sym = np.dot(A_val, A_val.T)
x_val = numpy.random.uniform(-0.4, 0.4, (A_val.shape[1], x_val = np.random.uniform(-0.4, 0.4, (A_val.shape[1],
1)).astype("float32") 1)).astype("float32")
self.run_gpu_solve(A_sym, x_val, 'symmetric') self.run_gpu_solve(A_sym, x_val, 'symmetric')
def test_orth_solve(self): def test_orth_solve(self):
numpy.random.seed(1) np.random.seed(1)
A_val = numpy.random.uniform(-0.4, 0.4, (5, 5)).astype("float32") A_val = np.random.uniform(-0.4, 0.4, (5, 5)).astype("float32")
A_orth = numpy.linalg.svd(A_val)[0] A_orth = np.linalg.svd(A_val)[0]
x_val = numpy.random.uniform(-0.4, 0.4, (A_orth.shape[1], x_val = np.random.uniform(-0.4, 0.4, (A_orth.shape[1],
1)).astype("float32") 1)).astype("float32")
self.run_gpu_solve(A_orth, x_val) self.run_gpu_solve(A_orth, x_val)
def test_uni_rand_solve(self): def test_uni_rand_solve(self):
numpy.random.seed(1) np.random.seed(1)
A_val = numpy.random.uniform(-0.4, 0.4, (5, 5)).astype("float32") A_val = np.random.uniform(-0.4, 0.4, (5, 5)).astype("float32")
x_val = numpy.random.uniform(-0.4, 0.4, x_val = np.random.uniform(-0.4, 0.4,
(A_val.shape[1], 4)).astype("float32") (A_val.shape[1], 4)).astype("float32")
self.run_gpu_solve(A_val, x_val) self.run_gpu_solve(A_val, x_val)
def test_linalgerrsym_solve(self): def test_linalgerrsym_solve(self):
numpy.random.seed(1) np.random.seed(1)
A_val = numpy.random.uniform(-0.4, 0.4, (5, 5)).astype("float32") A_val = np.random.uniform(-0.4, 0.4, (5, 5)).astype("float32")
x_val = numpy.random.uniform(-0.4, 0.4, x_val = np.random.uniform(-0.4, 0.4,
(A_val.shape[1], 4)).astype("float32") (A_val.shape[1], 4)).astype("float32")
A_val = numpy.dot(A_val.T, A_val) A_val = np.dot(A_val.T, A_val)
# make A singular # make A singular
A_val[:, 2] = A_val[:, 1] + A_val[:, 3] A_val[:, 2] = A_val[:, 1] + A_val[:, 3]
...@@ -99,10 +99,10 @@ class TestCusolver(unittest.TestCase): ...@@ -99,10 +99,10 @@ class TestCusolver(unittest.TestCase):
self.assertRaises(LinAlgError, fn, A_val, x_val) self.assertRaises(LinAlgError, fn, A_val, x_val)
def test_linalgerr_solve(self): def test_linalgerr_solve(self):
numpy.random.seed(1) np.random.seed(1)
A_val = numpy.random.uniform(-0.4, 0.4, (5, 5)).astype("float32") A_val = np.random.uniform(-0.4, 0.4, (5, 5)).astype("float32")
x_val = numpy.random.uniform(-0.4, 0.4, x_val = np.random.uniform(-0.4, 0.4,
(A_val.shape[1], 4)).astype("float32") (A_val.shape[1], 4)).astype("float32")
# make A singular # make A singular
A_val[:, 2] = 0 A_val[:, 2] = 0
......
theano/gpuarray/tests/test_multinomial.py
浏览文件 @ 824cb369
from __future__ import absolute_import, print_function, division from __future__ import absolute_import, print_function, division
import numpy import numpy as np
import unittest import unittest
...@@ -62,10 +62,10 @@ def test_multinomial_large(): ...@@ -62,10 +62,10 @@ def test_multinomial_large():
assert any([type(node.op) is GPUAMultinomialFromUniform assert any([type(node.op) is GPUAMultinomialFromUniform
for node in f.maker.fgraph.toposort()]) for node in f.maker.fgraph.toposort()])
pval = numpy.arange(10000 * 4, pval = np.arange(10000 * 4,
dtype='float32').reshape((10000, 4)) + 0.1 dtype='float32').reshape((10000, 4)) + 0.1
pval = pval / pval.sum(axis=1)[:, None] pval = pval / pval.sum(axis=1)[:, None]
uval = numpy.ones_like(pval[:, 0]) * 0.5 uval = np.ones_like(pval[:, 0]) * 0.5
mval = f(pval, uval) mval = f(pval, uval)
assert mval.shape == pval.shape assert mval.shape == pval.shape
...@@ -78,7 +78,7 @@ def test_multinomial_large(): ...@@ -78,7 +78,7 @@ def test_multinomial_large():
else: else:
raise NotImplementedError(config.cast_policy) raise NotImplementedError(config.cast_policy)
utt.assert_allclose(mval.sum(axis=1), 2) utt.assert_allclose(mval.sum(axis=1), 2)
asdf = numpy.asarray([0, 0, 2, 0]) + 0 * pval asdf = np.asarray([0, 0, 2, 0]) + 0 * pval
utt.assert_allclose(mval, asdf) # broadcast over all rows utt.assert_allclose(mval, asdf) # broadcast over all rows
...@@ -92,9 +92,9 @@ def test_gpu_opt_dtypes(): ...@@ -92,9 +92,9 @@ def test_gpu_opt_dtypes():
f = function([p, u], m, allow_input_downcast=True, mode=mode_with_gpu) f = function([p, u], m, allow_input_downcast=True, mode=mode_with_gpu)
assert any([type(node.op) is GPUAMultinomialFromUniform assert any([type(node.op) is GPUAMultinomialFromUniform
for node in f.maker.fgraph.toposort()]) for node in f.maker.fgraph.toposort()])
pval = numpy.arange(10000 * 4, dtype='float32').reshape((10000, 4)) + 0.1 pval = np.arange(10000 * 4, dtype='float32').reshape((10000, 4)) + 0.1
pval = pval / pval.sum(axis=1)[:, None] pval = pval / pval.sum(axis=1)[:, None]
uval = numpy.ones_like(pval[:, 0]) * 0.5 uval = np.ones_like(pval[:, 0]) * 0.5
samples = f(pval, uval) samples = f(pval, uval)
assert samples.dtype == dtype, "%s != %s" % (samples.dtype, dtype) assert samples.dtype == dtype, "%s != %s" % (samples.dtype, dtype)
...@@ -112,9 +112,9 @@ def test_gpu_opt(): ...@@ -112,9 +112,9 @@ def test_gpu_opt():
f = function([p, u], m, allow_input_downcast=True, mode=mode_with_gpu) f = function([p, u], m, allow_input_downcast=True, mode=mode_with_gpu)
assert any([type(node.op) is GPUAMultinomialFromUniform assert any([type(node.op) is GPUAMultinomialFromUniform
for node in f.maker.fgraph.toposort()]) for node in f.maker.fgraph.toposort()])
pval = numpy.arange(10000 * 4, dtype='float32').reshape((10000, 4)) + 0.1 pval = np.arange(10000 * 4, dtype='float32').reshape((10000, 4)) + 0.1
pval = pval / pval.sum(axis=1)[:, None] pval = pval / pval.sum(axis=1)[:, None]
uval = numpy.ones_like(pval[:, 0]) * 0.5 uval = np.ones_like(pval[:, 0]) * 0.5
f(pval, uval) f(pval, uval)
# Test with a row, it was failing in the past. # Test with a row, it was failing in the past.
...@@ -125,9 +125,9 @@ def test_gpu_opt(): ...@@ -125,9 +125,9 @@ def test_gpu_opt():
f = function([r, u], m, allow_input_downcast=True, mode=mode_with_gpu) f = function([r, u], m, allow_input_downcast=True, mode=mode_with_gpu)
assert any([type(node.op) is GPUAMultinomialFromUniform assert any([type(node.op) is GPUAMultinomialFromUniform
for node in f.maker.fgraph.toposort()]) for node in f.maker.fgraph.toposort()])
pval = numpy.arange(1 * 4, dtype='float32').reshape((1, 4)) + 0.1 pval = np.arange(1 * 4, dtype='float32').reshape((1, 4)) + 0.1
pval = pval / pval.sum(axis=1)[:, None] pval = pval / pval.sum(axis=1)[:, None]
uval = numpy.ones_like(pval[:, 0]) * 0.5 uval = np.ones_like(pval[:, 0]) * 0.5
f(pval, uval) f(pval, uval)
...@@ -146,15 +146,15 @@ class test_OP_wor(unittest.TestCase): ...@@ -146,15 +146,15 @@ class test_OP_wor(unittest.TestCase):
n_elements = 1000 n_elements = 1000
all_indices = range(n_elements) all_indices = range(n_elements)
numpy.random.seed(12345) np.random.seed(12345)
for i in [5, 10, 50, 100, 500, n_elements]: for i in [5, 10, 50, 100, 500, n_elements]:
uni = numpy.random.rand(i).astype(config.floatX) uni = np.random.rand(i).astype(config.floatX)
pvals = numpy.random.randint(1, 100, (1, n_elements)).astype(config.floatX) pvals = np.random.randint(1, 100, (1, n_elements)).astype(config.floatX)
pvals /= pvals.sum(1) pvals /= pvals.sum(1)
res = f(pvals, uni, i) res = f(pvals, uni, i)
res = numpy.squeeze(res) res = np.squeeze(res)
assert len(res) == i, res assert len(res) == i, res
assert numpy.all(numpy.in1d(numpy.unique(res), all_indices)), res assert np.all(np.in1d(np.unique(res), all_indices)), res
def test_fail_select_alot(self): def test_fail_select_alot(self):
""" """
...@@ -170,9 +170,9 @@ class test_OP_wor(unittest.TestCase): ...@@ -170,9 +170,9 @@ class test_OP_wor(unittest.TestCase):
n_elements = 100 n_elements = 100
n_selected = 200 n_selected = 200
numpy.random.seed(12345) np.random.seed(12345)
uni = numpy.random.rand(n_selected).astype(config.floatX) uni = np.random.rand(n_selected).astype(config.floatX)
pvals = numpy.random.randint(1, 100, (1, n_elements)).astype(config.floatX) pvals = np.random.randint(1, 100, (1, n_elements)).astype(config.floatX)
pvals /= pvals.sum(1) pvals /= pvals.sum(1)
self.assertRaises(ValueError, f, pvals, uni, n_selected) self.assertRaises(ValueError, f, pvals, uni, n_selected)
...@@ -191,18 +191,18 @@ class test_OP_wor(unittest.TestCase): ...@@ -191,18 +191,18 @@ class test_OP_wor(unittest.TestCase):
n_elements = 100 n_elements = 100
n_selected = 10 n_selected = 10
mean_rtol = 0.0005 mean_rtol = 0.0005
numpy.random.seed(12345) np.random.seed(12345)
pvals = numpy.random.randint(1, 100, (1, n_elements)).astype(config.floatX) pvals = np.random.randint(1, 100, (1, n_elements)).astype(config.floatX)
pvals /= pvals.sum(1) pvals /= pvals.sum(1)
avg_pvals = numpy.zeros((n_elements,), dtype=config.floatX) avg_pvals = np.zeros((n_elements,), dtype=config.floatX)
for rep in range(10000): for rep in range(10000):
uni = numpy.random.rand(n_selected).astype(config.floatX) uni = np.random.rand(n_selected).astype(config.floatX)
res = f(pvals, uni, n_selected) res = f(pvals, uni, n_selected)
res = numpy.squeeze(res) res = np.squeeze(res)
avg_pvals[res] += 1 avg_pvals[res] += 1
avg_pvals /= avg_pvals.sum() avg_pvals /= avg_pvals.sum()
avg_diff = numpy.mean(abs(avg_pvals - pvals)) avg_diff = np.mean(abs(avg_pvals - pvals))
assert avg_diff < mean_rtol, avg_diff assert avg_diff < mean_rtol, avg_diff
...@@ -222,14 +222,14 @@ class test_function_wor(unittest.TestCase): ...@@ -222,14 +222,14 @@ class test_function_wor(unittest.TestCase):
n_elements = 1000 n_elements = 1000
all_indices = range(n_elements) all_indices = range(n_elements)
numpy.random.seed(12345) np.random.seed(12345)
for i in [5, 10, 50, 100, 500, n_elements]: for i in [5, 10, 50, 100, 500, n_elements]:
pvals = numpy.random.randint(1, 100, (1, n_elements)).astype(config.floatX) pvals = np.random.randint(1, 100, (1, n_elements)).astype(config.floatX)
pvals /= pvals.sum(1) pvals /= pvals.sum(1)
res = f(pvals, i) res = f(pvals, i)
res = numpy.squeeze(res) res = np.squeeze(res)
assert len(res) == i assert len(res) == i
assert numpy.all(numpy.in1d(numpy.unique(res), all_indices)), res assert np.all(np.in1d(np.unique(res), all_indices)), res
def test_fail_select_alot(self): def test_fail_select_alot(self):
""" """
...@@ -246,8 +246,8 @@ class test_function_wor(unittest.TestCase): ...@@ -246,8 +246,8 @@ class test_function_wor(unittest.TestCase):
n_elements = 100 n_elements = 100
n_selected = 200 n_selected = 200
numpy.random.seed(12345) np.random.seed(12345)
pvals = numpy.random.randint(1, 100, (1, n_elements)).astype(config.floatX) pvals = np.random.randint(1, 100, (1, n_elements)).astype(config.floatX)
pvals /= pvals.sum(1) pvals /= pvals.sum(1)
self.assertRaises(ValueError, f, pvals, n_selected) self.assertRaises(ValueError, f, pvals, n_selected)
...@@ -267,17 +267,17 @@ class test_function_wor(unittest.TestCase): ...@@ -267,17 +267,17 @@ class test_function_wor(unittest.TestCase):
n_elements = 100 n_elements = 100
n_selected = 10 n_selected = 10
mean_rtol = 0.0005 mean_rtol = 0.0005
numpy.random.seed(12345) np.random.seed(12345)
pvals = numpy.random.randint(1, 100, (1, n_elements)).astype(config.floatX) pvals = np.random.randint(1, 100, (1, n_elements)).astype(config.floatX)
pvals /= pvals.sum(1) pvals /= pvals.sum(1)
avg_pvals = numpy.zeros((n_elements,), dtype=config.floatX) avg_pvals = np.zeros((n_elements,), dtype=config.floatX)
for rep in range(10000): for rep in range(10000):
res = f(pvals, n_selected) res = f(pvals, n_selected)
res = numpy.squeeze(res) res = np.squeeze(res)
avg_pvals[res] += 1 avg_pvals[res] += 1
avg_pvals /= avg_pvals.sum() avg_pvals /= avg_pvals.sum()
avg_diff = numpy.mean(abs(avg_pvals - pvals)) avg_diff = np.mean(abs(avg_pvals - pvals))
assert avg_diff < mean_rtol assert avg_diff < mean_rtol
...@@ -294,9 +294,9 @@ def test_gpu_opt_wor(): ...@@ -294,9 +294,9 @@ def test_gpu_opt_wor():
assert any([type(node.op) is GPUAMultinomialWOReplacementFromUniform assert any([type(node.op) is GPUAMultinomialWOReplacementFromUniform
for node in f.maker.fgraph.toposort()]) for node in f.maker.fgraph.toposort()])
n_samples = 3 n_samples = 3
pval = numpy.arange(10000 * 4, dtype='float32').reshape((10000, 4)) + 0.1 pval = np.arange(10000 * 4, dtype='float32').reshape((10000, 4)) + 0.1
pval = pval / pval.sum(axis=1)[:, None] pval = pval / pval.sum(axis=1)[:, None]
uval = numpy.ones(pval.shape[0] * n_samples) * 0.5 uval = np.ones(pval.shape[0] * n_samples) * 0.5
f(pval, uval, n_samples) f(pval, uval, n_samples)
# Test with a row, it was failing in the past. # Test with a row, it was failing in the past.
...@@ -307,7 +307,7 @@ def test_gpu_opt_wor(): ...@@ -307,7 +307,7 @@ def test_gpu_opt_wor():
f = function([r, u, n], m, allow_input_downcast=True, mode=mode_with_gpu) f = function([r, u, n], m, allow_input_downcast=True, mode=mode_with_gpu)
assert any([type(node.op) is GPUAMultinomialWOReplacementFromUniform assert any([type(node.op) is GPUAMultinomialWOReplacementFromUniform
for node in f.maker.fgraph.toposort()]) for node in f.maker.fgraph.toposort()])
pval = numpy.arange(1 * 4, dtype='float32').reshape((1, 4)) + 0.1 pval = np.arange(1 * 4, dtype='float32').reshape((1, 4)) + 0.1
pval = pval / pval.sum(axis=1)[:, None] pval = pval / pval.sum(axis=1)[:, None]
uval = numpy.ones_like(pval[:, 0]) * 0.5 uval = np.ones_like(pval[:, 0]) * 0.5
f(pval, uval, 1) f(pval, uval, 1)
theano/gpuarray/tests/test_nerv.py
浏览文件 @ 824cb369
from __future__ import absolute_import, print_function, division from __future__ import absolute_import, print_function, division
from nose.plugins.skip import SkipTest from nose.plugins.skip import SkipTest
import numpy import numpy as np
from theano import function from theano import function
from theano.tests import unittest_tools as utt from theano.tests import unittest_tools as utt
...@@ -40,10 +40,10 @@ def test_gemm16_value(): ...@@ -40,10 +40,10 @@ def test_gemm16_value():
f = function([m, m2], dot(m, m2), mode=mode_with_gpu) f = function([m, m2], dot(m, m2), mode=mode_with_gpu)
v1 = numpy.random.random((3, 4)).astype('float16') v1 = np.random.random((3, 4)).astype('float16')
v2 = numpy.random.random((4, 2)).astype('float16') v2 = np.random.random((4, 2)).astype('float16')
of = f(v1, v2) of = f(v1, v2)
on = numpy.dot(v1, v2) on = np.dot(v1, v2)
utt.assert_allclose(of, on) utt.assert_allclose(of, on)
theano/gpuarray/tests/test_nnet.py
浏览文件 @ 824cb369
from __future__ import absolute_import, print_function, division from __future__ import absolute_import, print_function, division
import numpy import numpy as np
import unittest import unittest
import theano import theano
...@@ -46,13 +46,13 @@ def test_GpuCrossentropySoftmaxArgmax1HotWithBias(): ...@@ -46,13 +46,13 @@ def test_GpuCrossentropySoftmaxArgmax1HotWithBias():
# Seed numpy.random with config.unittests.rseed # Seed numpy.random with config.unittests.rseed
utt.seed_rng() utt.seed_rng()
xx = numpy.asarray(numpy.random.rand(batch_size, n_in), xx = np.asarray(np.random.rand(batch_size, n_in),
dtype=numpy.float32) dtype=np.float32)
yy = numpy.ones((batch_size,), dtype='int32') yy = np.ones((batch_size,), dtype='int32')
b_values = numpy.zeros((n_out,), dtype='float32') b_values = np.zeros((n_out,), dtype='float32')
W_values = numpy.asarray(numpy.random.rand(n_in, n_out), dtype='float32') W_values = np.asarray(np.random.rand(n_in, n_out), dtype='float32')
dot_value = numpy.asarray(numpy.dot(xx, W_values), dtype='float32') dot_value = np.asarray(np.dot(xx, W_values), dtype='float32')
del W_values del W_values
p_y_given_x = T.nnet.softmax(dot_result + b) p_y_given_x = T.nnet.softmax(dot_result + b)
y_pred = T.argmax(p_y_given_x, axis=-1) y_pred = T.argmax(p_y_given_x, axis=-1)
...@@ -97,10 +97,10 @@ def test_GpuCrossentropySoftmax1HotWithBiasDx(): ...@@ -97,10 +97,10 @@ def test_GpuCrossentropySoftmax1HotWithBiasDx():
# Seed numpy.random with config.unittests.rseed # Seed numpy.random with config.unittests.rseed
utt.seed_rng() utt.seed_rng()
softmax_output_value = numpy.random.rand(batch_size, softmax_output_value = np.random.rand(batch_size,
n_out).astype('float32') n_out).astype('float32')
dnll_value = numpy.asarray(numpy.random.rand(batch_size), dtype='float32') dnll_value = np.asarray(np.random.rand(batch_size), dtype='float32')
y_idx_value = numpy.random.randint(low=0, high=5, size=batch_size) y_idx_value = np.random.randint(low=0, high=5, size=batch_size)
softmax_output = T.fmatrix() softmax_output = T.fmatrix()
softmax_output /= softmax_output.sum(axis=1).reshape( softmax_output /= softmax_output.sum(axis=1).reshape(
...@@ -174,8 +174,8 @@ def softmax_with_bias_unittest_template(dtypeInput, dtypeBias): ...@@ -174,8 +174,8 @@ def softmax_with_bias_unittest_template(dtypeInput, dtypeBias):
GpuSoftmaxWithBias) GpuSoftmaxWithBias)
def cmp(n, m): def cmp(n, m):
data = numpy.random.uniform(1e-7, 1, (n, m)).astype(dtype=dtypeInput) data = np.random.uniform(1e-7, 1, (n, m)).astype(dtype=dtypeInput)
b_data = numpy.random.uniform(1e-7, 1, (m,)).astype(dtype=dtypeBias) b_data = np.random.uniform(1e-7, 1, (m,)).astype(dtype=dtypeBias)
out = f(data, b_data) out = f(data, b_data)
gout = f_gpu(data, b_data) gout = f_gpu(data, b_data)
...@@ -227,7 +227,7 @@ def softmax_unittest_template(dtypeInput): ...@@ -227,7 +227,7 @@ def softmax_unittest_template(dtypeInput):
GpuSoftmax) GpuSoftmax)
def cmp(n, m): def cmp(n, m):
data = numpy.random.uniform(0, 1, (n, m)).astype(dtype=dtypeInput) data = np.random.uniform(0, 1, (n, m)).astype(dtype=dtypeInput)
out = f(data) out = f(data)
gout = f_gpu(data) gout = f_gpu(data)
...@@ -301,7 +301,7 @@ class test_SoftMax(unittest.TestCase): ...@@ -301,7 +301,7 @@ class test_SoftMax(unittest.TestCase):
return f, f_gpu return f, f_gpu
def _cmp(self, n, m, f, f_gpu): def _cmp(self, n, m, f, f_gpu):
data = numpy.arange(n * m, dtype='float32').reshape(n, m) data = np.arange(n * m, dtype='float32').reshape(n, m)
out = f(data) out = f(data)
gout = f_gpu(data) gout = f_gpu(data)
utt.assert_allclose(out, gout) utt.assert_allclose(out, gout)
......
theano/gpuarray/tests/test_opt.py
浏览文件 @ 824cb369
from __future__ import absolute_import, print_function, division from __future__ import absolute_import, print_function, division
import numpy import numpy as np
from nose.tools import assert_raises from nose.tools import assert_raises
import theano import theano
...@@ -79,13 +79,13 @@ def test_local_gpu_contiguous(): ...@@ -79,13 +79,13 @@ def test_local_gpu_contiguous():
def test_flatten(): def test_flatten():
m = theano.tensor.fmatrix() m = theano.tensor.fmatrix()
f = theano.function([m], m.flatten(), mode=mode_with_gpu) f = theano.function([m], m.flatten(), mode=mode_with_gpu)
val = numpy.random.rand(10, 11).astype("float32") val = np.random.rand(10, 11).astype("float32")
res = f(val) res = f(val)
utt.assert_allclose(res, val.flatten()) utt.assert_allclose(res, val.flatten())
assert res.shape == val.flatten().shape assert res.shape == val.flatten().shape
assert GpuReshape in [type(node.op) assert GpuReshape in [type(node.op)
for node in f.maker.fgraph.toposort()] for node in f.maker.fgraph.toposort()]
val = numpy.random.rand(10, 11).astype("float32") val = np.random.rand(10, 11).astype("float32")
res = f(val) res = f(val)
utt.assert_allclose(res, val.flatten()) utt.assert_allclose(res, val.flatten())
assert res.shape == val.flatten().shape assert res.shape == val.flatten().shape
...@@ -93,7 +93,7 @@ def test_flatten(): ...@@ -93,7 +93,7 @@ def test_flatten():
for node in f.maker.fgraph.toposort()] for node in f.maker.fgraph.toposort()]
f = theano.function([m], m.flatten(ndim=2), mode=mode_with_gpu) f = theano.function([m], m.flatten(ndim=2), mode=mode_with_gpu)
val = numpy.random.rand(10, 11).astype("float32") val = np.random.rand(10, 11).astype("float32")
res = f(val) res = f(val)
utt.assert_allclose(res, val) utt.assert_allclose(res, val)
assert res.shape == val.shape assert res.shape == val.shape
...@@ -102,7 +102,7 @@ def test_flatten(): ...@@ -102,7 +102,7 @@ def test_flatten():
m = theano.tensor.tensor3() m = theano.tensor.tensor3()
f = theano.function([m], m.flatten(ndim=2), mode=mode_with_gpu) f = theano.function([m], m.flatten(ndim=2), mode=mode_with_gpu)
val = numpy.random.rand(10, 11, 12).astype("float32") val = np.random.rand(10, 11, 12).astype("float32")
res = f(val) res = f(val)
utt.assert_allclose(res, val.reshape(10, -1)) utt.assert_allclose(res, val.reshape(10, -1))
assert res.shape == val.reshape(10, -1).shape assert res.shape == val.reshape(10, -1).shape
...@@ -120,7 +120,7 @@ def test_reduce(): ...@@ -120,7 +120,7 @@ def test_reduce():
f = theano.function([m], getattr(m, method)(axis=0, f = theano.function([m], getattr(m, method)(axis=0,
**param), **param),
mode=mode_with_gpu) mode=mode_with_gpu)
val = numpy.random.rand(10, 11).astype("float32") val = np.random.rand(10, 11).astype("float32")
res = f(val) res = f(val)
utt.assert_allclose(res, getattr(val, method)(axis=0)) utt.assert_allclose(res, getattr(val, method)(axis=0))
assert res.shape == (11,) assert res.shape == (11,)
...@@ -135,9 +135,9 @@ def test_reduce(): ...@@ -135,9 +135,9 @@ def test_reduce():
def test_local_gpualloc_memset_0(): def test_local_gpualloc_memset_0():
i = theano.tensor.iscalar() i = theano.tensor.iscalar()
z = numpy.zeros((1,), dtype='float32') z = np.zeros((1,), dtype='float32')
o = numpy.ones((1,), dtype='float32') o = np.ones((1,), dtype='float32')
ones = numpy.ones((2,), dtype='float32') ones = np.ones((2,), dtype='float32')
# Test with 0 from CPU op. # Test with 0 from CPU op.
# Should not be transfered as the only client is the output # Should not be transfered as the only client is the output
...@@ -146,7 +146,7 @@ def test_local_gpualloc_memset_0(): ...@@ -146,7 +146,7 @@ def test_local_gpualloc_memset_0():
topo = f.maker.fgraph.toposort() topo = f.maker.fgraph.toposort()
assert len(topo) == 1 assert len(topo) == 1
assert isinstance(topo[0].op, theano.tensor.Alloc) assert isinstance(topo[0].op, theano.tensor.Alloc)
assert (numpy.asarray(f(6)) == 0).all() assert (np.asarray(f(6)) == 0).all()
# Test with 0 from CPU op. # Test with 0 from CPU op.
# Should be transfered as it is used by another op. # Should be transfered as it is used by another op.
...@@ -155,7 +155,7 @@ def test_local_gpualloc_memset_0(): ...@@ -155,7 +155,7 @@ def test_local_gpualloc_memset_0():
topo = f.maker.fgraph.toposort() topo = f.maker.fgraph.toposort()
assert len(topo) == 3 assert len(topo) == 3
assert isinstance(topo[0].op, GpuAlloc) assert isinstance(topo[0].op, GpuAlloc)
assert (numpy.asarray(f(6)) == 0).all() assert (np.asarray(f(6)) == 0).all()
# Test with 0 # Test with 0
a = GpuAlloc(test_ctx_name)(z, i) a = GpuAlloc(test_ctx_name)(z, i)
...@@ -163,7 +163,7 @@ def test_local_gpualloc_memset_0(): ...@@ -163,7 +163,7 @@ def test_local_gpualloc_memset_0():
topo = f.maker.fgraph.toposort() topo = f.maker.fgraph.toposort()
assert len(topo) == 1 assert len(topo) == 1
assert isinstance(topo[0].op, GpuAlloc) and topo[0].op.memset_0 assert isinstance(topo[0].op, GpuAlloc) and topo[0].op.memset_0
assert (numpy.asarray(f(6)) == 0).all() assert (np.asarray(f(6)) == 0).all()
# Test with 1 # Test with 1
a = GpuAlloc(test_ctx_name)(o, i) a = GpuAlloc(test_ctx_name)(o, i)
...@@ -172,7 +172,7 @@ def test_local_gpualloc_memset_0(): ...@@ -172,7 +172,7 @@ def test_local_gpualloc_memset_0():
assert len(topo) == 1 assert len(topo) == 1
assert isinstance(topo[0].op, GpuAlloc) assert isinstance(topo[0].op, GpuAlloc)
assert not topo[0].op.memset_0 assert not topo[0].op.memset_0
assert (numpy.asarray(f(6)) == 1).all() assert (np.asarray(f(6)) == 1).all()
# Test with 1, 1 # Test with 1, 1
a = GpuAlloc(test_ctx_name)(ones, i) a = GpuAlloc(test_ctx_name)(ones, i)
...@@ -181,7 +181,7 @@ def test_local_gpualloc_memset_0(): ...@@ -181,7 +181,7 @@ def test_local_gpualloc_memset_0():
assert len(topo) == 1 assert len(topo) == 1
assert isinstance(topo[0].op, GpuAlloc) assert isinstance(topo[0].op, GpuAlloc)
assert not topo[0].op.memset_0 assert not topo[0].op.memset_0
assert (numpy.asarray(f(2)) == 1).all() assert (np.asarray(f(2)) == 1).all()
def test_local_gpualloc_empty(): def test_local_gpualloc_empty():
...@@ -219,7 +219,7 @@ def test_local_gpualloc_empty(): ...@@ -219,7 +219,7 @@ def test_local_gpualloc_empty():
def test_rebroadcast(): def test_rebroadcast():
d = numpy.random.rand(10, 10).astype('float32') d = np.random.rand(10, 10).astype('float32')
v = theano.tensor.fmatrix() v = theano.tensor.fmatrix()
up = tensor.unbroadcast(v.sum().dimshuffle('x', 'x'), 0, 1) up = tensor.unbroadcast(v.sum().dimshuffle('x', 'x'), 0, 1)
f = theano.function([v], [up], mode=mode_with_gpu) f = theano.function([v], [up], mode=mode_with_gpu)
...@@ -257,14 +257,14 @@ class test_gpu_ifelse(test_ifelse.test_ifelse): ...@@ -257,14 +257,14 @@ class test_gpu_ifelse(test_ifelse.test_ifelse):
f = theano.function([x, cond], f = theano.function([x, cond],
theano.ifelse.ifelse(cond, x.mean(), x.sum()), theano.ifelse.ifelse(cond, x.mean(), x.sum()),
mode=mode_with_gpu) mode=mode_with_gpu)
assert f(numpy.float32([1, 2, 3]), 0) == 6 assert f(np.float32([1, 2, 3]), 0) == 6
x = tensor.vector() x = tensor.vector()
cond = tensor.scalar() cond = tensor.scalar()
f = theano.function([x, cond], f = theano.function([x, cond],
theano.ifelse.ifelse(cond, x.mean(), x.sum()), theano.ifelse.ifelse(cond, x.mean(), x.sum()),
mode=mode_with_gpu) mode=mode_with_gpu)
assert f(numpy.float32([1, 2, 3]), 0) == 6 assert f(np.float32([1, 2, 3]), 0) == 6
def test_print_op(): def test_print_op():
...@@ -277,7 +277,7 @@ def test_print_op(): ...@@ -277,7 +277,7 @@ def test_print_op():
assert isinstance(topo[1].op, theano.printing.Print) assert isinstance(topo[1].op, theano.printing.Print)
assert isinstance(topo[2].op, GpuElemwise) assert isinstance(topo[2].op, GpuElemwise)
assert topo[3].op == host_from_gpu assert topo[3].op == host_from_gpu
f(numpy.random.random((5, 5)).astype('float32')) f(np.random.random((5, 5)).astype('float32'))
def test_pdbbreakpoint_op(): def test_pdbbreakpoint_op():
...@@ -306,7 +306,7 @@ def test_local_gpu_elemwise_careduce(): ...@@ -306,7 +306,7 @@ def test_local_gpu_elemwise_careduce():
topo = f.maker.fgraph.toposort() topo = f.maker.fgraph.toposort()
assert len(topo) == 3 assert len(topo) == 3
assert topo[1].op.pre_scalar_op == theano.scalar.sqr assert topo[1].op.pre_scalar_op == theano.scalar.sqr
data = numpy.random.rand(3, 4).astype(theano.config.floatX) data = np.random.rand(3, 4).astype(theano.config.floatX)
utt.assert_allclose(f(data), (data * data).sum()) utt.assert_allclose(f(data), (data * data).sum())
o = (x * x).sum(axis=1) o = (x * x).sum(axis=1)
...@@ -328,15 +328,15 @@ def test_local_lift_dot22scalar(): ...@@ -328,15 +328,15 @@ def test_local_lift_dot22scalar():
for n in f_gpu.maker.fgraph.apply_nodes) for n in f_gpu.maker.fgraph.apply_nodes)
assert any(isinstance(n.op, GpuGemm) assert any(isinstance(n.op, GpuGemm)
for n in f_gpu.maker.fgraph.apply_nodes) for n in f_gpu.maker.fgraph.apply_nodes)
x_val = numpy.random.random((2, 3)).astype(theano.config.floatX) x_val = np.random.random((2, 3)).astype(theano.config.floatX)
y_val = numpy.random.random((3, 4)).astype(theano.config.floatX) y_val = np.random.random((3, 4)).astype(theano.config.floatX)
a_val = 0.5 a_val = 0.5
utt.assert_allclose(f_cpu(x_val, y_val, a_val), f_gpu(x_val, y_val, a_val)) utt.assert_allclose(f_cpu(x_val, y_val, a_val), f_gpu(x_val, y_val, a_val))
def test_local_gpu_subtensor(): def test_local_gpu_subtensor():
# Test shared forced on CPU. # Test shared forced on CPU.
t = tensor._shared(numpy.zeros(20, "float32")) t = tensor._shared(np.zeros(20, "float32"))
f = theano.function([], t[3:4], mode=mode_with_gpu) f = theano.function([], t[3:4], mode=mode_with_gpu)
topo = f.maker.fgraph.toposort() topo = f.maker.fgraph.toposort()
assert any([type(node.op) is tensor.Subtensor for node in topo]) assert any([type(node.op) is tensor.Subtensor for node in topo])
...@@ -367,7 +367,7 @@ def test_local_gpu_subtensor(): ...@@ -367,7 +367,7 @@ def test_local_gpu_subtensor():
# Test shared forced on CPU end we do computation on the output of # Test shared forced on CPU end we do computation on the output of
# the subtensor. # the subtensor.
t = tensor._shared(numpy.zeros(20, "float32")) t = tensor._shared(np.zeros(20, "float32"))
f = theano.function([], t[3:4] + 1, mode=mode_with_gpu) f = theano.function([], t[3:4] + 1, mode=mode_with_gpu)
topo = f.maker.fgraph.toposort() topo = f.maker.fgraph.toposort()
assert any([type(node.op) is tensor.Subtensor for node in topo]) assert any([type(node.op) is tensor.Subtensor for node in topo])
...@@ -386,9 +386,9 @@ def test_local_gpu_elemwise(): ...@@ -386,9 +386,9 @@ def test_local_gpu_elemwise():
b = tensor.fmatrix() b = tensor.fmatrix()
c = tensor.fmatrix() c = tensor.fmatrix()
a_v = (numpy.random.rand(4, 5) * 10).astype("int8") a_v = (np.random.rand(4, 5) * 10).astype("int8")
b_v = (numpy.random.rand(4, 5) * 10).astype("float32") b_v = (np.random.rand(4, 5) * 10).astype("float32")
c_v = (numpy.random.rand(4, 5) * 10).astype("float32") c_v = (np.random.rand(4, 5) * 10).astype("float32")
# Due to optimization order, this composite is created when all # Due to optimization order, this composite is created when all
# the op are on the gpu. # the op are on the gpu.
...@@ -440,7 +440,7 @@ def test_local_gpu_elemwise(): ...@@ -440,7 +440,7 @@ def test_local_gpu_elemwise():
utt.assert_allclose(out[1], a_v * c_v) utt.assert_allclose(out[1], a_v * c_v)
# Test non-contiguous input # Test non-contiguous input
c = gpuarray_shared_constructor(numpy.asarray(c_v, dtype='float32')) c = gpuarray_shared_constructor(np.asarray(c_v, dtype='float32'))
f = theano.function([a, b], outs_op(a[::2], b[::2], c[::2]), f = theano.function([a, b], outs_op(a[::2], b[::2], c[::2]),
mode=mode_with_gpu) mode=mode_with_gpu)
out = f(a_v, b_v) out = f(a_v, b_v)
...@@ -462,7 +462,7 @@ def test_local_lift_abstractconv_gpu_shape(): ...@@ -462,7 +462,7 @@ def test_local_lift_abstractconv_gpu_shape():
def test_local_assert_no_cpu_op(): def test_local_assert_no_cpu_op():
rng = numpy.random.RandomState(utt.fetch_seed()) rng = np.random.RandomState(utt.fetch_seed())
m = rng.uniform(-1, 1, (10, 10)).astype("float32") m = rng.uniform(-1, 1, (10, 10)).astype("float32")
ms = gpuarray_shared_constructor(m, name="m_shared") ms = gpuarray_shared_constructor(m, name="m_shared")
out = theano.tensor.tanh(ms).dot(ms.T) out = theano.tensor.tanh(ms).dot(ms.T)
...@@ -512,6 +512,6 @@ def test_local_lift_solve(): ...@@ -512,6 +512,6 @@ def test_local_lift_solve():
for n in f_gpu.maker.fgraph.apply_nodes) for n in f_gpu.maker.fgraph.apply_nodes)
assert any(isinstance(n.op, GpuCusolverSolve) assert any(isinstance(n.op, GpuCusolverSolve)
for n in f_gpu.maker.fgraph.apply_nodes) for n in f_gpu.maker.fgraph.apply_nodes)
A_val = numpy.random.uniform(-0.4, 0.4, (5, 5)).astype("float32") A_val = np.random.uniform(-0.4, 0.4, (5, 5)).astype("float32")
b_val = numpy.random.uniform(-0.4, 0.4, (5, 3)).astype("float32") b_val = np.random.uniform(-0.4, 0.4, (5, 3)).astype("float32")
utt.assert_allclose(f_cpu(A_val, b_val), f_gpu(A_val, b_val)) utt.assert_allclose(f_cpu(A_val, b_val), f_gpu(A_val, b_val))
theano/gpuarray/tests/test_pickle.py
浏览文件 @ 824cb369
...@@ -12,7 +12,7 @@ from six import reraise ...@@ -12,7 +12,7 @@ from six import reraise
from nose.plugins.skip import SkipTest from nose.plugins.skip import SkipTest
from nose.tools import assert_raises from nose.tools import assert_raises
import numpy import numpy as np
from theano.compat import PY3 from theano.compat import PY3
from theano import config from theano import config
...@@ -75,7 +75,7 @@ def test_unpickle_gpuarray_as_numpy_ndarray_flag2(): ...@@ -75,7 +75,7 @@ def test_unpickle_gpuarray_as_numpy_ndarray_flag2():
reraise(SkipTest, exc_value, exc_trace) reraise(SkipTest, exc_value, exc_trace)
raise raise
assert isinstance(mat, numpy.ndarray) assert isinstance(mat, np.ndarray)
assert mat[0] == -42.0 assert mat[0] == -42.0
finally: finally:
......
theano/gpuarray/tests/test_pool.py
浏览文件 @ 824cb369
...@@ -4,7 +4,7 @@ import unittest ...@@ -4,7 +4,7 @@ import unittest
import copy import copy
import itertools import itertools
import numpy import numpy as np
import theano import theano
from theano import gradient from theano import gradient
from theano import tensor from theano import tensor
...@@ -81,7 +81,7 @@ def test_pool2d(): ...@@ -81,7 +81,7 @@ def test_pool2d():
(3, 2, 6, 6, 6, 5), (3, 2, 6, 6, 6, 5),
(3, 2, 6, 6, 6, 5, 7), ] (3, 2, 6, 6, 6, 5, 7), ]
numpy.random.RandomState(utt.fetch_seed()).shuffle(shps) np.random.RandomState(utt.fetch_seed()).shuffle(shps)
test_ws = (2, 2), (3, 2), (1, 1) test_ws = (2, 2), (3, 2), (1, 1)
test_st = (2, 2), (3, 2), (1, 1) test_st = (2, 2), (3, 2), (1, 1)
test_mode = ['max', 'sum', 'average_inc_pad', 'average_exc_pad'] test_mode = ['max', 'sum', 'average_inc_pad', 'average_exc_pad']
...@@ -113,7 +113,7 @@ def test_pool2d(): ...@@ -113,7 +113,7 @@ def test_pool2d():
for node in f.maker.fgraph.toposort()]) for node in f.maker.fgraph.toposort()])
assert any([isinstance(node.op, Pool) assert any([isinstance(node.op, Pool)
for node in f2.maker.fgraph.toposort()]) for node in f2.maker.fgraph.toposort()])
assert numpy.allclose(f(), f2()), (shp, ws, st, pad, mode, ignore_border) assert np.allclose(f(), f2()), (shp, ws, st, pad, mode, ignore_border)
a_pooled_grad = tensor.grad(a_pooled.sum(), a) a_pooled_grad = tensor.grad(a_pooled.sum(), a)
...@@ -131,7 +131,7 @@ def test_pool2d(): ...@@ -131,7 +131,7 @@ def test_pool2d():
assert any([isinstance(node.op, gop2) assert any([isinstance(node.op, gop2)
for node in g2.maker.fgraph.toposort()]) for node in g2.maker.fgraph.toposort()])
assert numpy.allclose(g(), g2()), (shp, ws, st, pad, mode, ignore_border) assert np.allclose(g(), g2()), (shp, ws, st, pad, mode, ignore_border)
# test rop and grad grad for max pooling # test rop and grad grad for max pooling
# for average pooling grad grad is just average pooling grad # for average pooling grad grad is just average pooling grad
...@@ -151,7 +151,7 @@ def test_pool2d(): ...@@ -151,7 +151,7 @@ def test_pool2d():
isinstance(node.op, DownsampleFactorMaxGradGrad) isinstance(node.op, DownsampleFactorMaxGradGrad)
for node in gr2.maker.fgraph.toposort() for node in gr2.maker.fgraph.toposort()
]) ])
assert numpy.allclose(gr(), gr2()), (shp, ws, st, pad, mode, ignore_border) assert np.allclose(gr(), gr2()), (shp, ws, st, pad, mode, ignore_border)
ggf = gradient.Lop(tensor.grad((a_pooled**2).sum(), a), a, a) ggf = gradient.Lop(tensor.grad((a_pooled**2).sum(), a), a, a)
...@@ -166,7 +166,7 @@ def test_pool2d(): ...@@ -166,7 +166,7 @@ def test_pool2d():
isinstance(node.op, DownsampleFactorMaxGradGrad) isinstance(node.op, DownsampleFactorMaxGradGrad)
for node in gg2.maker.fgraph.toposort() for node in gg2.maker.fgraph.toposort()
]) ])
assert numpy.allclose(gg(), gg2()), (shp, ws, st, pad, mode, ignore_border) assert np.allclose(gg(), gg2()), (shp, ws, st, pad, mode, ignore_border)
def test_pool3d(): def test_pool3d():
...@@ -191,7 +191,7 @@ def test_pool3d(): ...@@ -191,7 +191,7 @@ def test_pool3d():
(3, 2, 6, 6, 6, 5), (3, 2, 6, 6, 6, 5),
(3, 2, 6, 6, 6, 5, 7), ] (3, 2, 6, 6, 6, 5, 7), ]
numpy.random.RandomState(utt.fetch_seed()).shuffle(shps) np.random.RandomState(utt.fetch_seed()).shuffle(shps)
test_ws = (2, 2, 2), (3, 2, 3), (1, 1, 1) test_ws = (2, 2, 2), (3, 2, 3), (1, 1, 1)
test_st = (2, 2, 2), (2, 3, 2), (1, 1, 1) test_st = (2, 2, 2), (2, 3, 2), (1, 1, 1)
test_mode = ['max', 'sum', 'average_inc_pad', 'average_exc_pad'] test_mode = ['max', 'sum', 'average_inc_pad', 'average_exc_pad']
...@@ -223,7 +223,7 @@ def test_pool3d(): ...@@ -223,7 +223,7 @@ def test_pool3d():
for node in f.maker.fgraph.toposort()]) for node in f.maker.fgraph.toposort()])
assert any([isinstance(node.op, Pool) assert any([isinstance(node.op, Pool)
for node in f2.maker.fgraph.toposort()]) for node in f2.maker.fgraph.toposort()])
assert numpy.allclose(f(), f2()), (shp, ws, st, pad, mode, ignore_border) assert np.allclose(f(), f2()), (shp, ws, st, pad, mode, ignore_border)
a_pooled_grad = tensor.grad(a_pooled.sum(), a) a_pooled_grad = tensor.grad(a_pooled.sum(), a)
...@@ -241,7 +241,7 @@ def test_pool3d(): ...@@ -241,7 +241,7 @@ def test_pool3d():
assert any([isinstance(node.op, gop2) assert any([isinstance(node.op, gop2)
for node in g2.maker.fgraph.toposort()]) for node in g2.maker.fgraph.toposort()])
assert numpy.allclose(g(), g2()), (shp, ws, st, pad, mode, ignore_border) assert np.allclose(g(), g2()), (shp, ws, st, pad, mode, ignore_border)
# test rop and grad grad for max pooling # test rop and grad grad for max pooling
# for average pooling grad grad is just average pooling grad # for average pooling grad grad is just average pooling grad
...@@ -261,7 +261,7 @@ def test_pool3d(): ...@@ -261,7 +261,7 @@ def test_pool3d():
isinstance(node.op, DownsampleFactorMaxGradGrad) isinstance(node.op, DownsampleFactorMaxGradGrad)
for node in gr2.maker.fgraph.toposort() for node in gr2.maker.fgraph.toposort()
]) ])
assert numpy.allclose(gr(), gr2()), (shp, ws, st, pad, mode, ignore_border) assert np.allclose(gr(), gr2()), (shp, ws, st, pad, mode, ignore_border)
ggf = gradient.Lop(tensor.grad((a_pooled**2).sum(), a), a, a) ggf = gradient.Lop(tensor.grad((a_pooled**2).sum(), a), a, a)
...@@ -276,4 +276,4 @@ def test_pool3d(): ...@@ -276,4 +276,4 @@ def test_pool3d():
isinstance(node.op, DownsampleFactorMaxGradGrad) isinstance(node.op, DownsampleFactorMaxGradGrad)
for node in gg2.maker.fgraph.toposort() for node in gg2.maker.fgraph.toposort()
]) ])
assert numpy.allclose(gg(), gg2()), (shp, ws, st, pad, mode, ignore_border) assert np.allclose(gg(), gg2()), (shp, ws, st, pad, mode, ignore_border)
theano/gpuarray/tests/test_scan.py
浏览文件 @ 824cb369
from __future__ import absolute_import, print_function, division from __future__ import absolute_import, print_function, division
from unittest import TestCase from unittest import TestCase
import numpy import numpy as np
from six.moves import xrange from six.moves import xrange
import theano import theano
...@@ -43,19 +43,19 @@ class T_Scan(TestCase): ...@@ -43,19 +43,19 @@ class T_Scan(TestCase):
allow_input_downcast=True, allow_input_downcast=True,
mode=mode) mode=mode)
rng = numpy.random.RandomState(utt.fetch_seed()) rng = np.random.RandomState(utt.fetch_seed())
v_u = rng.uniform(size=(4,), low=-5., high=5.) v_u = rng.uniform(size=(4,), low=-5., high=5.)
v_x0 = rng.uniform() v_x0 = rng.uniform()
W = rng.uniform() W = rng.uniform()
W_in = rng.uniform() W_in = rng.uniform()
v_u = numpy.asarray(v_u, dtype='float32') v_u = np.asarray(v_u, dtype='float32')
v_x0 = numpy.asarray(v_x0, dtype='float32') v_x0 = np.asarray(v_x0, dtype='float32')
W = numpy.asarray(W, dtype='float32') W = np.asarray(W, dtype='float32')
W_in = numpy.asarray(W_in, dtype='float32') W_in = np.asarray(W_in, dtype='float32')
# compute the output in numpy # compute the output in numpy
v_out = numpy.zeros((4,)) v_out = np.zeros((4,))
v_out[0] = v_u[0] * W_in + v_x0 * W v_out[0] = v_u[0] * W_in + v_x0 * W
for step in xrange(1, 4): for step in xrange(1, 4):
v_out[step] = v_u[step] * W_in + v_out[step - 1] * W v_out[step] = v_u[step] * W_in + v_out[step - 1] * W
...@@ -115,14 +115,14 @@ class T_Scan(TestCase): ...@@ -115,14 +115,14 @@ class T_Scan(TestCase):
mode=mode_with_gpu) mode=mode_with_gpu)
# get random initial values # get random initial values
rng = numpy.random.RandomState(utt.fetch_seed()) rng = np.random.RandomState(utt.fetch_seed())
v_u = rng.uniform(size=(4,), low=-5., high=5.) v_u = rng.uniform(size=(4,), low=-5., high=5.)
v_x0 = rng.uniform() v_x0 = rng.uniform()
W = rng.uniform() W = rng.uniform()
W_in = rng.uniform() W_in = rng.uniform()
# compute the output in numpy # compute the output in numpy
v_out = numpy.zeros((4,)) v_out = np.zeros((4,))
v_out[0] = v_u[0] * W_in + v_x0 * W v_out[0] = v_u[0] * W_in + v_x0 * W
for step in xrange(1, 4): for step in xrange(1, 4):
v_out[step] = v_u[step] * W_in + v_out[step - 1] * W v_out[step] = v_u[step] * W_in + v_out[step - 1] * W
...@@ -177,20 +177,20 @@ class T_Scan(TestCase): ...@@ -177,20 +177,20 @@ class T_Scan(TestCase):
mode=mode_with_gpu) mode=mode_with_gpu)
# get random initial values # get random initial values
rng = numpy.random.RandomState(utt.fetch_seed()) rng = np.random.RandomState(utt.fetch_seed())
v_u = rng.uniform(size=(4,), low=-5., high=5.) v_u = rng.uniform(size=(4,), low=-5., high=5.)
v_x0 = rng.uniform() v_x0 = rng.uniform()
W = rng.uniform() W = rng.uniform()
W_in = rng.uniform() W_in = rng.uniform()
# compute the output in numpy # compute the output in numpy
v_out1 = numpy.zeros((4,)) v_out1 = np.zeros((4,))
v_out2 = numpy.zeros((4,), dtype='int64') v_out2 = np.zeros((4,), dtype='int64')
v_out1[0] = v_u[0] * W_in + v_x0 * W v_out1[0] = v_u[0] * W_in + v_x0 * W
v_out2[0] = v_u[0] + v_x0 v_out2[0] = v_u[0] + v_x0
for step in xrange(1, 4): for step in xrange(1, 4):
v_out1[step] = v_u[step] * W_in + v_out1[step - 1] * W v_out1[step] = v_u[step] * W_in + v_out1[step - 1] * W
v_out2[step] = numpy.int64(v_u[step] + v_out1[step - 1]) v_out2[step] = np.int64(v_u[step] + v_out1[step - 1])
theano_out1, theano_out2 = f2(v_u, v_x0, W_in, W) theano_out1, theano_out2 = f2(v_u, v_x0, W_in, W)
utt.assert_allclose(theano_out1, v_out1) utt.assert_allclose(theano_out1, v_out1)
...@@ -212,9 +212,9 @@ class T_Scan(TestCase): ...@@ -212,9 +212,9 @@ class T_Scan(TestCase):
for node in scan_node_topo]) for node in scan_node_topo])
def test_gpu4_gibbs_chain(self): def test_gpu4_gibbs_chain(self):
rng = numpy.random.RandomState(utt.fetch_seed()) rng = np.random.RandomState(utt.fetch_seed())
v_vsample = numpy.array(rng.binomial(1, .5, size=(3, 20),), v_vsample = np.array(rng.binomial(1, .5, size=(3, 20),),
dtype='float32') dtype='float32')
vsample = theano.shared(v_vsample) vsample = theano.shared(v_vsample)
trng = theano.sandbox.rng_mrg.MRG_RandomStreams( trng = theano.sandbox.rng_mrg.MRG_RandomStreams(
utt.fetch_seed()) utt.fetch_seed())
......
theano/gpuarray/tests/test_subtensor.py
浏览文件 @ 824cb369
from __future__ import absolute_import, print_function, division from __future__ import absolute_import, print_function, division
import numpy import numpy as np
import unittest import unittest
import theano import theano
...@@ -47,8 +47,8 @@ def test_advinc_subtensor1(): ...@@ -47,8 +47,8 @@ def test_advinc_subtensor1():
# Test the second case in the opt local_gpu_advanced_incsubtensor1 # Test the second case in the opt local_gpu_advanced_incsubtensor1
for shp in [(3, 3), (3, 3, 3)]: for shp in [(3, 3), (3, 3, 3)]:
shared = gpuarray_shared_constructor shared = gpuarray_shared_constructor
xval = numpy.arange(numpy.prod(shp), dtype='float32').reshape(shp) + 1 xval = np.arange(np.prod(shp), dtype='float32').reshape(shp) + 1
yval = numpy.empty((2,) + shp[1:], dtype='float32') yval = np.empty((2,) + shp[1:], dtype='float32')
yval[:] = 10 yval[:] = 10
x = shared(xval, name='x') x = shared(xval, name='x')
y = tensor.tensor(dtype='float32', y = tensor.tensor(dtype='float32',
...@@ -61,7 +61,7 @@ def test_advinc_subtensor1(): ...@@ -61,7 +61,7 @@ def test_advinc_subtensor1():
rval = f(yval) rval = f(yval)
rep = xval.copy() rep = xval.copy()
rep[[0, 2]] += yval rep[[0, 2]] += yval
assert numpy.allclose(rval, rep) assert np.allclose(rval, rep)
def test_advinc_subtensor1_dtype(): def test_advinc_subtensor1_dtype():
...@@ -69,8 +69,8 @@ def test_advinc_subtensor1_dtype(): ...@@ -69,8 +69,8 @@ def test_advinc_subtensor1_dtype():
shp = (3, 4) shp = (3, 4)
for dtype1, dtype2 in [('float32', 'int8'), ('float32', 'float64')]: for dtype1, dtype2 in [('float32', 'int8'), ('float32', 'float64')]:
shared = gpuarray_shared_constructor shared = gpuarray_shared_constructor
xval = numpy.arange(numpy.prod(shp), dtype=dtype1).reshape(shp) + 1 xval = np.arange(np.prod(shp), dtype=dtype1).reshape(shp) + 1
yval = numpy.empty((2,) + shp[1:], dtype=dtype2) yval = np.empty((2,) + shp[1:], dtype=dtype2)
yval[:] = 10 yval[:] = 10
x = shared(xval, name='x') x = shared(xval, name='x')
y = tensor.tensor(dtype=yval.dtype, y = tensor.tensor(dtype=yval.dtype,
...@@ -83,7 +83,7 @@ def test_advinc_subtensor1_dtype(): ...@@ -83,7 +83,7 @@ def test_advinc_subtensor1_dtype():
rval = f(yval) rval = f(yval)
rep = xval.copy() rep = xval.copy()
rep[[0, 2]] += yval rep[[0, 2]] += yval
assert numpy.allclose(rval, rep) assert np.allclose(rval, rep)
def test_advinc_subtensor1_vector_scalar(): def test_advinc_subtensor1_vector_scalar():
...@@ -91,8 +91,8 @@ def test_advinc_subtensor1_vector_scalar(): ...@@ -91,8 +91,8 @@ def test_advinc_subtensor1_vector_scalar():
shp = (3,) shp = (3,)
for dtype1, dtype2 in [('float32', 'int8'), ('float32', 'float64')]: for dtype1, dtype2 in [('float32', 'int8'), ('float32', 'float64')]:
shared = gpuarray_shared_constructor shared = gpuarray_shared_constructor
xval = numpy.arange(numpy.prod(shp), dtype=dtype1).reshape(shp) + 1 xval = np.arange(np.prod(shp), dtype=dtype1).reshape(shp) + 1
yval = numpy.asarray(10, dtype=dtype2) yval = np.asarray(10, dtype=dtype2)
x = shared(xval, name='x') x = shared(xval, name='x')
y = tensor.tensor(dtype=yval.dtype, y = tensor.tensor(dtype=yval.dtype,
broadcastable=(False,) * len(yval.shape), broadcastable=(False,) * len(yval.shape),
...@@ -104,14 +104,14 @@ def test_advinc_subtensor1_vector_scalar(): ...@@ -104,14 +104,14 @@ def test_advinc_subtensor1_vector_scalar():
rval = f(yval) rval = f(yval)
rep = xval.copy() rep = xval.copy()
rep[[0, 2]] += yval rep[[0, 2]] += yval
assert numpy.allclose(rval, rep) assert np.allclose(rval, rep)
def test_incsub_f16(): def test_incsub_f16():
shp = (3, 3) shp = (3, 3)
shared = gpuarray_shared_constructor shared = gpuarray_shared_constructor
xval = numpy.arange(numpy.prod(shp), dtype='float16').reshape(shp) + 1 xval = np.arange(np.prod(shp), dtype='float16').reshape(shp) + 1
yval = numpy.empty((2,) + shp[1:], dtype='float16') yval = np.empty((2,) + shp[1:], dtype='float16')
yval[:] = 2 yval[:] = 2
x = shared(xval, name='x') x = shared(xval, name='x')
y = tensor.tensor(dtype='float16', y = tensor.tensor(dtype='float16',
...@@ -124,7 +124,7 @@ def test_incsub_f16(): ...@@ -124,7 +124,7 @@ def test_incsub_f16():
rval = f(yval) rval = f(yval)
rep = xval.copy() rep = xval.copy()
rep[[0, 2]] += yval rep[[0, 2]] += yval
assert numpy.allclose(rval, rep) assert np.allclose(rval, rep)
expr = tensor.inc_subtensor(x[1:], y) expr = tensor.inc_subtensor(x[1:], y)
f = theano.function([y], expr, mode=mode_with_gpu) f = theano.function([y], expr, mode=mode_with_gpu)
...@@ -133,7 +133,7 @@ def test_incsub_f16(): ...@@ -133,7 +133,7 @@ def test_incsub_f16():
rval = f(yval) rval = f(yval)
rep = xval.copy() rep = xval.copy()
rep[1:] += yval rep[1:] += yval
assert numpy.allclose(rval, rep) assert np.allclose(rval, rep)
class G_advancedsubtensor(test_subtensor.TestAdvancedSubtensor): class G_advancedsubtensor(test_subtensor.TestAdvancedSubtensor):
...@@ -159,7 +159,7 @@ def test_adv_subtensor(): ...@@ -159,7 +159,7 @@ def test_adv_subtensor():
# Test the advancedsubtensor on gpu. # Test the advancedsubtensor on gpu.
shp = (2, 3, 4) shp = (2, 3, 4)
shared = gpuarray_shared_constructor shared = gpuarray_shared_constructor
xval = numpy.arange(numpy.prod(shp), dtype=theano.config.floatX).reshape(shp) xval = np.arange(np.prod(shp), dtype=theano.config.floatX).reshape(shp)
idx1, idx2 = tensor.ivectors('idx1', 'idx2') idx1, idx2 = tensor.ivectors('idx1', 'idx2')
idxs = [idx1, None, slice(0, 2, 1), idx2, None] idxs = [idx1, None, slice(0, 2, 1), idx2, None]
x = shared(xval, name='x') x = shared(xval, name='x')
...@@ -171,27 +171,27 @@ def test_adv_subtensor(): ...@@ -171,27 +171,27 @@ def test_adv_subtensor():
idx2_val = [0, 1] idx2_val = [0, 1]
rval = f(idx1_val, idx2_val) rval = f(idx1_val, idx2_val)
rep = xval[idx1_val, None, slice(0, 2, 1), idx2_val, None] rep = xval[idx1_val, None, slice(0, 2, 1), idx2_val, None]
assert numpy.allclose(rval, rep) assert np.allclose(rval, rep)
class test_gpudiagonal(unittest.TestCase): class test_gpudiagonal(unittest.TestCase):
def test_matrix(self): def test_matrix(self):
x = tensor.matrix() x = tensor.matrix()
np_x = numpy.arange(77).reshape(7, 11).astype(theano.config.floatX) np_x = np.arange(77).reshape(7, 11).astype(theano.config.floatX)
fn = theano.function([x], GpuDiagonal()(x), mode=mode_with_gpu) fn = theano.function([x], GpuDiagonal()(x), mode=mode_with_gpu)
assert numpy.allclose(fn(np_x), np_x.diagonal()) assert np.allclose(fn(np_x), np_x.diagonal())
fn = theano.function([x], GpuDiagonal(2)(x), mode=mode_with_gpu) fn = theano.function([x], GpuDiagonal(2)(x), mode=mode_with_gpu)
assert numpy.allclose(fn(np_x), np_x.diagonal(2)) assert np.allclose(fn(np_x), np_x.diagonal(2))
fn = theano.function([x], GpuDiagonal(-3)(x), mode=mode_with_gpu) fn = theano.function([x], GpuDiagonal(-3)(x), mode=mode_with_gpu)
assert numpy.allclose(fn(np_x), np_x.diagonal(-3)) assert np.allclose(fn(np_x), np_x.diagonal(-3))
def test_tensor(self): def test_tensor(self):
x = tensor.tensor4() x = tensor.tensor4()
np_x = numpy.arange(30107).reshape(7, 11, 17, 23).astype(theano.config.floatX) np_x = np.arange(30107).reshape(7, 11, 17, 23).astype(theano.config.floatX)
for offset, axis1, axis2 in [ for offset, axis1, axis2 in [
(1, 0, 1), (-1, 0, 1), (0, 1, 0), (-2, 1, 0), (1, 0, 1), (-1, 0, 1), (0, 1, 0), (-2, 1, 0),
(-3, 1, 0), (-2, 2, 0), (3, 3, 0), (-1, 3, 2), (-3, 1, 0), (-2, 2, 0), (3, 3, 0), (-1, 3, 2),
(2, 2, 3), (-1, 2, 1), (1, 3, 1), (-1, 1, 3)]: (2, 2, 3), (-1, 2, 1), (1, 3, 1), (-1, 1, 3)]:
assert numpy.allclose( assert np.allclose(
GpuDiagonal(offset, axis1, axis2)(x).eval({x: np_x}), GpuDiagonal(offset, axis1, axis2)(x).eval({x: np_x}),
np_x.diagonal(offset, axis1, axis2)) np_x.diagonal(offset, axis1, axis2))
theano/gpuarray/tests/test_type.py
浏览文件 @ 824cb369
from __future__ import absolute_import, print_function, division from __future__ import absolute_import, print_function, division
import os import os
import numpy import numpy as np
import theano import theano
from theano.compat import PY3 from theano.compat import PY3
...@@ -33,10 +33,10 @@ def test_values_eq_approx(): ...@@ -33,10 +33,10 @@ def test_values_eq_approx():
a = rand_gpuarray(20, dtype='float32') a = rand_gpuarray(20, dtype='float32')
assert GpuArrayType.values_eq_approx(a, a) assert GpuArrayType.values_eq_approx(a, a)
b = a.copy() b = a.copy()
b[0] = numpy.asarray(b[0]) + 1. b[0] = np.asarray(b[0]) + 1.
assert not GpuArrayType.values_eq_approx(a, b) assert not GpuArrayType.values_eq_approx(a, b)
b = a.copy() b = a.copy()
b[0] = -numpy.asarray(b[0]) b[0] = -np.asarray(b[0])
assert not GpuArrayType.values_eq_approx(a, b) assert not GpuArrayType.values_eq_approx(a, b)
...@@ -50,7 +50,7 @@ def test_specify_shape(): ...@@ -50,7 +50,7 @@ def test_specify_shape():
def test_filter_float(): def test_filter_float():
theano.compile.shared_constructor(gpuarray_shared_constructor) theano.compile.shared_constructor(gpuarray_shared_constructor)
try: try:
s = theano.shared(numpy.array(0.0, dtype='float32'), s = theano.shared(np.array(0.0, dtype='float32'),
target=test_ctx_name) target=test_ctx_name)
theano.function([], updates=[(s, 0.0)]) theano.function([], updates=[(s, 0.0)])
finally: finally:
...@@ -73,6 +73,6 @@ def test_unpickle_gpuarray_as_numpy_ndarray_flag0(): ...@@ -73,6 +73,6 @@ def test_unpickle_gpuarray_as_numpy_ndarray_flag0():
u = CompatUnpickler(fp) u = CompatUnpickler(fp)
mat = u.load() mat = u.load()
assert isinstance(mat, pygpu.gpuarray.GpuArray) assert isinstance(mat, pygpu.gpuarray.GpuArray)
assert numpy.asarray(mat)[0] == -42.0 assert np.asarray(mat)[0] == -42.0
finally: finally:
config.experimental.unpickle_gpu_on_cpu = oldflag config.experimental.unpickle_gpu_on_cpu = oldflag
theano/gpuarray/type.py
浏览文件 @ 824cb369
from __future__ import absolute_import, print_function, division from __future__ import absolute_import, print_function, division
import numpy import numpy as np
import six.moves.copyreg as copyreg import six.moves.copyreg as copyreg
from six import iteritems from six import iteritems
import warnings import warnings
...@@ -226,7 +226,7 @@ class GpuArrayType(Type): ...@@ -226,7 +226,7 @@ class GpuArrayType(Type):
converted_data = theano._asarray(data, self.dtype) converted_data = theano._asarray(data, self.dtype)
# We use the `values_eq` static function from TensorType # We use the `values_eq` static function from TensorType
# to handle NaN values. # to handle NaN values.
if TensorType.values_eq(numpy.asarray(data), if TensorType.values_eq(np.asarray(data),
converted_data, converted_data,
force_same_dtype=False): force_same_dtype=False):
data = converted_data data = converted_data
...@@ -293,18 +293,18 @@ class GpuArrayType(Type): ...@@ -293,18 +293,18 @@ class GpuArrayType(Type):
return False return False
if force_same_dtype and a.typecode != b.typecode: if force_same_dtype and a.typecode != b.typecode:
return False return False
a_eq_b = numpy.asarray(compare(a, '==', b)) a_eq_b = np.asarray(compare(a, '==', b))
if a_eq_b.all(): if a_eq_b.all():
return True return True
# maybe the trouble is that there are NaNs # maybe the trouble is that there are NaNs
a = numpy.asarray(a) a = np.asarray(a)
b = numpy.asarray(b) b = np.asarray(b)
a_missing = numpy.isnan(a) a_missing = np.isnan(a)
if a_missing.any(): if a_missing.any():
b_missing = numpy.isnan(b) b_missing = np.isnan(b)
return numpy.all(a_eq_b + (a_missing == b_missing)) return np.all(a_eq_b + (a_missing == b_missing))
else: else:
return False return False
...@@ -326,16 +326,16 @@ class GpuArrayType(Type): ...@@ -326,16 +326,16 @@ class GpuArrayType(Type):
rtol_ = rtol rtol_ = rtol
if atol is not None: if atol is not None:
atol_ = atol atol_ = atol
res = elemwise2(a, '', b, a, odtype=numpy.dtype('bool'), res = elemwise2(a, '', b, a, odtype=np.dtype('bool'),
op_tmpl="res = (fabs(a - b) <" op_tmpl="res = (fabs(a - b) <"
"(%(atol_)s + %(rtol_)s * fabs(b)))" % "(%(atol_)s + %(rtol_)s * fabs(b)))" %
locals()) locals())
ret = numpy.asarray(res).all() ret = np.asarray(res).all()
if ret: if ret:
return True return True
# maybe the trouble is that there are NaNs # maybe the trouble is that there are NaNs
an = numpy.asarray(a) an = np.asarray(a)
bn = numpy.asarray(b) bn = np.asarray(b)
return tensor.TensorType.values_eq_approx( return tensor.TensorType.values_eq_approx(
an, bn, allow_remove_inf=allow_remove_inf, an, bn, allow_remove_inf=allow_remove_inf,
allow_remove_nan=allow_remove_nan, rtol=rtol, atol=atol) allow_remove_nan=allow_remove_nan, rtol=rtol, atol=atol)
...@@ -408,9 +408,9 @@ class GpuArrayType(Type): ...@@ -408,9 +408,9 @@ class GpuArrayType(Type):
def get_size(self, shape_info): def get_size(self, shape_info):
if shape_info: if shape_info:
return numpy.prod(shape_info) * numpy.dtype(self.dtype).itemsize return np.prod(shape_info) * np.dtype(self.dtype).itemsize
else: else:
return numpy.dtype(self.dtype).itemsize return np.dtype(self.dtype).itemsize
def c_declare(self, name, sub, check_input=True): def c_declare(self, name, sub, check_input=True):
return """ return """
...@@ -470,7 +470,7 @@ class GpuArrayType(Type): ...@@ -470,7 +470,7 @@ class GpuArrayType(Type):
'<gpuarray_api.h>'] '<gpuarray_api.h>']
def c_header_dirs(self): def c_header_dirs(self):
return [pygpu.get_include(), numpy.get_include()] return [pygpu.get_include(), np.get_include()]
def c_libraries(self): def c_libraries(self):
return ['gpuarray'] return ['gpuarray']
...@@ -509,7 +509,7 @@ class GpuArrayVariable(_operators, Variable): ...@@ -509,7 +509,7 @@ class GpuArrayVariable(_operators, Variable):
# override the default # override the default
def __repr_test_value__(self): def __repr_test_value__(self):
return repr(numpy.array(theano.gof.op.get_test_value(self))) return repr(np.array(theano.gof.op.get_test_value(self)))
GpuArrayType.Variable = GpuArrayVariable GpuArrayType.Variable = GpuArrayVariable
...@@ -534,13 +534,13 @@ class GpuArrayConstant(_operators, Constant): ...@@ -534,13 +534,13 @@ class GpuArrayConstant(_operators, Constant):
""" """
def signature(self): def signature(self):
return GpuArraySignature((self.type, numpy.asarray(self.data))) return GpuArraySignature((self.type, np.asarray(self.data)))
def __str__(self): def __str__(self):
if self.name is not None: if self.name is not None:
return self.name return self.name
try: try:
np_data = numpy.asarray(self.data) np_data = np.asarray(self.data)
except gpuarray.GpuArrayException: except gpuarray.GpuArrayException:
np_data = self.data np_data = self.data
return "GpuArrayConstant{%s}" % np_data return "GpuArrayConstant{%s}" % np_data
...@@ -568,7 +568,7 @@ class GpuArraySharedVariable(_operators, SharedVariable): ...@@ -568,7 +568,7 @@ class GpuArraySharedVariable(_operators, SharedVariable):
else: else:
return self.container.value.copy() return self.container.value.copy()
else: else:
return numpy.asarray(self.container.value) return np.asarray(self.container.value)
def set_value(self, value, borrow=False): def set_value(self, value, borrow=False):
if isinstance(value, pygpu.gpuarray.GpuArray): if isinstance(value, pygpu.gpuarray.GpuArray):
...@@ -601,7 +601,7 @@ def gpuarray_shared_constructor(value, name=None, strict=False, ...@@ -601,7 +601,7 @@ def gpuarray_shared_constructor(value, name=None, strict=False,
if target == 'gpu' or target == 'cpu': if target == 'gpu' or target == 'cpu':
raise TypeError('not for me') raise TypeError('not for me')
if not isinstance(value, (numpy.ndarray, pygpu.gpuarray.GpuArray)): if not isinstance(value, (np.ndarray, pygpu.gpuarray.GpuArray)):
raise TypeError('ndarray or GpuArray required') raise TypeError('ndarray or GpuArray required')
if target is notset: if target is notset:
...@@ -809,7 +809,7 @@ copyreg.constructor(GpuArray_unpickler) ...@@ -809,7 +809,7 @@ copyreg.constructor(GpuArray_unpickler)
def GpuArray_pickler(cnda): def GpuArray_pickler(cnda):
ctx_name = _name_for_ctx(cnda.context) ctx_name = _name_for_ctx(cnda.context)
return (GpuArray_unpickler, (numpy.asarray(cnda), ctx_name)) return (GpuArray_unpickler, (np.asarray(cnda), ctx_name))
# In case pygpu is not imported. # In case pygpu is not imported.
if pygpu is not None: if pygpu is not None:
......
Markdown 格式
0% 或
您添加了 0 人 到此讨论。请谨慎行事。
请先完成此评论的编辑！
请 注册 或者 后发表评论