Skip to content

P
pytensor
提交 19f7768f authored 作者: David Warde-Farley's avatar David Warde-Farley
浏览文件
操作

in range() -> in xrange() where possible

上级 94349257
隐藏空白字符变更
内嵌 并排
正在显示 包含 91 行增加 和 91 行删除
theano/gof/vm.py
浏览文件 @ 19f7768f
......@@ -94,7 +94,7 @@ class VM(object):
profile.apply_cimpl[node] = hasattr(thunk,'cthunk')
# clear the timer info out of the buffers
for i in range(len(self.call_times)):
for i in xrange(len(self.call_times)):
self.call_times[i] = 0.0
self.call_counts[i] = 0
......@@ -459,9 +459,9 @@ class VM_Linker(link.LocalLinker):
# put storage_map and compute_map into a int-based scheme
n_applies = len(nodes)
storage_map_list = [storage_map[vars_idx_inv[i]]
for i in range(len(vars_idx_inv))]
for i in xrange(len(vars_idx_inv))]
compute_map_list = [compute_map[vars_idx_inv[i]]
for i in range(len(vars_idx_inv))]
for i in xrange(len(vars_idx_inv))]
if nodes:
assert type(storage_map_list[0]) is list
assert type(compute_map_list[0]) is list
......
theano/sandbox/cuda/basic_ops.py
浏览文件 @ 19f7768f
......@@ -803,8 +803,8 @@ class GpuSum(Op):
assert N in [1,2,3]
makecall = self._makecall(node, name, x, z, fail)
N_pattern = ''.join(['1']*N)
param_dim = ",".join(["CudaNdarray_HOST_DIMS(%(x)s)[%(i)s]"%locals() for i in range(N+1)])
strides_dim = ",".join(["CudaNdarray_HOST_STRIDES(%(x)s)[%(i)s]"%locals() for i in range(N+1)])
param_dim = ",".join(["CudaNdarray_HOST_DIMS(%(x)s)[%(i)s]"%locals() for i in xrange(N+1)])
strides_dim = ",".join(["CudaNdarray_HOST_STRIDES(%(x)s)[%(i)s]"%locals() for i in xrange(N+1)])
threads_y = """
//get as many y threads as we can fit
while (n_threads.x * (n_threads.y+1) <= NUM_VECTOR_OP_THREADS_PER_BLOCK)
......@@ -1326,8 +1326,8 @@ class GpuSum(Op):
for_i3 = "for (int i3 = threadIdx.x; i3 < d3; i3 += blockDim.x)"
reducebuf = self._k_reduce_buf('Z[i0 * sZ0]')
param_dim = ",".join(["const int d%(i)s"%locals() for i in range(nd_in)])
param_strides = ",".join(["const int sA%(i)s"%locals() for i in range(nd_in)])
param_dim = ",".join(["const int d%(i)s"%locals() for i in xrange(nd_in)])
param_strides = ",".join(["const int sA%(i)s"%locals() for i in xrange(nd_in)])
decl = self._k_decl(node,nodename)
init = self._k_init(node,nodename)
print >> sio, """
......@@ -1988,7 +1988,7 @@ class GpuContiguous(Op):
Py_INCREF(%(z)s);
} else if ((NULL == %(z)s)"""%locals()
for i in range(len(node.inputs[0].type.broadcastable)):
for i in xrange(len(node.inputs[0].type.broadcastable)):
str += "\n|| (CudaNdarray_HOST_DIMS(%(input)s)[%(i)s] != CudaNdarray_HOST_DIMS(%(z)s)[%(i)s])"%locals()
str += """)
{
......
theano/sandbox/cuda/elemwise.py
浏览文件 @ 19f7768f
......@@ -552,9 +552,9 @@ class NaiveAlgo(object):
print >> sio, 'std::cerr << "\\n";'
for ipos in xrange(len(node.inputs)):
print >> sio, 'std::cerr << " local_str inputs %(ipos)s: " <<'%locals()+' << " " << '.join(["local_str[%(ipos)s][%(x)s]"%locals() for x in range(nd)])+'<<"\\n";'
print >> sio, 'std::cerr << " local_str inputs %(ipos)s: " <<'%locals()+' << " " << '.join(["local_str[%(ipos)s][%(x)s]"%locals() for x in xrange(nd)])+'<<"\\n";'
for ipos in xrange(len(node.outputs)):
print >> sio, 'std::cerr << " local_ostr inputs %(ipos)s: " <<'%locals()+' << " " << '.join(["local_ostr[%(ipos)s][%(x)s]"%locals() for x in range(nd)])+'<<"\\n";'
print >> sio, 'std::cerr << " local_ostr inputs %(ipos)s: " <<'%locals()+' << " " << '.join(["local_ostr[%(ipos)s][%(x)s]"%locals() for x in xrange(nd)])+'<<"\\n";'
print >> sio, """
for(int id=0;id<nd_collapse;id++){
......@@ -593,17 +593,17 @@ class NaiveAlgo(object):
print >> sio, 'std::cerr << "\\n";'
for ipos in xrange(len(node.inputs)):
print >> sio, 'std::cerr << " local_str %(ipos)s: " <<'%locals()+' << " " << '.join(["local_str[%(ipos)s][%(x)s]"%locals() for x in range(nd)])+'<<"\\n";'
print >> sio, 'std::cerr << " local_str %(ipos)s: " <<'%locals()+' << " " << '.join(["local_str[%(ipos)s][%(x)s]"%locals() for x in xrange(nd)])+'<<"\\n";'
for ipos in xrange(len(node.outputs)):
print >> sio, 'std::cerr << " local_ostr %(ipos)s: " <<'%locals()+' << " " << '.join(["local_ostr[%(ipos)s][%(x)s]"%locals() for x in range(nd)])+'<<"\\n";'
print >> sio, 'std::cerr << " local_ostr %(ipos)s: " <<'%locals()+' << " " << '.join(["local_ostr[%(ipos)s][%(x)s]"%locals() for x in xrange(nd)])+'<<"\\n";'
# collapse contiguous dimensions (ignoring scalars, generic version(collapse any dimensions, right, left, middle))
# this is a good idea because we make less index calculation in the gpu.
print >> sio, "int nd_collapse_[%(nd)s] = {"%locals() +','.join(['1' for x in range(nd)]) +"};"
print >> sio, "int nd_collapse_[%(nd)s] = {"%locals() +','.join(['1' for x in xrange(nd)]) +"};"
for ipos in xrange(len(node.inputs)):
if not _logical_scalar(node.inputs[ipos]):
print >> sio, """
int nd_collapse_%(ipos)s[%(nd)s] = {"""%locals() +','.join(['1' for x in range(nd)]) +"};"
int nd_collapse_%(ipos)s[%(nd)s] = {"""%locals() +','.join(['1' for x in xrange(nd)]) +"};"
print >> sio, """
can_collapse_%(nodename)s(nd_collapse, local_dims, local_str[%(ipos)s], nd_collapse_%(ipos)s);
for(int i=0;i<nd_collapse;i++){
......@@ -615,12 +615,12 @@ nd_collapse_[i]=0;
print >>sio, """
std::cerr<< "nd_collapse_%(ipos)s "<<
"""%locals()
print >>sio, ' << " " << '.join(["nd_collapse_%(ipos)s["%locals()+str(i)+"]" for i in range(nd)])
print >>sio, ' << " " << '.join(["nd_collapse_%(ipos)s["%locals()+str(i)+"]" for i in xrange(nd)])
print >>sio, '<< "\\n";'
print >>sio, """
std::cerr<< "nd_collapse_ "<<
"""%locals()
print >>sio, ' << " " << '.join(["nd_collapse_["%locals()+str(i)+"]" for i in range(nd)])
print >>sio, ' << " " << '.join(["nd_collapse_["%locals()+str(i)+"]" for i in xrange(nd)])
print >>sio, '<< "\\n";'
# update the local stride.
......@@ -664,8 +664,8 @@ nd_collapse_[i]=0;
if(nd_collapse_[i]==1)nd_collapse--;
}
if(nd_collapse == 1 """%locals()
l=["local_str[%(ipos)s][nd_collapse-1]==1 "%locals()for ipos in range(len(node.inputs)) if not _logical_scalar(node.inputs[ipos])]
l+=["local_ostr[%(ipos)s][nd_collapse-1]==1 "%locals()for ipos in range(len(node.outputs)) if not _logical_scalar(node.outputs[ipos])]
l=["local_str[%(ipos)s][nd_collapse-1]==1 "%locals()for ipos in xrange(len(node.inputs)) if not _logical_scalar(node.inputs[ipos])]
l+=["local_ostr[%(ipos)s][nd_collapse-1]==1 "%locals()for ipos in xrange(len(node.outputs)) if not _logical_scalar(node.outputs[ipos])]
if len(l)>0:
print >> sio," && "," && ".join(l)
print >> sio,"""){nd_collapse=0;} """
......@@ -679,9 +679,9 @@ nd_collapse_[i]=0;
print >> sio, 'std::cerr << "\\n";'
for ipos in xrange(len(node.inputs)):
print >> sio, 'std::cerr << " local_str %(ipos)s: " <<'%locals()+' << " " << '.join(["local_str[%(ipos)s][%(x)s]"%locals() for x in range(nd)])+'<<"\\n";'
print >> sio, 'std::cerr << " local_str %(ipos)s: " <<'%locals()+' << " " << '.join(["local_str[%(ipos)s][%(x)s]"%locals() for x in xrange(nd)])+'<<"\\n";'
for ipos in xrange(len(node.outputs)):
print >> sio, 'std::cerr << " local_ostr %(ipos)s: " <<'%locals()+' << " " << '.join(["local_ostr[%(ipos)s][%(x)s]"%locals() for x in range(nd)])+'<<"\\n";'
print >> sio, 'std::cerr << " local_ostr %(ipos)s: " <<'%locals()+' << " " << '.join(["local_ostr[%(ipos)s][%(x)s]"%locals() for x in xrange(nd)])+'<<"\\n";'
def launch_Ccontiguous(nodename, scalar_op, sync=True):
......@@ -784,7 +784,7 @@ nd_collapse_[i]=0;
print >> sio, "case 0: {"
launch_Ccontiguous(nodename, scalar_op, self.sync)
print >> sio, " } break;"
for i in range(1, nd+1):
for i in xrange(1, nd+1):
print >> sio, "case "+str(i)+": {"
launch_General(nodename, scalar_op, i, self.sync)
print >> sio, " } break;"
......@@ -800,7 +800,7 @@ nd_collapse_[i]=0;
def c_support_code_apply(self, node, nodename):
nd = node.outputs[0].type.ndim
return "".join(
[self.c_src_kernel(node, nodename,x) for x in range(1,nd+1)]+
[self.c_src_kernel(node, nodename,x) for x in xrange(1,nd+1)]+
[
self.c_src_kernel_Ccontiguous(node, nodename),
self.c_src_callkernel(node, nodename),
......
theano/sandbox/cuda/opt.py
浏览文件 @ 19f7768f
......@@ -478,7 +478,7 @@ def local_gpu_sum(node):
new_in_shp = [x_shape[0]]
new_mask = [reduce_mask[0]]
for i in range(1, x.type.ndim):
for i in xrange(1, x.type.ndim):
if reduce_mask[i] == reduce_mask[i-1]:
new_in_shp[-1] *= x_shape[i]
else:
......@@ -993,7 +993,7 @@ def split_huge_add_or_mul(node):
return False
while len(node.inputs)>max_nb_inputs:
inner_op = []
for i in range(0,len(node.inputs),max_nb_inputs):
for i in xrange(0,len(node.inputs),max_nb_inputs):
inner_op.append(node.op(*node.inputs[i:i+max_nb_inputs]))
node = node.op(*inner_op).owner
return node
......
theano/sandbox/neighbourhoods.py
浏览文件 @ 19f7768f
......@@ -76,9 +76,9 @@ class NeighbourhoodsFromImages(Op):
self.code_string, self.code = self.make_py_code()
def _compute_neigh_strides(self):
neigh_strides = [1 for i in range(len(self.strides))]
neigh_strides = [1 for i in xrange(len(self.strides))]
cur_stride = 1
for i in range(len(self.strides)-1, -1, -1):
for i in xrange(len(self.strides)-1, -1, -1):
neigh_strides[i] = cur_stride
cur_stride *= self.dims_neighbourhoods[i]
return neigh_strides
......@@ -107,7 +107,7 @@ class NeighbourhoodsFromImages(Op):
def out_shape(self, input_shape):
dims = list(input_shape[:self.n_dims_before])
num_strides = [0 for i in range(len(self.strides))]
num_strides = [0 for i in xrange(len(self.strides))]
neigh_flattened_dim = 1
for i, ds in enumerate(self.dims_neighbourhoods):
cur_stride = self.strides[i]
......@@ -238,13 +238,13 @@ class NeighbourhoodsFromImages(Op):
def _py_flattened_idx(self):
return "+".join(["neigh_strides[%d]*neigh_idx_%d" % (i,i) \
for i in range(len(self.strides))])
for i in xrange(len(self.strides))])
def _py_assignment(self):
input_idx = "".join(["outer_idx_%d," % (i,) \
for i in xrange(self.n_dims_before)])
input_idx += "".join(["dim_%d_offset+neigh_idx_%d," % \
(i,i) for i in range(len(self.strides))])
(i,i) for i in xrange(len(self.strides))])
out_idx = "".join(\
["outer_idx_%d," % (i,) for i in \
range(self.n_dims_before)] + \
......
theano/scalar/basic.py
浏览文件 @ 19f7768f
......@@ -1958,9 +1958,9 @@ class Composite(ScalarOp):
raise ValueError("The env to Composite must be exclusively composed of ScalarOp instances.")
subd = dict(zip(inputs,
["%%(i%i)s"%i for i in range(len(inputs))]) +
["%%(i%i)s"%i for i in xrange(len(inputs))]) +
zip(outputs,
["%%(o%i)s"%i for i in range(len(outputs))]))
["%%(o%i)s"%i for i in xrange(len(outputs))]))
for orphan in env.variables: #env.orphans:
if orphan.owner is None and orphan not in env.inputs:
......@@ -2048,9 +2048,9 @@ class Composite(ScalarOp):
raise NotImplementedError("grad is not implemented for Composite")
def c_code(self, node, name, inames, onames, sub):
d = dict(zip(["i%i"%i for i in range(len(inames))],
d = dict(zip(["i%i"%i for i in xrange(len(inames))],
inames) +
zip(["o%i"%i for i in range(len(onames))],
zip(["o%i"%i for i in xrange(len(onames))],
onames),
**sub)
d['name'] = name
......
theano/sparse/sandbox/sp.py
浏览文件 @ 19f7768f
......@@ -382,8 +382,8 @@ class ConvolutionIndices(Op):
# of the column values (order in which you write the values determines how the
# vectorized data will get used later one)
for fmapi in range(inshp[0]): # loop over input features
for n in range(nkern): # loop over number of kernels (nkern=1 for weight sharing)
for fmapi in xrange(inshp[0]): # loop over input features
for n in xrange(nkern): # loop over number of kernels (nkern=1 for weight sharing)
# FOR EACH OUTPUT PIXEL...
for oy in N.arange(lbound[0],ubound[0],dy): # loop over output image height
......
theano/tensor/basic.py
浏览文件 @ 19f7768f
......@@ -1815,7 +1815,7 @@ class SpecifyShape(Op):
def infer_shape(self, node, shapes):
xshape, sshape = shapes
new_shape=[]
for dim in range(node.inputs[0].ndim):
for dim in xrange(node.inputs[0].ndim):
try:
s=get_constant_value(node.inputs[1][dim])
s=as_tensor_variable(s)
......@@ -2677,7 +2677,7 @@ def var(input, axis = None):
#make a pattern that will undo the reduction of dimensions caused by mean
pattern = []
next_dim = 0
for i in range(input_ndim):
for i in xrange(input_ndim):
if i in axis:
pattern.append('x')
else:
......@@ -3928,7 +3928,7 @@ class Rebroadcast(Op):
if len(self.axis) == 0:
broadcast_pattern = []
else:
broadcast_pattern = ['?' for i in range(1+numpy.max(self.axis.keys()))]
broadcast_pattern = ['?' for i in xrange(1+numpy.max(self.axis.keys()))]
for k,v in self.axis.iteritems():
broadcast_pattern[k] = str(int(v))
return '%s{%s}' % (self.__class__.__name__, ','.join(broadcast_pattern))
......@@ -3955,7 +3955,7 @@ class Rebroadcast(Op):
assert len(ishapes)==1
l = []
one = constant(1)
for ax in range(len(ishapes[0])):
for ax in xrange(len(ishapes[0])):
if self.axis.get(ax, False):
l.append(one)
else:
......@@ -3994,7 +3994,7 @@ def patternbroadcast(x, broadcastable):
We apply the opt here not to pollute the graph especially during the gpu optimization
"""
rval = Rebroadcast(*[(i,broadcastable[i]) for i in range(len(broadcastable))])(x)
rval = Rebroadcast(*[(i,broadcastable[i]) for i in xrange(len(broadcastable))])(x)
return theano.tensor.opt.apply_rebroadcast_opt(rval)
class Join(Op):
......@@ -4158,7 +4158,7 @@ class Join(Op):
# 'axis' dimension.
return [gz[[slice(None)] * axis + [slice(idx[k], idx[k + 1])] + \
[slice(None)] * (n_dims - axis - 1)] \
for k in range(len(sizes_along_axis))]
for k in xrange(len(sizes_along_axis))]
def vec_length(self, node):
"""Guess the length of a Join Variable"""
......@@ -4235,7 +4235,7 @@ def shape_padleft(t, n_ones=1):
"""
_t = as_tensor_variable(t)
pattern = ['x']*n_ones + [i for i in range(_t.type.ndim)]
pattern = ['x']*n_ones + [i for i in xrange(_t.type.ndim)]
return DimShuffle(_t.broadcastable, pattern)(_t)
@constructor
......@@ -4246,7 +4246,7 @@ def shape_padright(t, n_ones=1):
"""
_t = as_tensor_variable(t)
pattern = [i for i in range(_t.type.ndim)] + ['x']*n_ones
pattern = [i for i in xrange(_t.type.ndim)] + ['x']*n_ones
return DimShuffle(_t.broadcastable, pattern)(_t)
@constructor
......@@ -4382,7 +4382,7 @@ if 0: #vertical and horizontal stacking are deprecated. Better to use stack() a
out, = out_
assert x.ndim == y.ndim
# Make sure every dimension (save the first) is the same
for i in range(x.ndim): assert i == 0 or x.shape[i] == y.shape[i]
for i in xrange(x.ndim): assert i == 0 or x.shape[i] == y.shape[i]
out[0] = numpy.vstack([x, y])
def grad(self, inp, grads):
"""
......@@ -4390,7 +4390,7 @@ if 0: #vertical and horizontal stacking are deprecated. Better to use stack() a
that way we can do more sanity-checking::
assert x.ndim == y.ndim
# Make sure every dimension (save the first) is the same
for i in range(x.data.ndim): assert i == 0 or x.data.shape[i] == y.shape[i]
for i in xrange(x.data.ndim): assert i == 0 or x.data.shape[i] == y.shape[i]
etc...
"""
x, y = inp
......@@ -4482,13 +4482,13 @@ class Reshape(Op):
#return [tuple([switch(eq(node.inputs[1][i], -1),
# theano.tensor.opt.Shape_i(i)(node.outputs[0]),
# node.inputs[1][i])
# for i in range(self.ndim)]
# for i in xrange(self.ndim)]
# )]
# Here, we only simplify if the shape (node.inputs[1]) is a constant,
# ideally it would suffice to check that it is always non-negative.
oshape = []
for i in range(self.ndim):
for i in xrange(self.ndim):
default_os_i = theano.tensor.opt.Shape_i(i)(node.outputs[0])
try:
os_i = get_constant_value(node.inputs[1][i]).item()
......@@ -4801,15 +4801,15 @@ class PermuteRowElements(Op):
xs0 = x.shape[0]
ys0 = y.shape[0]
if xs0 == ys0:
for i in range(xs0):
for i in xrange(xs0):
self._rec_perform(node, x[i], y[i], inverse, out[i], curdim+1)
elif ys0 == 1 and node.inputs[1].type.broadcastable[curdim]:
# Broadcast y
for i in range(xs0):
for i in xrange(xs0):
self._rec_perform(node, x[i], y[0], inverse, out[i], curdim+1)
elif xs0 == 1 and node.inputs[0].type.broadcastable[curdim]:
# Broadcast x
for i in range(ys0):
for i in xrange(ys0):
self._rec_perform(node, x[0], y[i], inverse, out[i], curdim+1)
else:
raise ValueError('Dimension mismatch: %s, %s' % (xs0, ys0))
......@@ -4850,7 +4850,7 @@ class PermuteRowElements(Op):
# If x has been broadcasted along some axes, we need to sum
# the gradient over these axes, but keep the dimension (as
# broadcastable)
broadcasted_dims = [dim for dim in range(gz.type.ndim)\
broadcasted_dims = [dim for dim in xrange(gz.type.ndim)\
if x.type.broadcastable[dim] and not gz.type.broadcastable[dim]]
gx = Sum(axis = broadcasted_dims)(gx)
......@@ -4858,7 +4858,7 @@ class PermuteRowElements(Op):
# so we need to put them back.
newdims = []
i = 0
for dim in range(gz.type.ndim):
for dim in xrange(gz.type.ndim):
if dim in broadcasted_dims:
newdims.append('x')
else:
......@@ -5348,13 +5348,13 @@ class TensorDotGrad(Op):
_gx = numpy.tensordot(gz, y, [range(x.ndim-len(self.axes[0]),gz.ndim), sum_over_y])
idx = numpy.hstack((sum_over_x, self.axes[0]))
newshapex = numpy.zeros(x.ndim)
newshapex[[newpos for newpos in idx]] = [i for i in range(x.ndim)]
newshapex[[newpos for newpos in idx]] = [i for i in xrange(x.ndim)]
gx[0] = numpy.transpose(_gx, newshapex)
_gy = numpy.tensordot(x, gz, [sum_over_x, range(x.ndim-len(self.axes[0]))])
idy = numpy.hstack((self.axes[1], sum_over_y))
newshapey = numpy.zeros(y.ndim)
newshapey[[newpos for newpos in idy]] = [i for i in range(y.ndim)]
newshapey[[newpos for newpos in idy]] = [i for i in xrange(y.ndim)]
gy[0] = numpy.transpose(_gy, newshapey)
tensordot_grad = TensorDotGrad
......
theano/tensor/blas.py
浏览文件 @ 19f7768f
......@@ -1603,7 +1603,7 @@ def local_dot22_to_dot22scalar(node):
# I'm asserting there are no such inputs here:
assert dot22_idx != mul_idx
assert all((i in (dot22_idx, mul_idx))
for i in range(len(node.inputs)))
for i in xrange(len(node.inputs)))
return [T.mul(m.owner.inputs[1-i],dot)]
elif m.owner and m.owner.op == T.mul:
......
theano/tensor/elemwise_cgen.py
浏览文件 @ 19f7768f
......@@ -315,7 +315,7 @@ def make_reordered_loop(init_loop_orders, olv_index, dtypes, inner_task, sub):
%(ovar)s_loops_it = %(ovar)s_loops.begin();
""" % locals()
for i in range(nnested):
for i in xrange(nnested):
declare_totals += """
int TOTAL_%(i)i = init_totals[%(ovar)s_loops_it->second];
++%(ovar)s_loops_it;
......@@ -357,7 +357,7 @@ def make_reordered_loop(init_loop_orders, olv_index, dtypes, inner_task, sub):
std::vector< std::pair<int, int> >::reverse_iterator %(ovar)s_loops_rit;
""" % locals()
for i in range(nvars):
for i in xrange(nvars):
var = sub["lv%i" % i]
declare_strides_jumps += """
%(ovar)s_loops_rit = %(ovar)s_loops.rbegin();""" % locals()
......@@ -382,7 +382,7 @@ def make_reordered_loop(init_loop_orders, olv_index, dtypes, inner_task, sub):
iterv = 'ITER_%i' % i
total = 'TOTAL_%i' % i
update = ''
for j in range(nvars):
for j in xrange(nvars):
var = sub["lv%i" % j]
update += "%(var)s_iter += %(var)s_jump_l%(i)i;\n" % locals()
......
theano/tensor/nnet/conv.py
浏览文件 @ 19f7768f
......@@ -73,14 +73,14 @@ def conv2d(input, filters, image_shape=None, filter_shape=None,
#accept Constant value for image_shape and filter_shape.
if image_shape is not None:
image_shape = list(image_shape)
for i in range(len(image_shape)):
for i in xrange(len(image_shape)):
if image_shape[i] is not None:
image_shape[i] = get_constant_value(as_tensor_variable(image_shape[i]))
assert str(image_shape[i].dtype).startswith('int')
image_shape[i] = int(image_shape[i])
if filter_shape is not None:
filter_shape = list(filter_shape)
for i in range(len(filter_shape)):
for i in xrange(len(filter_shape)):
if filter_shape[i] is not None:
filter_shape[i] = get_constant_value(as_tensor_variable(filter_shape[i]))
assert str(filter_shape[i].dtype).startswith('int')
......@@ -450,7 +450,7 @@ class ConvOp(Op):
else:
time_unroll_patch = self.speed_unroll_patch_noshape[mode_idx]
time_unroll_batch_kern = 9999999
for i in range(len(self.speed_unroll_batch_kern)):
for i in xrange(len(self.speed_unroll_batch_kern)):
if bsize%self.speed_unroll_batch_kern[i][0]==0 and nkern%self.speed_unroll_batch_kern[i][1]==0:
if self.speed_unroll_batch_kern[i][2+mode_idx]<time_unroll_batch_kern:
time_unroll_batch_kern=self.speed_unroll_batch_kern[i][2+mode_idx]
......@@ -506,9 +506,9 @@ class ConvOp(Op):
else: #full mode not implemented
self.flops=0
for out_row in range(self.outshp[0]):#loop over output row
for out_col in range(self.outshp[0]):#loop over output col
for row in range(self.kshp[0]):#loop over kern row
for out_row in xrange(self.outshp[0]):#loop over output row
for out_col in xrange(self.outshp[0]):#loop over output col
for row in xrange(self.kshp[0]):#loop over kern row
if (row+out_row-self.kshp[0]+1<0 or
row+out_row-self.kshp[0]+1>=self.imshp[1]):
......@@ -664,10 +664,10 @@ class ConvOp(Op):
val = _valfrommode(self.out_mode)
bval = _bvalfromboundary('fill')
for b in range(bsize):
for n in range(nkern):
for b in xrange(bsize):
for n in xrange(nkern):
zz[b,n,...].fill(0)
for im0 in range(stacklen):
for im0 in xrange(stacklen):
zz[b,n,...] += _convolve2d(\
img2d[b,im0,...], filtersflipped[n,im0,...],1,val, bval, 0)
......@@ -681,12 +681,12 @@ class ConvOp(Op):
img2d = img2d2
#N_image_shape = image_data.shape
for b in range(bsize):
for n in range(nkern):
for b in xrange(bsize):
for n in xrange(nkern):
zz[b,n,...].fill(0)
for im0 in range(stacklen):
for row in range(0,zz.shape[2],self.dx):
for col in range(0,zz.shape[3],self.dy):
for im0 in xrange(stacklen):
for row in xrange(0,zz.shape[2],self.dx):
for col in xrange(0,zz.shape[3],self.dy):
zz[b,n,row,col] += (img2d[b,im0,row:row+kshp[0],col:col+kshp[1]]*\
filtersflipped[n,im0,::-1,::-1]).sum()
......@@ -1555,16 +1555,16 @@ def gen_conv_code_unroll_batch_kern(d,unroll_bsize=1, unroll_ksize=1):
d["unroll_ksize"]=unroll_ksize
def my_dup(st,size):
s=""
for i in range(size):
for i in xrange(size):
d["unroll_iter"]=i
s+=st%d
return s+"\n"
def my_dup2(st):
s=""
iter=0
for i in range(unroll_bsize):
for i in xrange(unroll_bsize):
d["unroll_biter"]=i
for j in range(unroll_ksize):
for j in xrange(unroll_ksize):
d["unroll_kiter"]=j
d["unroll_iter"]=iter
iter+=1
......
theano/tensor/opt.py
浏览文件 @ 19f7768f
......@@ -1043,7 +1043,7 @@ class Assert(T.Op):
out = onames[0]
check = []
fail = sub['fail']
for idx in range(len(inames)-1):
for idx in xrange(len(inames)-1):
i=inames[idx+1]
dtype=node.inputs[idx+1].dtype
check.append('if(!((npy_%(dtype)s*)PyArray_DATA(%(i)s))[0]){PyErr_SetString(PyExc_AssertionError,"Theano Assert failed!");%(fail)s}'%locals())
......@@ -1138,7 +1138,7 @@ def local_alloc_elemwise(node):
assert i.type.ndim == cmp_op.ndim
if theano.config.experimental.local_alloc_elemwise_assert:
assert_op = assert_(assert_op,*[T.eq(i.shape[idx],cmp_op.shape[idx])\
for idx in range(i.type.ndim) \
for idx in xrange(i.type.ndim) \
if not i.type.broadcastable[idx]])
new.append(i.owner.inputs[0])
elif i.owner and isinstance(i.owner.op, T.DimShuffle) and i.owner.inputs[0].owner \
......@@ -1146,7 +1146,7 @@ def local_alloc_elemwise(node):
assert i.type.ndim == cmp_op.type.ndim
if theano.config.experimental.local_alloc_elemwise_assert:
assert_op = assert_(assert_op,*[T.eq(i.shape[idx],cmp_op.shape[idx]) for idx \
in range(i.type.ndim) if not i.type.broadcastable[idx]])
in xrange(i.type.ndim) if not i.type.broadcastable[idx]])
new.append(i.owner.inputs[0].owner.inputs[0])
else: new.append(i)
new[no_broad_idx]=assert_op
......@@ -1322,7 +1322,7 @@ def local_subtensor_lift(node):
Handles the following unary ops:
elemwise(x,...)[idx] -> elemwise(x[idx],...)
when x,... are broadcasted scalar or not broadcasted at all
rebroadcast(x)[idx] => rebroadcast(x[idx])
rebroadcast(x)[idx] => rebroadcast(x[idx])
"""
if isinstance(node.op, T.Subtensor):
u = node.inputs[0]
......@@ -1375,7 +1375,7 @@ def local_subtensor_lift(node):
if isinstance(x, slice):
new_axis += [(j, u.broadcastable[i])]
j += 1
# now keep the broadcastable pattern of all
# now keep the broadcastable pattern of all
# items not appearing in subtensor list
for i in xrange(len(node.op.idx_list), len(u.broadcastable)):
new_axis += [(j,u.broadcastable[i])]
......@@ -1897,7 +1897,7 @@ def local_remove_switch_const_cond(node):
out = T.cast(out, node.outputs[0].dtype)
if out.type.broadcastable != node.outputs[0].type.broadcastable:
# We need to copy data to the new dimensions during execution
out = T.alloc(out, *[node.outputs[0].shape[i] for i in range(out.ndim)])
out = T.alloc(out, *[node.outputs[0].shape[i] for i in xrange(out.ndim)])
return [out]
return False
......@@ -2723,11 +2723,11 @@ def local_sum_alloc(node):
val = get_constant_value(input)
assert val.size == 1
val = val.reshape(1)[0]
to_prod = [shapes[i] for i in range(len(shapes)) if i in node.op.axis]
to_prod = [shapes[i] for i in xrange(len(shapes)) if i in node.op.axis]
if to_prod:
val *= T.mul(*to_prod)
return [T.alloc(T.cast(val, dtype=node.outputs[0].dtype),
*[shapes[i] for i in range(len(shapes)) if i not in node.op.axis])]
*[shapes[i] for i in xrange(len(shapes)) if i not in node.op.axis])]
except TypeError, e:
pass
......@@ -2905,7 +2905,7 @@ def local_pow_specialize_device(node):
pow2 = [xsym]
pow2_scal = [theano.scalar.Scalar(xsym.dtype)()]
y_to_do = abs(y)
for i in range(int(numpy.log2(y_to_do))):
for i in xrange(int(numpy.log2(y_to_do))):
pow2.append(T.sqr(pow2[i]))
pow2_scal.append(theano.scalar.sqr(pow2_scal[i]))
rval1 = None
......@@ -3546,7 +3546,7 @@ def local_grad_log_erfc_neg(node):
mul_inputs = check_input(mul_neg.owner.inputs)
#put the constant first
for i in range(len(mul_inputs)):
for i in xrange(len(mul_inputs)):
if isinstance(i, Constant):
if i==0:
break
......@@ -3636,7 +3636,7 @@ a64=[x*((1-1/(2*x**2)+3/(4*x**4)-15/(8*x**6))**(-1))*numpy.sqrt(numpy.pi) for x
a32=[x*((1-1/(2*x**2)+3/(4*x**4)-15/(8*x**6))**(-1))*numpy.float32(numpy.sqrt(numpy.pi)) for x in numpy.asarray(r,dtype='float32')]
for idx,(a,b,c,d,e) in enumerate(zip(r,p64,p32,a64,a32)):print a,b,c,d,e,c-b,e-b,numpy.absolute(c-b)<numpy.absolute(e-b)
for i in range(1,len(p32)): print r[i], p32[i]-p32[i-1]#, show that the value don't look stable at some point before inf.
for i in xrange(1,len(p32)): print r[i], p32[i]-p32[i-1]#, show that the value don't look stable at some point before inf.
#float64 threshold is 26.63 the approx seam more precise at that point.
r = numpy.arange(26.2,26.7,.001)
......@@ -3649,7 +3649,7 @@ a128=[x*((1-1/(2*x**2)+3/(4*x**4)-15/(8*x**6))**(-1))*numpy.float128(numpy.sqrt(
a64=[x*((1-1/(2*x**2)+3/(4*x**4)-15/(8*x**6)+63/(7*x**8))**(-1))*numpy.sqrt(numpy.pi) for x in r]
for a,b,c,d in zip(r,p128,p64,a64):print a,b,c,d,c-b,d-b
for i in range(1,len(p64)): print i, 64[i]-p64[i-1]
for i in xrange(1,len(p64)): print i, 64[i]-p64[i-1]
"""
# ###############
......
theano/tensor/raw_random.py
浏览文件 @ 19f7768f
......@@ -357,7 +357,7 @@ def _generate_broadcasting_indices(out_shape, *shapes):
# Will contain the return value: a list of indices for each argument
ret_indices = [ [()] for shape in all_shapes ]
for dim in range(len(out_shape)):
for dim in xrange(len(out_shape)):
# Temporary list to generate the indices
_ret_indices = [ [] for shape in all_shapes ]
......@@ -377,7 +377,7 @@ def _generate_broadcasting_indices(out_shape, *shapes):
for prev_index in zip(*ret_indices):
for dim_index in zip(*ranges):
for i in range(len(all_shapes)):
for i in xrange(len(all_shapes)):
_ret_indices[i].append(prev_index[i] + (dim_index[i],))
ret_indices = _ret_indices
......@@ -488,7 +488,7 @@ def random_integers_helper(random_state, low, high, size):
out_size = tuple(size)
else:
out_size = ()
for dim in range(out_ndim):
for dim in xrange(out_ndim):
dim_len = max(low.shape[dim], high.shape[dim])
out_size = out_size + (dim_len,)
......@@ -602,7 +602,7 @@ def multinomial_helper(random_state, n, pvals, size):
size = tuple(size)
else:
size = ()
for dim in range(ndim):
for dim in xrange(ndim):
dim_len = max(n.shape[dim], pvals.shape[dim])
size = size + (dim_len,)
out_size = size+(pvals.shape[-1],)
......
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