Skip to content

P
pytensor
提交 8ae399a1 authored 作者: Razvan Pascanu's avatar Razvan Pascanu
浏览文件
操作

I used to store information either in self as properties or in the info

dictionary and from time to time tried to sync the two. As Josh pointed out, this was confusing, so I changed the code to use only self. I left the usage of the info dictionary to construct the op and in __eq__ and __hash__ because at those points I don't care about the individual terms in the dictionary and it would make the code considerably bigger to spell out all keys in the info dictionary.
上级 09bd9be8
隐藏空白字符变更
内嵌 并排
正在显示 包含 42 行增加 和 36 行删除
theano/scan_module/scan_op.py
浏览文件 @ 8ae399a1
...@@ -62,22 +62,25 @@ class Scan(Op): ...@@ -62,22 +62,25 @@ class Scan(Op):
the scan op. the scan op.
""" """
# adding properties into self # adding properties into self
self.info = info
self.inputs = inputs self.inputs = inputs
self.outputs = outputs self.outputs = outputs
self.__dict__.update(info) self.__dict__.update(info)
# I keep a version of info in self, to use in __eq__ and __hash__,
# since info contains all tunable parameters of the op, so for two
# scan to be equal this tunable parameters should be the same
self.info = info
# build a list of output types for any Apply node using this op. # build a list of output types for any Apply node using this op.
info['output_types'] = [] self.output_types = []
idx = 0 idx = 0
jdx = 0 jdx = 0
if info['gpu']: if self.gpu:
# mit_mot # mit_mot
while idx < self.n_mit_mot_outs: while idx < self.n_mit_mot_outs:
# Not that for mit_mot there are several output slices per # Not that for mit_mot there are several output slices per
# output sequence # output sequence
o = outputs[idx] o = outputs[idx]
info['output_types'].append( self.output_types.append(
cuda.CudaNdarrayType( cuda.CudaNdarrayType(
broadcastable = (False,) + o.type.broadcastable)) broadcastable = (False,) + o.type.broadcastable))
idx += len(self.mit_mot_out_slices[jdx]) idx += len(self.mit_mot_out_slices[jdx])
...@@ -86,22 +89,22 @@ class Scan(Op): ...@@ -86,22 +89,22 @@ class Scan(Op):
# mit_sot / sit_sot / nit_sot # mit_sot / sit_sot / nit_sot
end = idx + self.n_mit_sot + self.n_sit_sot + self.n_nit_sot end = idx + self.n_mit_sot + self.n_sit_sot + self.n_nit_sot
for o in outputs[idx:end]: for o in outputs[idx:end]:
info['output_types'].append( self.output_types.append(
cuda.CudaNdarrayType( broadcastable = (False,) + cuda.CudaNdarrayType( broadcastable = (False,) +
o.type.broadcastable)) o.type.broadcastable))
# shared outputs # shared outputs
for o in outputs[end:]: for o in outputs[end:]:
if isinstance(o.type, TensorType): if isinstance(o.type, TensorType):
info['output_types'].append(cuda.CudaNdarrayType( self.output_types.append(cuda.CudaNdarrayType(
broadcastable = o.type.broadcastable)) broadcastable = o.type.broadcastable))
else: else:
info['output_types'].append( o.type ) self.output_types.append( o.type )
else: else:
while idx < self.n_mit_mot_outs: while idx < self.n_mit_mot_outs:
# Not that for mit_mot there are several output slices per # Not that for mit_mot there are several output slices per
# output sequence # output sequence
o = outputs[idx] o = outputs[idx]
info['output_types'].append( self.output_types.append(
TensorType( TensorType(
broadcastable = (False,) + o.type.broadcastable broadcastable = (False,) + o.type.broadcastable
, dtype = o.type.dtype) , dtype = o.type.dtype)
...@@ -112,7 +115,7 @@ class Scan(Op): ...@@ -112,7 +115,7 @@ class Scan(Op):
# mit_sot / sit_sot / nit_sot # mit_sot / sit_sot / nit_sot
end = idx + self.n_mit_sot + self.n_sit_sot + self.n_nit_sot end = idx + self.n_mit_sot + self.n_sit_sot + self.n_nit_sot
for o in outputs[idx:end]: for o in outputs[idx:end]:
info['output_types'].append( self.output_types.append(
TensorType( TensorType(
broadcastable = (False,) + o.type.broadcastable broadcastable = (False,) + o.type.broadcastable
, dtype = o.type.dtype )) , dtype = o.type.dtype ))
...@@ -120,50 +123,55 @@ class Scan(Op): ...@@ -120,50 +123,55 @@ class Scan(Op):
for o in outputs[end:]: for o in outputs[end:]:
if cuda.cuda_available and isinstance(o.type, if cuda.cuda_available and isinstance(o.type,
cuda.CudaNdarrayType): cuda.CudaNdarrayType):
info['output_types'].append( TensorType( self.output_types.append( TensorType(
broadcastable = o.type.broadcastable broadcastable = o.type.broadcastable
, dtype = theano.config.floatX) ) , dtype = theano.config.floatX) )
else: else:
info['output_types'].append( o.type ) self.output_types.append( o.type )
self.destroy_map = {} self.destroy_map = {}
if 'inplace' in info and info['inplace']: if hasattr(self,'inplace') and self.inplace:
for idx in xrange(info['n_mit_mot'] + info['n_mit_sot'] + for idx in xrange(self.n_mit_mot + self.n_mit_sot +
info['n_sit_sot'] ): self.n_sit_sot ):
self.destroy_map[idx] = [idx + 1 + info['n_seqs']] self.destroy_map[idx] = [idx + 1 + self.n_seqs]
# I consider all inputs of the inner function non mutable # I consider all inputs of the inner function non mutable
nonmutable = range(len(inputs)) nonmutable = range(len(inputs))
mode_instance = compile.mode.get_mode(info['mode']) mode_instance = compile.mode.get_mode(self.mode)
# if the default mode is used, and that mode is ProfileMode # if the default mode is used, and that mode is ProfileMode
# then we need to copy the mode otherwise the time for a given # then we need to copy the mode otherwise the time for a given
# op will be counted multiple times # op will be counted multiple times
if ( info['mode'] is None and if ( self.mode is None and
isinstance(mode_instance, compile.profilemode.ProfileMode) ): isinstance(mode_instance, compile.profilemode.ProfileMode) ):
mode_instance = compile.profilemode.ProfileMode( mode_instance = compile.profilemode.ProfileMode(
optimizer = mode_instance.provided_optimizer optimizer = mode_instance.provided_optimizer
, linker = mode_instance.provided_linker ) , linker = mode_instance.provided_linker )
compile.profilemode.prof_mode_instance_to_print.append(mode_instance) compile.profilemode.prof_mode_instance_to_print.append(mode_instance)
info['mode_instance'] = mode_instance self.mode_instance = mode_instance
if self.name: if self.name:
info['mode_instance'].message = self.name + " sub profile" self.mode_instance.message = self.name + " sub profile"
else: else:
info['mode_instance'].message = "Scan sub profile" self.mode_instance.message = "Scan sub profile"
else: else:
info['mode_instance'] = mode_instance self.mode_instance = mode_instance
if 'name' not in info or info['name'] is None: if not hasattr(self,'name') or self.name is None:
info['name'] = 'scan_fn' self.name = 'scan_fn'
# to have a fair __eq__ comparison later on, we update the info with
# the actual mode used to compile the function and the name of the
# function that we set in case none was given
self.info['name'] = self.name
self.info['mode_instance'] = self.mode_instance
if isinstance(info['mode_instance'], compile.debugmode.DebugMode): if isinstance(self.mode_instance, compile.debugmode.DebugMode):
theano_fn = function( theano_fn = function(
inputs inputs
, outputs , outputs
, mode = info['mode_instance'] , mode = self.mode_instance
, name = info['name'] ) , name = self.name )
def fn_wrapper(ins_storage, outs_storage): def fn_wrapper(ins_storage, outs_storage):
''' '''
...@@ -188,20 +196,18 @@ class Scan(Op): ...@@ -188,20 +196,18 @@ class Scan(Op):
inputs inputs
, outputs , outputs
, nonmutable , nonmutable
, mode = info['mode_instance'] , mode = self.mode_instance
, name = info['name'] , name = self.name
, slices = ( info['n_mit_mot_outs'] + , slices = ( self.n_mit_mot_outs +
info['n_mit_sot'] + self.n_mit_sot +
info['n_sit_sot'] + self.n_sit_sot +
info['n_nit_sot'] ) self.n_nit_sot )
) )
# check for shared variables in the inputs # check for shared variables in the inputs
assert not numpy.any( [isinstance(x, SharedVariable) for x assert not numpy.any( [isinstance(x, SharedVariable) for x
in self.fn.maker.inputs]) in self.fn.maker.inputs])
self.__dict__.update(info)
self.info = info
# Pre-computing some values to speed up perform # Pre-computing some values to speed up perform
self.mintaps = [ numpy.min(x) for x in self.tap_array] self.mintaps = [ numpy.min(x) for x in self.tap_array]
self.mintaps += [ 0 for x in xrange(self.n_nit_sot) ] self.mintaps += [ 0 for x in xrange(self.n_nit_sot) ]
...@@ -320,7 +326,7 @@ class Scan(Op): ...@@ -320,7 +326,7 @@ class Scan(Op):
apply_node = Apply(self apply_node = Apply(self
, inputs , inputs
, [t() for t in self.info['output_types']]) , [t() for t in self.output_types])
return apply_node return apply_node
def __eq__(self, other): def __eq__(self, other):
...@@ -522,7 +528,7 @@ class Scan(Op): ...@@ -522,7 +528,7 @@ class Scan(Op):
outs[j][0].shape[0] < store_steps[j] or outs[j][0].shape[0] < store_steps[j] or
outs[j][0].shape[1:] != shape[1:] or outs[j][0].shape[1:] != shape[1:] or
outs[j][0].dtype != dtype ): outs[j][0].dtype != dtype ):
if self.info['gpu']: if self.gpu:
outs[j][0] = cuda.cuda_ndarray.cuda_ndarray.CudaNdarray.zeros(shape) outs[j][0] = cuda.cuda_ndarray.cuda_ndarray.CudaNdarray.zeros(shape)
else: else:
outs[j][0] = numpy.zeros(shape, dtype) outs[j][0] = numpy.zeros(shape, dtype)
......
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