Skip to content

P
pytensor
提交 c89e1bc2 authored 作者: Frédéric Bastien's avatar Frédéric Bastien
浏览文件
操作

Merge pull request #3107 from harlouci/flake8_v3

Flake8 tensor
上级 de826376 cceee86e
隐藏空白字符变更
内嵌 并排
正在显示 包含 224 行增加 和 248 行删除
theano/sandbox/linalg/tests/test_linalg.py
浏览文件 @ c89e1bc2
......@@ -174,3 +174,13 @@ def test_tag_solve_triangular():
for node in f.maker.fgraph.toposort():
if isinstance(node.op, Solve):
assert node.op.A_structure == 'upper_triangular'
def test_matrix_inverse_solve():
if not imported_scipy:
raise SkipTest("Scipy needed for the Solve op.")
A = theano.tensor.dmatrix('A')
b = theano.tensor.dmatrix('b')
node = matrix_inverse(A).dot(b).owner
[out] = inv_as_solve.transform(node)
assert isinstance(out.owner.op, Solve)
theano/tensor/blas.py
浏览文件 @ c89e1bc2
......@@ -164,7 +164,7 @@ def default_blas_ldflags():
global numpy
try:
if (hasattr(numpy.distutils, '__config__') and
numpy.distutils.__config__):
numpy.distutils.__config__):
# If the old private interface is available use it as it
# don't print information to the user.
blas_info = numpy.distutils.__config__.blas_opt_info
......@@ -177,7 +177,7 @@ def default_blas_ldflags():
# ignored"
# This happen with Python 2.7.3 |EPD 7.3-1 and numpy 1.8.1
import numpy.distutils.system_info # noqa
import numpy.distutils.system_info # noqa
# We need to catch warnings as in some cases NumPy print
# stuff that we don't want the user to see like this:
......@@ -276,7 +276,7 @@ SOMEPATH/Canopy_64bit/User/lib/python2.7/site-packages/numpy/distutils/system_in
# Using "conda install mkl" will install both, as well as
# optimized versions of numpy and scipy.
try:
import mkl #noqa
import mkl # noqa
except ImportError as e:
_logger.info('Conda mkl is not available: %s', e)
else:
......@@ -319,8 +319,8 @@ SOMEPATH/Canopy_64bit/User/lib/python2.7/site-packages/numpy/distutils/system_in
AddConfigVar('blas.ldflags',
"lib[s] to include for [Fortran] level-3 blas implementation",
StrParam(default_blas_ldflags))
"lib[s] to include for [Fortran] level-3 blas implementation",
StrParam(default_blas_ldflags))
try:
......@@ -333,12 +333,10 @@ try:
# `scipy.linalg.blas.fblas` with `scipy.linalg.blas`.
# See http://github.com/scipy/scipy/pull/358
fblas = scipy.linalg.blas
_blas_gemv_fns = {
numpy.dtype('float32'): fblas.sgemv,
numpy.dtype('float64'): fblas.dgemv,
numpy.dtype('complex64'): fblas.cgemv,
numpy.dtype('complex128'): fblas.zgemv,
}
_blas_gemv_fns = {numpy.dtype('float32'): fblas.sgemv,
numpy.dtype('float64'): fblas.dgemv,
numpy.dtype('complex64'): fblas.cgemv,
numpy.dtype('complex128'): fblas.zgemv}
except ImportError as e:
have_fblas = False
# This is used in Gemv and ScipyGer. We use CGemv and CGer
......@@ -400,8 +398,8 @@ class Gemv(Op):
# The following is not grounds for error because as long as
# sizes are 1 at time of perform() there is no problem
# if x.broadcastable[0] != A.broadcastable[1]:
# raise TypeError('broadcastable mismatch between x and A',
#(x.type, A.type))
# raise TypeError('broadcastable mismatch between x and A',
# (x.type, A.type))
return Apply(self, [y, alpha, A, x, beta], [y.type()])
def perform(self, node, inputs, out_storage):
......@@ -411,9 +409,10 @@ class Gemv(Op):
gemv = _blas_gemv_fns[y.dtype]
if (A.shape[0] != y.shape[0] or A.shape[1] != x.shape[0]):
raise ValueError('Incompatible shapes for gemv '
'(beta * y + alpha * dot(A, x)). y: %s, A: %s, x: %s '
% (y.shape, A.shape, x.shape))
raise ValueError(
'Incompatible shapes for gemv '
'(beta * y + alpha * dot(A, x)). y: %s, A: %s, x: %s '
% (y.shape, A.shape, x.shape))
# Here I suppose that A is in c order. If we don't make it
# explicitly as fortran order, scipy 0.7.2 seam to create
......@@ -479,7 +478,7 @@ class Ger(Op):
alpha = T.as_tensor_variable(alpha)
if len(set([A.dtype, alpha.dtype, x.dtype, y.dtype])) != 1:
raise TypeError('ger requires matching dtypes',
(A.dtype, alpha.dtype, x.dtype, y.dtype))
(A.dtype, alpha.dtype, x.dtype, y.dtype))
if alpha.ndim != 0:
raise TypeError('ger requires scalar alpha', alpha.type)
if A.ndim != 2:
......@@ -567,13 +566,14 @@ def _ldflags(ldflags_str, libs, flags, libs_dir, include_dir):
for d in dirs:
for f in os.listdir(d):
if (f.endswith('.so') or f.endswith('.dylib') or
f.endswith('.dll')):
f.endswith('.dll')):
if any([f.find(ll) >= 0 for ll in l]):
found_dyn = True
if not found_dyn and dirs:
_logger.warning("We did not found a dynamic library into the "
"library_dir of the library we use for blas. If you use "
"ATLAS, make sure to compile it with dynamics library.")
_logger.warning(
"We did not found a dynamic library into the "
"library_dir of the library we use for blas. If you use "
"ATLAS, make sure to compile it with dynamics library.")
for t in ldflags_str.split():
# Remove extra quote.
......@@ -644,7 +644,7 @@ class GemmRelated(Op):
return ldflags()
# code_cache_version is built by subclasses from
# build_gemm_version
# build_gemm_version
def c_compile_args(self):
return ldflags(libs=False, flags=True)
......@@ -673,7 +673,7 @@ class GemmRelated(Op):
int sx_0, sx_1, sy_0, sy_1, sz_0, sz_1;
"""
#setup_z_Nz_Sz = None
# setup_z_Nz_Sz = None
check_xyz_rank2 = """
if (PyArray_NDIM(%(_x)s) != 2) {
......@@ -823,7 +823,7 @@ class GemmRelated(Op):
{
"""
#case_float_ab_constants = None
# case_float_ab_constants = None
case_float_gemm = """
float* x = (float*)PyArray_DATA(%(_x)s);
......@@ -856,7 +856,7 @@ class GemmRelated(Op):
{
"""
#case_double_ab_constants = None
# case_double_ab_constants = None
case_double_gemm = """
double* x = (double*)PyArray_DATA(%(_x)s);
......@@ -1028,10 +1028,10 @@ class Gemm(GemmRelated):
if not (z.dtype == a.dtype == x.dtype == y.dtype == b.dtype):
raise TypeError(Gemm.E_mixed,
(z.dtype, a.dtype, x.dtype, y.dtype, b.dtype))
(z.dtype, a.dtype, x.dtype, y.dtype, b.dtype))
if (not z.dtype.startswith('float')
and not z.dtype.startswith('complex')):
if (not z.dtype.startswith('float') and
not z.dtype.startswith('complex')):
raise TypeError(Gemm.E_float, (z.dtype))
output = z.type()
......@@ -1173,8 +1173,8 @@ class Gemm(GemmRelated):
_z, _a, _x, _y, _b = inp
_zout, = out
if node.inputs[0].type.dtype.startswith('complex'):
raise utils.MethodNotDefined('%s.c_code' \
% self.__class__.__name__)
raise utils.MethodNotDefined('%s.c_code'
% self.__class__.__name__)
if not config.blas.ldflags:
return super(Gemm, self).c_code(node, name,
(_z, _a, _x, _y, _b), (_zout, ),
......@@ -1203,9 +1203,9 @@ def res_is_a(node, op, maxclients=None):
else:
retval = True
return node.owner \
and node.owner.op == op \
and retval
return (node.owner and
node.owner.op == op and
retval)
def _as_scalar(res, dtype=None):
......@@ -1235,16 +1235,16 @@ def _as_scalar(res, dtype=None):
def _is_real_matrix(res):
return res.type.dtype in ('float32', 'float64') \
and res.type.ndim == 2 \
and res.type.broadcastable[0] == False \
and res.type.broadcastable[1] == False # cope with tuple vs. list
return (res.type.dtype in ('float32', 'float64') and
res.type.ndim == 2 and
res.type.broadcastable[0] == False and
res.type.broadcastable[1] == False) # cope with tuple vs. list
def _is_real_vector(res):
return res.type.dtype in ('float32', 'float64') \
and res.type.ndim == 1 \
and res.type.broadcastable[0] == False
return (res.type.dtype in ('float32', 'float64') and
res.type.ndim == 1 and
res.type.broadcastable[0] == False)
def _beta_L_plus_alpha_M(beta, L, alpha, M, recurse_flip=True):
......@@ -1262,8 +1262,8 @@ def _beta_L_plus_alpha_M(beta, L, alpha, M, recurse_flip=True):
# it also might be the case that there is a dimshuffle between the +
# and the dot22. local_dot_to_dot22 in particular will put in such things.
if (M.owner and isinstance(M.owner.op, T.DimShuffle) and
M.owner.inputs[0].owner and
isinstance(M.owner.inputs[0].owner.op, Dot22)):
M.owner.inputs[0].owner and
isinstance(M.owner.inputs[0].owner.op, Dot22)):
MM = M.owner.inputs[0]
if M.owner.op.new_order == (0,):
# it is making a column MM into a vector
......@@ -1493,7 +1493,7 @@ def _gemm_from_factored_list(lst):
assert len(gemm_of_sM_list) == 1
add_inputs = [item_to_var(input)
for k, input in enumerate(lst) if k not in (i, j)]
for k, input in enumerate(lst) if k not in (i, j)]
add_inputs.extend(gemm_of_sM_list)
if len(add_inputs) > 1:
rval = [T.add(*add_inputs)]
......@@ -1583,7 +1583,7 @@ class GemmOptimizer(Optimizer):
(theano.scalar.Add, theano.scalar.Sub,
theano.scalar.Neg, theano.scalar.Mul))):
continue
if not node in fgraph.apply_nodes:
if node not in fgraph.apply_nodes:
# This mean that we already removed this node from
# the graph
continue
......@@ -1592,7 +1592,7 @@ class GemmOptimizer(Optimizer):
time_canonicalize += time1
time_factor_can += time2
time_factor_list += time3
except InconsistencyError as e:
except InconsistencyError:
nb_inconsistency_make += 1
continue
if new_outputs:
......@@ -1725,8 +1725,8 @@ class Dot22(GemmRelated):
_x, _y = inp
_zout, = out
if node.inputs[0].type.dtype.startswith('complex'):
raise utils.MethodNotDefined('%s.c_code' \
% self.__class__.__name__)
raise utils.MethodNotDefined('%s.c_code'
% self.__class__.__name__)
if len(self.c_libraries()) <= 0:
return super(Dot22, self).c_code(node, name, (_x, _y),
(_zout, ), sub)
......@@ -1895,17 +1895,16 @@ blas_optdb.register('local_dot_to_dot22',
in2out(local_dot_to_dot22),
0, 'fast_run', 'fast_compile')
blas_optdb.register('gemm_optimizer',
GemmOptimizer(),
10, 'fast_run')
GemmOptimizer(),
10, 'fast_run')
blas_optdb.register('local_gemm_to_gemv',
EquilibriumOptimizer([
local_gemm_to_gemv,
local_gemm_to_ger,
local_dot22_to_ger_or_gemv,
local_dimshuffle_lift],
max_use_ratio=5,
ignore_newtrees=False),
15, 'fast_run')
EquilibriumOptimizer([local_gemm_to_gemv,
local_gemm_to_ger,
local_dot22_to_ger_or_gemv,
local_dimshuffle_lift],
max_use_ratio=5,
ignore_newtrees=False),
15, 'fast_run')
# After destroyhandler(49.5) but before we try to make elemwise things
......@@ -1936,12 +1935,12 @@ class Dot22Scalar(GemmRelated):
if not (a.dtype == x.dtype == y.dtype):
raise TypeError('Dot22Scalar requires matching dtypes',
(a.dtype, x.dtype, y.dtype))
(a.dtype, x.dtype, y.dtype))
if (not a.dtype.startswith('float')
and not a.dtype.startswith('complex')):
if (not a.dtype.startswith('float') and
not a.dtype.startswith('complex')):
raise TypeError('Dot22Scalar requires float or complex args',
a.dtype)
a.dtype)
bz = [x.type.broadcastable[0], y.type.broadcastable[1]]
outputs = [T.tensor(x.type.dtype, bz)]
......@@ -1992,8 +1991,8 @@ class Dot22Scalar(GemmRelated):
_x, _y, _a = inp
_zout, = out
if node.inputs[0].type.dtype.startswith('complex'):
raise utils.MethodNotDefined('%s.c_code' \
% self.__class__.__name__)
raise utils.MethodNotDefined('%s.c_code'
% self.__class__.__name__)
if len(self.c_libraries()) <= 0:
return super(Dot22Scalar, self).c_code(node, name, (_x, _y),
(_zout, ), sub)
......@@ -2051,7 +2050,7 @@ def local_dot22_to_dot22scalar(node):
# The canonizer should have merged those mul together.
i_mul = [x.owner and x.owner.op == T.mul and
any([_as_scalar(x_i, dtype=d.dtype)
for x_i in x.owner.inputs])
for x_i in x.owner.inputs])
for x in node.inputs]
if not any(i_mul):
# no scalar in input and no multiplication
......@@ -2065,8 +2064,7 @@ def local_dot22_to_dot22scalar(node):
scalar_idx = -1
for i, x in enumerate(m.owner.inputs):
if _as_scalar(x, dtype=d.dtype) and (theano.scalar.upcast(
x.type.dtype, d.type.dtype)
== d.type.dtype):
x.type.dtype, d.type.dtype) == d.type.dtype):
scalar_idx = i
break
......@@ -2103,8 +2101,8 @@ def local_dot22_to_dot22scalar(node):
break
if scalar_idx < 0:
_logger.info('Not optimizing dot22 with inputs %s %s, as the type '
'of the scalar cannot be upcasted to the matrix type',
node.inputs, [x.type for x in node.inputs])
'of the scalar cannot be upcasted to the matrix type',
node.inputs, [x.type for x in node.inputs])
return False
assert scalar_idx < len(node.inputs)
s = node.inputs[scalar_idx]
......@@ -2128,8 +2126,8 @@ blas_optdb.register('local_dot22_to_dot22scalar',
11, 'fast_run')
#from opt import register_specialize, register_canonicalize
#@register_specialize
# from opt import register_specialize, register_canonicalize
# @register_specialize
@local_optimizer([T.sub, T.add])
def local_print_as_we_go_along(node):
if node.op in (T.sub, T.add):
......
theano/tensor/blas_c.py
浏览文件 @ c89e1bc2
......@@ -4,8 +4,7 @@ from theano import config
from theano.tensor.opt import in2out
from theano.tensor.blas import ldflags, blas_header_text, blas_header_version
from theano.tensor.blas import (
blas_optdb, optdb, local_optimizer, EquilibriumOptimizer)
from theano.tensor.blas import blas_optdb, optdb, local_optimizer
from theano.tensor.blas import Ger, ger, ger_destructive
from theano.tensor.blas import Gemv, gemv_inplace, gemv_no_inplace
from theano.tensor import basic as T
......@@ -268,7 +267,7 @@ def ger_c_code(A, a, x, y, Z, destructive, fail):
(double*)x_data, &Sx,
(double*)y_data, &Sy,
(double*)(PyArray_DATA(%(Z)s)), &Sz1);
}
else {
......@@ -610,7 +609,7 @@ def gemv_c_code(aa, xx, yy, zz, alpha, beta, destructive, fail,
// so Sx1 == 1 is required for safety.
if (Nx0 == 1 && Sx1 == 1)
{
zz_data[0] = fbeta*zz_data[0] + alpha*sdot_(&Nx1,
zz_data[0] = fbeta*zz_data[0] + alpha*sdot_(&Nx1,
(float*)(PyArray_DATA(%(xx)s)), &Sx1,
(float*)yy_data, &Sy);
}
......@@ -633,7 +632,7 @@ def gemv_c_code(aa, xx, yy, zz, alpha, beta, destructive, fail,
// so Sx1 == 1 is required for safety.
if (Nx0 == 1 && Sx1 == 1)
{
zz_data[0] = dbeta*zz_data[0] + alpha*ddot_(&Nx1,
zz_data[0] = dbeta*zz_data[0] + alpha*ddot_(&Nx1,
(double*)(PyArray_DATA(%(xx)s)), &Sx1,
(double*)yy_data, &Sy);
}
......@@ -732,8 +731,7 @@ def check_force_gemv_init():
gemv_no_inplace(aa, 1., xx, yy, 0.),
theano.compile.Mode(optimizer='fast_compile').excluding('gpu',
'gpuarray'),
profile=False
)
profile=False)
finally:
theano.config.compute_test_value = tv
theano.config.compute_test_value_opt = tvo
......@@ -742,11 +740,11 @@ def check_force_gemv_init():
# then we want gemv_c_code to initiliaze the memory to 0 so that we
# don't inadvertantly introduce NaNs to the users data.
aa_data = numpy.array(
float('NaN')*numpy.ones((2,)),
float('NaN') * numpy.ones((2,)),
dtype=theano.config.floatX
)
yy_data = numpy.array(
numpy.ones((2,))*2,
numpy.ones((2,)) * 2,
dtype=theano.config.floatX
)
xx_data = numpy.array(
......
theano/tensor/blas_scipy.py
浏览文件 @ c89e1bc2
......@@ -12,11 +12,11 @@ from theano.tensor.opt import in2out
if have_fblas:
from theano.tensor.blas import fblas
_blas_ger_fns = {
numpy.dtype('float32'): fblas.sger,
numpy.dtype('float64'): fblas.dger,
numpy.dtype('complex64'): fblas.cgeru,
numpy.dtype('complex128'): fblas.zgeru,
}
numpy.dtype('float32'): fblas.sger,
numpy.dtype('float64'): fblas.dger,
numpy.dtype('complex64'): fblas.cgeru,
numpy.dtype('complex128'): fblas.zgeru,
}
class ScipyGer(Ger):
......@@ -47,10 +47,10 @@ class ScipyGer(Ger):
A = A.copy()
elif A.flags['C_CONTIGUOUS']:
A = local_ger(calpha[0], cy[0], cx[0], a=A.T,
overwrite_a=int(self.destructive)).T
overwrite_a=int(self.destructive)).T
else:
A = local_ger(calpha[0], cx[0], cy[0], a=A,
overwrite_a=int(self.destructive))
overwrite_a=int(self.destructive))
cZ[0] = A
for o in node_output_compute:
o[0] = True
......@@ -87,10 +87,10 @@ if have_fblas:
# precedence. Once the original Ger is replaced, then these optimizations
# have no effect.
blas_optdb.register('scipy_blas',
use_scipy_blas,
100, 'fast_run')
use_scipy_blas,
100, 'fast_run')
# this matches the InplaceBlasOpt defined in blas.py
optdb.register('make_scipy_blas_destructive',
make_scipy_blas_destructive,
70.0, 'fast_run', 'inplace')
make_scipy_blas_destructive,
70.0, 'fast_run', 'inplace')
theano/tensor/elemwise_cgen.py
浏览文件 @ c89e1bc2
......@@ -276,7 +276,7 @@ def make_reordered_loop(init_loop_orders, olv_index, dtypes, inner_task, sub, op
if index != 'x':
order_loops += """
%(ovar)s_loops_it->first = abs(PyArray_STRIDES(%(ovar)s)[%(index)i]);
""" % locals()
""" % locals()
else:
# Stride is 0 when dimension is broadcastable
order_loops += """
......@@ -311,15 +311,13 @@ def make_reordered_loop(init_loop_orders, olv_index, dtypes, inner_task, sub, op
total = "%(var)s_n%(candidate)s" % locals()
break
else:
total = '1';
total = '1'
totals.append(total)
declare_totals = """
int init_totals[%(nnested)s] = {%(totals)s};
""" % dict(
nnested=nnested,
totals=', '.join(totals)
)
""" % dict(nnested=nnested,
totals=', '.join(totals))
# Sort totals to match the new order that was computed by sorting
# the loop vector. One integer variable per loop is declared.
......@@ -355,13 +353,11 @@ def make_reordered_loop(init_loop_orders, olv_index, dtypes, inner_task, sub, op
declare_strides = """
int init_strides[%(nvars)i][%(nnested)i] = {
%(strides)s
};""" % dict(
nvars=nvars,
nnested=nnested,
strides=', \n'.join(
', '.join(get_loop_strides(lo, i))
for i, lo in enumerate(init_loop_orders)
if len(lo) > 0))
};""" % dict(nvars=nvars,
nnested=nnested,
strides=', \n'.join(', '.join(get_loop_strides(lo, i))
for i, lo in enumerate(init_loop_orders)
if len(lo) > 0))
# Declare (sorted) stride and for each variable
# we iterate from innermost loop to outermost loop
......@@ -385,9 +381,9 @@ def make_reordered_loop(init_loop_orders, olv_index, dtypes, inner_task, sub, op
declare_iter += "%(var)s_iter = (%(dtype)s*)(PyArray_DATA(%(var)s));\n" % locals()
pointer_update = ''
for j , dtype in enumerate(dtypes):
for j, dtype in enumerate(dtypes):
var = sub["lv%i" % j]
pointer_update += "%(dtype)s &%(var)s_i = * ( %(var)s_iter"%locals()
pointer_update += "%(dtype)s &%(var)s_i = * ( %(var)s_iter" % locals()
tot_jump = ''
for i in reversed(range(nnested)):
iterv = 'ITER_%i' % i
......@@ -401,7 +397,7 @@ def make_reordered_loop(init_loop_orders, olv_index, dtypes, inner_task, sub, op
update = ''
forloop = ''
# The pointers are defined only in the most inner loop
if i == nnested-1:
if i == nnested - 1:
update = pointer_update
if i == 0:
if openmp:
......@@ -413,19 +409,17 @@ def make_reordered_loop(init_loop_orders, olv_index, dtypes, inner_task, sub, op
%(forloop)s
{ // begin loop %(i)i
%(update)s
%(loop)s
%(loop)s
} // end loop %(i)i
""" % locals()
return '\n'.join([
'{',
order_loops,
declare_totals,
declare_strides,
declare_iter,
loop,
'}\n',
])
return '\n'.join(['{',
order_loops,
declare_totals,
declare_strides,
declare_iter,
loop,
'}\n'])
# print make_declare(((0, 1, 2, 3), ('x', 1, 0, 3), ('x', 'x', 'x', 0)),
# ('double', 'int', 'float'),
......@@ -451,16 +445,16 @@ def make_reordered_loop(init_loop_orders, olv_index, dtypes, inner_task, sub, op
##################
### DimShuffle ###
# DimShuffle #
##################
#################
### Broadcast ###
# Broadcast #
#################
################
### CAReduce ###
# CAReduce #
################
......@@ -527,4 +521,3 @@ def make_loop_careduce(loop_orders, dtypes, loop_tasks, sub):
s += loop_tasks[-1]
return "{%s}" % s
theano/tensor/extra_ops.py
浏览文件 @ c89e1bc2
......@@ -5,6 +5,7 @@ from six.moves import xrange
import theano
from theano.tensor import basic
from theano.tensor import nlinalg # noqa
from theano import gof, scalar
from theano.gradient import DisconnectedType
tensor = basic
......
theano/tensor/fourier.py
浏览文件 @ c89e1bc2
......@@ -62,7 +62,7 @@ class Fourier(gof.Op):
(axis.data < 0 or axis.data > a.ndim - 1)):
raise TypeError('%s: index of the transformed axis must be'
' a scalar not smaller than 0 and smaller than'
' dimension of array' % self.__class__.__name__)
' dimension of array' % self.__class__.__name__)
if n is None:
n = a.shape[axis]
n = tensor.as_tensor_variable(n)
......@@ -78,7 +78,7 @@ class Fourier(gof.Op):
' strictly positive scalar'
% self.__class__.__name__)
return gof.Apply(self, [a, n, axis], [tensor.TensorType('complex128',
a.type.broadcastable)()])
a.type.broadcastable)()])
def infer_shape(self, node, in_shapes):
shape_a = in_shapes[0]
......@@ -87,8 +87,8 @@ class Fourier(gof.Op):
if len(shape_a) == 1:
return [(n,)]
elif isinstance(axis, tensor.TensorConstant):
out_shape = list(shape_a[0: axis.data.item()]) + [n] +\
list(shape_a[axis.data + 1:])
out_shape = (list(shape_a[0: axis.data.item()]) + [n] +
list(shape_a[axis.data + 1:]))
else:
l = len(shape_a)
shape_a = tensor.stack(*shape_a)
......@@ -136,7 +136,8 @@ class Fourier(gof.Op):
flip_shape = list(numpy.arange(0, a.ndim)[::-1])
res = res.dimshuffle(flip_shape)
res = tensor.switch(tensor.lt(n, tensor.shape(a)[axis]),
tensor.set_subtensor(res[n::, ], 0, False, False), res)
tensor.set_subtensor(res[n::, ], 0, False, False),
res)
res = res.dimshuffle(flip_shape)
# insures that gradient shape conforms to input shape:
......
theano/tensor/nlinalg.py
浏览文件 @ c89e1bc2
from __future__ import print_function
import logging
import theano
logger = logging.getLogger(__name__)
import numpy
from six.moves import xrange
import theano
from theano.tensor import as_tensor_variable
from theano.gof import Op, Apply
from theano.tensor import as_tensor_variable, dot, DimShuffle, Dot
from theano.tensor.blas import Dot22
from theano.tensor.opt import (register_stabilize,
register_specialize, register_canonicalize)
from theano.gof import local_optimizer
from theano.gof.opt import Optimizer
from theano.gradient import DisconnectedType
from theano.tensor import basic as tensor
logger = logging.getLogger(__name__)
class MatrixPinv(Op):
"""Computes the pseudo-inverse of a matrix :math:`A`.
......@@ -427,8 +423,10 @@ class EighGrad(Op):
N = x.shape[0]
outer = numpy.outer
G = lambda n: sum(v[:, m] * V.T[n].dot(v[:, m]) / (w[n] - w[m])
for m in xrange(N) if m != n)
def G(n):
return sum(v[:, m] * V.T[n].dot(v[:, m]) / (w[n] - w[m])
for m in xrange(N) if m != n)
g = sum(outer(v[:, n], v[:, n] * W[n] + G(n))
for n in xrange(N))
......@@ -641,16 +639,6 @@ def svd(a, full_matrices=1, compute_uv=1):
return SVD(full_matrices, compute_uv)(a)
def test_matrix_inverse_solve():
if not imported_scipy:
raise SkipTest("Scipy needed for the Solve op.")
A = theano.tensor.dmatrix('A')
b = theano.tensor.dmatrix('b')
node = matrix_inverse(A).dot(b).owner
[out] = inv_as_solve.transform(node)
assert isinstance(out.owner.op, Solve)
class lstsq(Op):
def __eq__(self, other):
return type(self) == type(other)
......@@ -670,9 +658,6 @@ class lstsq(Op):
theano.tensor.lscalar(), theano.tensor.dvector()])
def perform(self, node, inputs, outputs):
x = inputs[0]
y = inputs[1]
rcond = inputs[2]
zz = numpy.linalg.lstsq(inputs[0], inputs[1], inputs[2])
outputs[0][0] = zz[0]
outputs[1][0] = zz[1]
......@@ -703,7 +688,7 @@ def norm(x, ord):
return x[x.nonzero()].shape[0]
else:
try:
z = tensor.sum(abs(x**ord))**(1./ord)
z = tensor.sum(abs(x**ord))**(1. / ord)
except TypeError:
raise ValueError("Invalid norm order for vectors.")
return z
......
theano/tensor/opt_uncanonicalize.py
浏览文件 @ c89e1bc2
......@@ -33,7 +33,6 @@ supposed to be canonical.
# TODO: intelligent merge for mul/add
# TODO: 0*x -> 0
import logging
_logger = logging.getLogger('theano.tensor.opt')
from theano import gof
from theano.tensor.elemwise import CAReduce
......@@ -44,6 +43,8 @@ from theano.tensor.basic import (get_scalar_constant_value,
from theano.tensor.opt import register_uncanonicalize
from theano import scalar as scal
_logger = logging.getLogger('theano.tensor.opt')
@register_uncanonicalize
@gof.local_optimizer([T._max_and_argmax])
......@@ -81,8 +82,8 @@ def local_max_to_min(node):
if node.op == T.neg and node.inputs[0].owner:
max = node.inputs[0]
if (max.owner and
isinstance(max.owner.op, CAReduce)
and max.owner.op.scalar_op == scal.maximum):
isinstance(max.owner.op, CAReduce) and
max.owner.op.scalar_op == scal.maximum):
neg = max.owner.inputs[0]
if neg.owner and neg.owner.op == T.neg:
return [CAReduce(scal.minimum,
......
theano/tensor/type.py
浏览文件 @ c89e1bc2
import logging
_logger = logging.getLogger("theano.tensor.type")
import warnings
import numpy
import theano
from theano import config
from theano.gof import Constant, hashtype, Type, Variable
from theano.gof.utils import MethodNotDefined
from theano.gof import hashtype, Type, Variable
from theano import scalar as scal
_logger = logging.getLogger("theano.tensor.type")
class TensorType(Type):
"""Symbolic `Type` representing a numpy.ndarray value."""
......@@ -39,7 +40,7 @@ class TensorType(Type):
if self.dtype == 'floatX':
self.dtype = config.floatX
# broadcastable is immutable, and all elements are either
### True or False
# True or False
self.broadcastable = tuple(bool(b) for b in broadcastable)
self.dtype_specs() # error checking is done there
self.name = name
......@@ -74,17 +75,17 @@ class TensorType(Type):
# input (typical mistake, especially with shared variables).
if isinstance(data, Variable):
raise TypeError(
'Expected an array-like object, but found a Variable: '
'maybe you are trying to call a function on a (possibly '
'shared) variable instead of a numeric array?')
'Expected an array-like object, but found a Variable: '
'maybe you are trying to call a function on a (possibly '
'shared) variable instead of a numeric array?')
if ((type(data) is numpy.ndarray)
and (data.dtype == self.numpy_dtype)):
if ((type(data) is numpy.ndarray) and
(data.dtype == self.numpy_dtype)):
if data.dtype.num != self.numpy_dtype.num:
data = theano._asarray(data, dtype=self.dtype)
# -- now fall through to ndim check
elif((type(data) is numpy.memmap)
and (data.dtype == self.numpy_dtype)):
elif ((type(data) is numpy.memmap) and
(data.dtype == self.numpy_dtype)):
# numpy.memmap is a "safe" subclass of ndarray,
# so we can use it whereever we expect a base ndarray.
# however, casting it would defeat the purpose of not
......@@ -95,11 +96,11 @@ class TensorType(Type):
# we raise a meaningful TypeError.
if not (type(data) is numpy.ndarray):
raise TypeError("%s expected a ndarray object." % self,
data, type(data))
data, type(data))
if data.dtype != self.numpy_dtype:
raise TypeError(("%s expected a ndarray object with "
"dtype = %s (got %s).") % (
self, self.numpy_dtype, data.dtype))
"dtype = %s (got %s).") %
(self, self.numpy_dtype, data.dtype))
assert False, "This point should never be reached."
else:
if allow_downcast:
......@@ -185,7 +186,7 @@ class TensorType(Type):
" dimension.", data.shape, self.broadcastable)
i += 1
if (self.filter_checks_isfinite and
not numpy.all(numpy.isfinite(data))):
not numpy.all(numpy.isfinite(data))):
raise ValueError("non-finite elements not allowed")
return data
......@@ -208,14 +209,12 @@ class TensorType(Type):
return other
raise TypeError(
'Cannot convert Type %(othertype)s '
'(of Variable %(other)s) into Type %(self)s. '
'You can try to manually convert %(other)s into a %(self)s.'
% dict(
othertype=other.type,
other=other,
self=self)
)
'Cannot convert Type %(othertype)s '
'(of Variable %(other)s) into Type %(self)s. '
'You can try to manually convert %(other)s into a %(self)s.' %
dict(othertype=other.type,
other=other,
self=self))
def value_validity_msg(self, a):
try:
......@@ -247,10 +246,10 @@ class TensorType(Type):
'int64': (int, 'npy_int64', 'NPY_INT64'),
'complex128': (complex, 'theano_complex128', 'NPY_COMPLEX128'),
'complex64': (complex, 'theano_complex64', 'NPY_COMPLEX64')
}[self.dtype]
}[self.dtype]
except KeyError:
raise TypeError("Unsupported dtype for %s: %s"
% (self.__class__.__name__, self.dtype))
% (self.__class__.__name__, self.dtype))
def to_scalar_type(self):
return scal.get_scalar_type(dtype=self.dtype)
......@@ -261,11 +260,11 @@ class TensorType(Type):
and other.broadcastable == self.broadcastable
def convert_variable(self, var):
if (type(self) == type(var.type) and
if (type(self) == type(var.type) and # noqa
self.dtype == var.type.dtype and
self.ndim == var.type.ndim and
all(sb == ob or ob for sb, ob in zip(self.broadcastable,
var.type.broadcastable))):
var.type.broadcastable))):
return theano.tensor.patternbroadcast(var, self.broadcastable)
@staticmethod
......@@ -351,7 +350,7 @@ class TensorType(Type):
rtol = 1.0000000000000001e-05
atol = 1e-8
cmp_elemwise = (numpy.absolute(a - b) <=
(atol + rtol * numpy.absolute(b)))
(atol + rtol * numpy.absolute(b)))
# Find places where both a and b have missing values.
both_missing = a_missing * numpy.isnan(b)
......@@ -361,9 +360,9 @@ class TensorType(Type):
# cmp_elemwise is weird when we have inf and -inf.
# set it to False
cmp_elemwise = numpy.where(
both_inf & cmp_elemwise,
a == b,
cmp_elemwise)
both_inf & cmp_elemwise,
a == b,
cmp_elemwise)
# check the sign of the inf
both_inf = numpy.where(both_inf, (a == b), both_inf)
......@@ -383,7 +382,7 @@ class TensorType(Type):
return hashtype(self) ^ hash(self.dtype) ^ hash(self.broadcastable)
ndim = property(lambda self: len(self.broadcastable),
doc="number of dimensions")
doc="number of dimensions")
"""Number of dimensions
This read-only property is the preferred way to get the number of
......@@ -407,10 +406,10 @@ class TensorType(Type):
else:
b = self.broadcastable
named_broadcastable = {(): 'scalar',
(False,): 'vector',
(False, True): 'col',
(True, False): 'row',
(False, False): 'matrix'}
(False,): 'vector',
(False, True): 'col',
(True, False): 'row',
(False, False): 'matrix'}
if b in named_broadcastable:
bcast = named_broadcastable[b]
else:
......@@ -422,7 +421,7 @@ class TensorType(Type):
def __repr__(self):
return str(self)
#"TensorType{%s, %s}" % (str(self.dtype), str(self.broadcastable))
# "TensorType{%s, %s}" % (str(self.dtype), str(self.broadcastable))
def c_declare(self, name, sub, check_input=True):
"""Override `CLinkerType.c_declare` """
......@@ -636,13 +635,13 @@ def values_eq_approx_always_true(a, b):
# Register TensorType C code for ViewOp.
theano.compile.register_view_op_c_code(
TensorType,
"""
Py_XDECREF(%(oname)s);
%(oname)s = %(iname)s;
Py_XINCREF(%(oname)s);
""",
version=1)
TensorType,
"""
Py_XDECREF(%(oname)s);
%(oname)s = %(iname)s;
Py_XINCREF(%(oname)s);
""",
version=1)
# Register TensorType C code for Shape Op.
......@@ -665,51 +664,51 @@ theano.compile.register_shape_c_code(
# Register TensorType C code for ViewOp.
theano.compile.register_shape_i_c_code(
TensorType,
"""
if(!%(oname)s)
%(oname)s=(PyArrayObject*)PyArray_EMPTY(0, NULL, NPY_INT64, 0);
((npy_int64*)PyArray_DATA(%(oname)s))[0]=PyArray_DIMS(%(iname)s)[%(i)s];
""",
"""
if (%(i)s>=PyArray_NDIM(%(iname)s)){
PyErr_SetString(PyExc_TypeError,
"Number of dimensions lower than expected");
%(fail)s
}
""",
version=3)
TensorType,
"""
if(!%(oname)s)
%(oname)s=(PyArrayObject*)PyArray_EMPTY(0, NULL, NPY_INT64, 0);
((npy_int64*)PyArray_DATA(%(oname)s))[0]=PyArray_DIMS(%(iname)s)[%(i)s];
""",
"""
if (%(i)s>=PyArray_NDIM(%(iname)s)){
PyErr_SetString(PyExc_TypeError,
"Number of dimensions lower than expected");
%(fail)s
}
""",
version=3)
# Register TensorType C code for DeepCopyOp
theano.compile.register_deep_copy_op_c_code(
TensorType,
"""
int alloc = %(oname)s == NULL;
for(int i=0; !alloc && i<PyArray_NDIM(%(oname)s); i++) {
if(PyArray_DIMS(%(iname)s)[i] != PyArray_DIMS(%(oname)s)[i]) {
alloc = true;
break;
}
TensorType,
"""
int alloc = %(oname)s == NULL;
for(int i=0; !alloc && i<PyArray_NDIM(%(oname)s); i++) {
if(PyArray_DIMS(%(iname)s)[i] != PyArray_DIMS(%(oname)s)[i]) {
alloc = true;
break;
}
}
if(alloc) {
Py_XDECREF(%(oname)s);
%(oname)s = (PyArrayObject*)PyArray_NewCopy(%(iname)s,
NPY_ANYORDER);
if (!%(oname)s)
{
PyErr_SetString(PyExc_ValueError,
"DeepCopyOp: the copy failed!");
%(fail)s;
}
if(alloc) {
Py_XDECREF(%(oname)s);
%(oname)s = (PyArrayObject*)PyArray_NewCopy(%(iname)s,
NPY_ANYORDER);
if (!%(oname)s)
{
PyErr_SetString(PyExc_ValueError,
"DeepCopyOp: the copy failed!");
%(fail)s;
}
} else {
if(PyArray_CopyInto(%(oname)s, %(iname)s)){
PyErr_SetString(PyExc_ValueError,
"DeepCopyOp: the copy failed into already allocated space!");
%(fail)s;
}
} else {
if(PyArray_CopyInto(%(oname)s, %(iname)s)){
PyErr_SetString(PyExc_ValueError,
"DeepCopyOp: the copy failed into already allocated space!");
%(fail)s;
}
""",
version=2)
}
""",
version=2)
theano.compile.register_rebroadcast_c_code(
......@@ -723,7 +722,7 @@ theano.compile.register_rebroadcast_c_code(
%(fail)s
}
""",
version=1)
version=1)
theano.compile.register_specify_shape_c_code(
......
theano/tests/test_flake8.py
浏览文件 @ c89e1bc2
......@@ -57,17 +57,7 @@ whitelist_flake8 = [
"typed_list/tests/test_type.py",
"typed_list/tests/test_opt.py",
"typed_list/tests/test_basic.py",
"tensor/blas_headers.py",
"tensor/type.py",
"tensor/fourier.py",
"tensor/__init__.py",
"tensor/opt_uncanonicalize.py",
"tensor/blas.py",
"tensor/extra_ops.py",
"tensor/nlinalg.py",
"tensor/blas_c.py",
"tensor/elemwise_cgen.py",
"tensor/blas_scipy.py",
"tensor/tests/test_subtensor.py",
"tensor/tests/test_utils.py",
"tensor/tests/test_nlinalg.py",
......
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