Skip to content

P
pytensor
提交 cfc588cc authored 作者: Razvan Pascanu's avatar Razvan Pascanu
浏览文件
操作

Removed trailing spaces

上级 ccb73dab
隐藏空白字符变更
内嵌 并排
正在显示 包含 164 行增加 和 166 行删除
theano/tensor/basic.py
浏览文件 @ cfc588cc
...@@ -57,7 +57,7 @@ __oplist_constructor_list = [] ...@@ -57,7 +57,7 @@ __oplist_constructor_list = []
"""List of functions to be listed as op constructors in the oplist (`gen_oplist`, doc/oplist.txt).""" """List of functions to be listed as op constructors in the oplist (`gen_oplist`, doc/oplist.txt)."""
def constructor(f): def constructor(f):
"""Add `f` to :doc:`oplist`. """Add `f` to :doc:`oplist`.
Make `f` appear as a constructor in the oplist (`gen_oplist`, doc/oplist.txt). Make `f` appear as a constructor in the oplist (`gen_oplist`, doc/oplist.txt).
""" """
__oplist_constructor_list.append(f) __oplist_constructor_list.append(f)
...@@ -80,7 +80,7 @@ if 0: ...@@ -80,7 +80,7 @@ if 0:
if hasattr(x, '_as_CudaNdarrayVariable'): if hasattr(x, '_as_CudaNdarrayVariable'):
return x._as_CudaNdarrayVariable() #TODO: pass name and ndim arguments return x._as_CudaNdarrayVariable() #TODO: pass name and ndim arguments
return as_tensor_variable(x, name, ndim) return as_tensor_variable(x, name, ndim)
def as_tensor_variable(x, name = None, ndim=None): def as_tensor_variable(x, name = None, ndim=None):
"""Return `x`, transformed into a `TensorType` """Return `x`, transformed into a `TensorType`
...@@ -158,7 +158,7 @@ class NumpyAutocaster(object): ...@@ -158,7 +158,7 @@ class NumpyAutocaster(object):
When config.floatX is float32 (at the time of calling), then this function downcasts float When config.floatX is float32 (at the time of calling), then this function downcasts float
and numpy.float arguments to numpy.float32, if float32 is in the self.dtypes list. and numpy.float arguments to numpy.float32, if float32 is in the self.dtypes list.
Python ints are always 64bit and floats are always double precision. Python ints are always 64bit and floats are always double precision.
This class uses the algorithm in __call__ to use a narrower dtype when no precision would This class uses the algorithm in __call__ to use a narrower dtype when no precision would
be lost, and to even lose precision when this is demanded by the list of dtypes (e.g. to be lost, and to even lose precision when this is demanded by the list of dtypes (e.g. to
...@@ -182,7 +182,7 @@ class NumpyAutocaster(object): ...@@ -182,7 +182,7 @@ class NumpyAutocaster(object):
# recall: float is numpy.float # recall: float is numpy.float
if isinstance(x, float) and config.floatX in self.dtypes and config.floatX == 'float32': if isinstance(x, float) and config.floatX in self.dtypes and config.floatX == 'float32':
return theano._asarray(x, dtype='float32') return theano._asarray(x, dtype='float32')
for dtype in self.dtypes: for dtype in self.dtypes:
x_ = theano._asarray(x, dtype=dtype) x_ = theano._asarray(x, dtype=dtype)
if numpy.all(x == x_): if numpy.all(x == x_):
...@@ -200,7 +200,7 @@ autocast_float = NumpyAutocaster(('float32', 'float64')) ...@@ -200,7 +200,7 @@ autocast_float = NumpyAutocaster(('float32', 'float64'))
# this autocasting, and in future, our ops might be smarter about factoring out upcasts. The # this autocasting, and in future, our ops might be smarter about factoring out upcasts. The
# advantage of this mechanism is to combine it with floatX so that 1.0 + xmatrix() will always # advantage of this mechanism is to combine it with floatX so that 1.0 + xmatrix() will always
# have the same type as the xmatrix(). # have the same type as the xmatrix().
# #
class autocast_float_as(object): class autocast_float_as(object):
"""This class makes it possible to temporarily and locally adjust autocasting behaviour. """This class makes it possible to temporarily and locally adjust autocasting behaviour.
...@@ -222,7 +222,7 @@ class autocast_float_as(object): ...@@ -222,7 +222,7 @@ class autocast_float_as(object):
def constant_or_value(x, rtype, name=None, ndim=None, dtype=None): def constant_or_value(x, rtype, name=None, ndim=None, dtype=None):
"""Return a symbolic `Constant` with value `x` """Return a symbolic `Constant` with value `x`
:Exceptions: :Exceptions:
- `TypeError`: `x` could not be converted to a numpy.ndarray - `TypeError`: `x` could not be converted to a numpy.ndarray
- `ValueError`: `x` could not be expanded to have ndim dimensions - `ValueError`: `x` could not be expanded to have ndim dimensions
...@@ -295,19 +295,19 @@ if int(config.tensor.cmp_sloppy)>1: ...@@ -295,19 +295,19 @@ if int(config.tensor.cmp_sloppy)>1:
# useful to test the GPU as they don't use extended precision and # useful to test the GPU as they don't use extended precision and
# this cause some difference bigger then the normal sloppy. # this cause some difference bigger then the normal sloppy.
float32_atol = 5e-4 float32_atol = 5e-4
float32_rtol = 1e-3 float32_rtol = 1e-3
float64_rtol = 1e-4 float64_rtol = 1e-4
float64_atol = 1e-3 float64_atol = 1e-3
elif int(config.tensor.cmp_sloppy): elif int(config.tensor.cmp_sloppy):
float32_atol = 1e-4 float32_atol = 1e-4
float32_rtol = 1e-3 float32_rtol = 1e-3
float64_rtol = 1e-4 float64_rtol = 1e-4
float64_atol = 1e-3 float64_atol = 1e-3
else: else:
#If you change those value in test don't forget to put them back when the test end. #If you change those value in test don't forget to put them back when the test end.
#Don't forget the case when the test fail. #Don't forget the case when the test fail.
float32_atol = 1e-5 float32_atol = 1e-5
float32_rtol = 1e-3 float32_rtol = 1e-3
# defaults in numpy.allclose # defaults in numpy.allclose
float64_rtol = 1.0000000000000001e-05 float64_rtol = 1.0000000000000001e-05
...@@ -395,7 +395,7 @@ class TensorType(Type): ...@@ -395,7 +395,7 @@ class TensorType(Type):
if self.dtype=='floatX': if self.dtype=='floatX':
self.dtype=config.floatX self.dtype=config.floatX
### broadcastable is immutable, and all elements are either True or False ### broadcastable is immutable, and all elements are either True or False
self.broadcastable = tuple(bool(b) for b in broadcastable) self.broadcastable = tuple(bool(b) for b in broadcastable)
self.dtype_specs() # error checking is done there self.dtype_specs() # error checking is done there
self.name = name self.name = name
self.numpy_dtype = numpy.dtype(self.dtype) self.numpy_dtype = numpy.dtype(self.dtype)
...@@ -438,12 +438,12 @@ class TensorType(Type): ...@@ -438,12 +438,12 @@ class TensorType(Type):
except Exception, e: except Exception, e:
return str(e) return str(e)
return "value is valid" return "value is valid"
def dtype_specs(self): def dtype_specs(self):
"""Return a tuple (python type, c type, numpy typenum) that corresponds to """Return a tuple (python type, c type, numpy typenum) that corresponds to
self.dtype. self.dtype.
This function is used internally as part of C code generation. This function is used internally as part of C code generation.
""" """
#TODO: add more type correspondances for e.g. int32, int64, float32, #TODO: add more type correspondances for e.g. int32, int64, float32,
...@@ -483,7 +483,7 @@ class TensorType(Type): ...@@ -483,7 +483,7 @@ class TensorType(Type):
a_eq_b = (a==b) a_eq_b = (a==b)
r = numpy.all(a_eq_b) r = numpy.all(a_eq_b)
if r: return True if r: return True
# maybe the trouble is that there are NaNs # maybe the trouble is that there are NaNs
a_missing = numpy.isnan(a) a_missing = numpy.isnan(a)
if a_missing.any(): if a_missing.any():
b_missing = numpy.isnan(b) b_missing = numpy.isnan(b)
...@@ -546,7 +546,7 @@ class TensorType(Type): ...@@ -546,7 +546,7 @@ class TensorType(Type):
#set it to False #set it to False
cmp_elemwise = numpy.where(both_inf&cmp_elemwise, cmp_elemwise = numpy.where(both_inf&cmp_elemwise,
a==b,cmp_elemwise) a==b,cmp_elemwise)
#check the sign of the inf #check the sign of the inf
both_inf = numpy.where(both_inf,a==b,both_inf) both_inf = numpy.where(both_inf,a==b,both_inf)
...@@ -554,7 +554,7 @@ class TensorType(Type): ...@@ -554,7 +554,7 @@ class TensorType(Type):
both_inf += a_inf both_inf += a_inf
if allow_remove_nan: if allow_remove_nan:
both_missing += a_missing both_missing += a_missing
# Combine all information. # Combine all information.
return (cmp_elemwise + both_missing + both_inf).all() return (cmp_elemwise + both_missing + both_inf).all()
...@@ -885,7 +885,7 @@ class _tensor_py_operators: ...@@ -885,7 +885,7 @@ class _tensor_py_operators:
def __abs__(self): return abs_(self) def __abs__(self): return abs_(self)
def __neg__(self): return neg(self) def __neg__(self): return neg(self)
#CASTS #CASTS
#### REMOVED THESE BECAUSE PYTHON appears to require __int__ to return an int. -JB 20081112 #### REMOVED THESE BECAUSE PYTHON appears to require __int__ to return an int. -JB 20081112
#def __int__(self): return convert_to_int32(self) #def __int__(self): return convert_to_int32(self)
#def __float__(self): return convert_to_float64(self) #def __float__(self): return convert_to_float64(self)
...@@ -898,7 +898,7 @@ class _tensor_py_operators: ...@@ -898,7 +898,7 @@ class _tensor_py_operators:
def __ge__(self,other): return ge(self, other) def __ge__(self,other): return ge(self, other)
#BITWISE #BITWISE
def __invert__(self): return invert(self) def __invert__(self): return invert(self)
def __and__(self,other): return and_(self, other) def __and__(self,other): return and_(self, other)
def __or__(self,other): return or_(self, other) def __or__(self,other): return or_(self, other)
def __xor__(self,other): return xor(self, other) def __xor__(self,other): return xor(self, other)
...@@ -910,27 +910,27 @@ class _tensor_py_operators: ...@@ -910,27 +910,27 @@ class _tensor_py_operators:
# def __ixor__(self, other): return _xor_inplace(self, other) # def __ixor__(self, other): return _xor_inplace(self, other)
#ARITHMETIC - NORMAL #ARITHMETIC - NORMAL
def __add__(self,other): def __add__(self,other):
try: try:
return add(self,other) return add(self,other)
except Exception, e: except Exception, e:
return NotImplemented return NotImplemented
def __sub__(self,other): def __sub__(self,other):
try: try:
return sub(self,other) return sub(self,other)
except Exception, e: except Exception, e:
return NotImplemented return NotImplemented
def __mul__(self,other): def __mul__(self,other):
try: try:
return mul(self,other) return mul(self,other)
except Exception, e: except Exception, e:
return NotImplemented return NotImplemented
def __div__(self,other): def __div__(self,other):
try: try:
return div_proxy(self,other) return div_proxy(self,other)
except Exception, e: except Exception, e:
return NotImplemented return NotImplemented
def __pow__(self,other): def __pow__(self,other):
try: try:
return pow(self,other) return pow(self,other)
except Exception, e: except Exception, e:
...@@ -1031,12 +1031,12 @@ class _tensor_py_operators: ...@@ -1031,12 +1031,12 @@ class _tensor_py_operators:
def __getslice__(self, *args): def __getslice__(self, *args):
args = slice(*args), args = slice(*args),
return self.__getitem__(args) return self.__getitem__(args)
#COPYING #COPYING
def copy(self): def copy(self):
return tensor_copy(self) return tensor_copy(self)
def __iter__(self): def __iter__(self):
try: try:
for i in xrange(get_vector_length(self)): for i in xrange(get_vector_length(self)):
yield self[i] yield self[i]
...@@ -1044,7 +1044,7 @@ class _tensor_py_operators: ...@@ -1044,7 +1044,7 @@ class _tensor_py_operators:
# This prevents accidental iteration via builtin.sum(self) # This prevents accidental iteration via builtin.sum(self)
raise TypeError('TensorType does not support iteration. ' raise TypeError('TensorType does not support iteration. '
'Maybe you are using builtin.sum instead of theano.tensor.sum? (Maybe .max?)') 'Maybe you are using builtin.sum instead of theano.tensor.sum? (Maybe .max?)')
# CONVENIENT ACCESS TO TYPE PROPERTIES # CONVENIENT ACCESS TO TYPE PROPERTIES
ndim = property(lambda self: self.type.ndim) ndim = property(lambda self: self.type.ndim)
...@@ -1053,7 +1053,7 @@ class _tensor_py_operators: ...@@ -1053,7 +1053,7 @@ class _tensor_py_operators:
"""The broadcastable signature of this tensor. """The broadcastable signature of this tensor.
See :doc:`broadcasting` for details. See :doc:`broadcasting` for details.
""" """
dtype = property(lambda self: self.type.dtype) dtype = property(lambda self: self.type.dtype)
""" The dtype of this tensor. """ """ The dtype of this tensor. """
...@@ -1095,7 +1095,7 @@ class _tensor_py_operators: ...@@ -1095,7 +1095,7 @@ class _tensor_py_operators:
def get_constant_value(self): def get_constant_value(self):
return get_constant_value(self) return get_constant_value(self)
class TensorVariable(Variable, _tensor_py_operators): class TensorVariable(Variable, _tensor_py_operators):
"""Subclass to add the tensor operators to the basic `Variable` class.""" """Subclass to add the tensor operators to the basic `Variable` class."""
TensorType.Variable = TensorVariable TensorType.Variable = TensorVariable
...@@ -1115,7 +1115,7 @@ class TensorConstantSignature(tuple): ...@@ -1115,7 +1115,7 @@ class TensorConstantSignature(tuple):
#N.B. compare shape to ensure no broadcasting in == #N.B. compare shape to ensure no broadcasting in ==
#N.B. compare elementwise last because it is the most expensive check #N.B. compare elementwise last because it is the most expensive check
return (t0 == t1) and (d0.shape == d1.shape) \ return (t0 == t1) and (d0.shape == d1.shape) \
and (self.sum == other.sum) and (numpy.all(d0 == d1)) and (self.sum == other.sum) and (numpy.all(d0 == d1))
def __hash__(self): def __hash__(self):
t, d = self t, d = self
return hashtype(self) ^ hash(t) ^ hash(d.shape) ^ hash(self.sum) return hashtype(self) ^ hash(t) ^ hash(d.shape) ^ hash(self.sum)
...@@ -1130,7 +1130,7 @@ class TensorConstantSignature(tuple): ...@@ -1130,7 +1130,7 @@ class TensorConstantSignature(tuple):
class TensorConstant(Constant, _tensor_py_operators): class TensorConstant(Constant, _tensor_py_operators):
"""Subclass to add the tensor operators to the basic `Constant` class. """Subclass to add the tensor operators to the basic `Constant` class.
To create a TensorConstant, use the `constant` function in this module. To create a TensorConstant, use the `constant` function in this module.
""" """
def signature(self): def signature(self):
...@@ -1139,7 +1139,7 @@ TensorType.Constant = TensorConstant ...@@ -1139,7 +1139,7 @@ TensorType.Constant = TensorConstant
class TensorValue(Value, _tensor_py_operators): class TensorValue(Value, _tensor_py_operators):
"""Subclass to add the tensor operators to the basic `Value` class. """Subclass to add the tensor operators to the basic `Value` class.
To create a TensorValue, use the `value` function in this module. To create a TensorValue, use the `value` function in this module.
:note: Value is deprecated by SharedVariable :note: Value is deprecated by SharedVariable
...@@ -1167,8 +1167,8 @@ def _elemwise(scalar_op, name, doc_prefix=''): ...@@ -1167,8 +1167,8 @@ def _elemwise(scalar_op, name, doc_prefix=''):
inplace = elemwise.Elemwise(inplace_scalar_op, {0: 0}, name = name+"_inplace") inplace = elemwise.Elemwise(inplace_scalar_op, {0: 0}, name = name+"_inplace")
# don't add the inplace versions, they aren't supposed to be part of the user interface # don't add the inplace versions, they aren't supposed to be part of the user interface
_constructor_list.append(straight) _constructor_list.append(straight)
# This is here so that gen_oplist can detect which module declared these variables. # This is here so that gen_oplist can detect which module declared these variables.
straight.__module__ = 'tensor' straight.__module__ = 'tensor'
...@@ -1181,7 +1181,7 @@ def _elemwise(scalar_op, name, doc_prefix=''): ...@@ -1181,7 +1181,7 @@ def _elemwise(scalar_op, name, doc_prefix=''):
def _redefine(real_symbol_value, module='tensor'): def _redefine(real_symbol_value, module='tensor'):
"""Replace the value associated with a function symbol. """Replace the value associated with a function symbol.
This is useful to trick epydoc into doing what we want. It's a hack. This is useful to trick epydoc into doing what we want. It's a hack.
""" """
real_symbol_value.__module__ = 'tensor.basic' real_symbol_value.__module__ = 'tensor.basic'
...@@ -1275,7 +1275,7 @@ def _conversion(real_value, name): ...@@ -1275,7 +1275,7 @@ def _conversion(real_value, name):
_convert_to_int8 = _conversion(elemwise.Elemwise(scal.convert_to_int8), 'int8') _convert_to_int8 = _conversion(elemwise.Elemwise(scal.convert_to_int8), 'int8')
"""Cast to 8-bit integer""" """Cast to 8-bit integer"""
_convert_to_int16 = _conversion(elemwise.Elemwise(scal.convert_to_int16), 'int16') _convert_to_int16 = _conversion(elemwise.Elemwise(scal.convert_to_int16), 'int16')
"""Cast to 16-bit integer""" """Cast to 16-bit integer"""
...@@ -1287,7 +1287,7 @@ _convert_to_int64 = _conversion(elemwise.Elemwise(scal.convert_to_int64), 'int64 ...@@ -1287,7 +1287,7 @@ _convert_to_int64 = _conversion(elemwise.Elemwise(scal.convert_to_int64), 'int64
_convert_to_uint8 = _conversion(elemwise.Elemwise(scal.convert_to_uint8), 'uint8') _convert_to_uint8 = _conversion(elemwise.Elemwise(scal.convert_to_uint8), 'uint8')
"""Cast to unsigned 8-bit integer""" """Cast to unsigned 8-bit integer"""
_convert_to_uint16 = _conversion(elemwise.Elemwise(scal.convert_to_uint16), 'uint16') _convert_to_uint16 = _conversion(elemwise.Elemwise(scal.convert_to_uint16), 'uint16')
"""Cast to unsigned 16-bit integer""" """Cast to unsigned 16-bit integer"""
...@@ -1324,9 +1324,9 @@ _cast_mapping = { ...@@ -1324,9 +1324,9 @@ _cast_mapping = {
'complex128': _convert_to_complex128} 'complex128': _convert_to_complex128}
@constructor @constructor
def cast(x, dtype): def cast(x, dtype):
"""Symbolically cast `x` to a Tensor of type `dtype`.""" """Symbolically cast `x` to a Tensor of type `dtype`."""
if dtype=='floatX': dtype = config.floatX if dtype=='floatX': dtype = config.floatX
_x = as_tensor_variable(x) _x = as_tensor_variable(x)
if _x.type.dtype == dtype: if _x.type.dtype == dtype:
return _x return _x
...@@ -1382,7 +1382,7 @@ pprint.assign(_shape, printing.MemberPrinter('shape')) ...@@ -1382,7 +1382,7 @@ pprint.assign(_shape, printing.MemberPrinter('shape'))
class MaxAndArgmax(Op): class MaxAndArgmax(Op):
"""Calculate the max and argmax over a given axis. """Calculate the max and argmax over a given axis.
.. note:: .. note::
If axis is None it means to calculate the max over the last dimension which is If axis is None it means to calculate the max over the last dimension which is
...@@ -1393,7 +1393,7 @@ class MaxAndArgmax(Op): ...@@ -1393,7 +1393,7 @@ class MaxAndArgmax(Op):
nin=2 # tensor, axis nin=2 # tensor, axis
nout=2 # max val, max idx nout=2 # max val, max idx
E_axis = 'invalid axis' E_axis = 'invalid axis'
def __eq__(self,other): def __eq__(self,other):
return type(self)==type(other) return type(self)==type(other)
def __hash__(self): def __hash__(self):
...@@ -1422,7 +1422,7 @@ class MaxAndArgmax(Op): ...@@ -1422,7 +1422,7 @@ class MaxAndArgmax(Op):
inputs = [x, axis] inputs = [x, axis]
#TODO: figure things out if axis is a constant #TODO: figure things out if axis is a constant
broadcastable = [False] * (x.type.ndim - 1) broadcastable = [False] * (x.type.ndim - 1)
outputs = [tensor(x.type.dtype, broadcastable,name='max'), outputs = [tensor(x.type.dtype, broadcastable,name='max'),
tensor('int32', broadcastable,name='argmax')] tensor('int32', broadcastable,name='argmax')]
return Apply(self, inputs, outputs) return Apply(self, inputs, outputs)
def perform(self, node, (x, axis), (max, max_idx)): def perform(self, node, (x, axis), (max, max_idx)):
...@@ -1445,7 +1445,7 @@ class MaxAndArgmax(Op): ...@@ -1445,7 +1445,7 @@ class MaxAndArgmax(Op):
# gMax * dMax/dx + gArgMax * dArgMax/dx, gMax * dMax/daxis + gArgMax * dArgMax/daxis # gMax * dMax/dx + gArgMax * dArgMax/dx, gMax * dMax/daxis + gArgMax * dArgMax/daxis
# g_max has one less dimension than x, so you need to complete g_max to x's shape # g_max has one less dimension than x, so you need to complete g_max to x's shape
# when axis=0 the broadcasting mechanism does it automatically # when axis=0 the broadcasting mechanism does it automatically
if not ( axis.data == 0 or axis.data == x.ndim-1): if not ( axis.data == 0 or axis.data == x.ndim-1):
raise NotImplementedError('MaxAndArgmax gradient with axis corresponding to internal dimension') raise NotImplementedError('MaxAndArgmax gradient with axis corresponding to internal dimension')
if axis.data==0: if axis.data==0:
...@@ -1874,7 +1874,7 @@ if 0: ...@@ -1874,7 +1874,7 @@ if 0:
class Alloc(gof.Op): class Alloc(gof.Op):
"""Create a Tensor from an initial value and a desired shape """Create a Tensor from an initial value and a desired shape
alloc(value, shape0, shape1, ..., shapeN) alloc(value, shape0, shape1, ..., shapeN)
Returns an N-dimensional tensor initialized by `value` using something equivalent to Returns an N-dimensional tensor initialized by `value` using something equivalent to
>>> z = numpy.zeros(shape, value.dtype) >>> z = numpy.zeros(shape, value.dtype)
...@@ -1883,7 +1883,7 @@ class Alloc(gof.Op): ...@@ -1883,7 +1883,7 @@ class Alloc(gof.Op):
The result has N dimensions, has the dtype of `value` and is obtained by broadcasting value The result has N dimensions, has the dtype of `value` and is obtained by broadcasting value
over the output ndarray. over the output ndarray.
This Op is used to replace fill() during optimizations because after shapes are lifted, This Op is used to replace fill() during optimizations because after shapes are lifted,
the first argument to fill can often be pruned from the graph. the first argument to fill can often be pruned from the graph.
""" """
def __init__(self): def __init__(self):
...@@ -1943,7 +1943,7 @@ class Alloc(gof.Op): ...@@ -1943,7 +1943,7 @@ class Alloc(gof.Op):
pass pass
return ret return ret
alloc = Alloc() alloc = Alloc()
pprint.assign(alloc, printing.FunctionPrinter('alloc')) pprint.assign(alloc, printing.FunctionPrinter('alloc'))
...@@ -2006,8 +2006,8 @@ def mean(input, axis = None, op = False): ...@@ -2006,8 +2006,8 @@ def mean(input, axis = None, op = False):
:param axis: compute the mean along this axis of the tensor. :param axis: compute the mean along this axis of the tensor.
None means all axes (like numpy). None means all axes (like numpy).
:type axis: None or int or (list of int) (see `Sum`) :type axis: None or int or (list of int) (see `Sum`)
:note: for gpu, if you manually cast the input to float32 before calling :note: for gpu, if you manually cast the input to float32 before calling
mean, everything will be done on the gpu. mean, everything will be done on the gpu.
""" """
if op: if op:
...@@ -2117,7 +2117,7 @@ class Default(gof.Op): ...@@ -2117,7 +2117,7 @@ class Default(gof.Op):
if x is None: if x is None:
# why copy? Theano can't yet understand out[0] being a view of either x or y, # why copy? Theano can't yet understand out[0] being a view of either x or y,
# so we can be a view of x, but only a copy of y. # so we can be a view of x, but only a copy of y.
out[0] = default.copy() out[0] = default.copy()
else: else:
out[0] = x out[0] = x
default = Default() default = Default()
...@@ -2221,7 +2221,7 @@ class Subtensor(Op): ...@@ -2221,7 +2221,7 @@ class Subtensor(Op):
integers are indexes into the inputs array, and the start/stop/step members integers are indexes into the inputs array, and the start/stop/step members
of each slice are also integer indexes into the inputs array (or None). The of each slice are also integer indexes into the inputs array (or None). The
inputs array is the tensor x, followed by scalar integer variables. inputs array is the tensor x, followed by scalar integer variables.
@todo: add support for advanced tensor indexing (in Subtensor_dx too). @todo: add support for advanced tensor indexing (in Subtensor_dx too).
The idx_list is a tuple similar in structure to the sort of key you might expect in numpy's The idx_list is a tuple similar in structure to the sort of key you might expect in numpy's
...@@ -2246,7 +2246,7 @@ class Subtensor(Op): ...@@ -2246,7 +2246,7 @@ class Subtensor(Op):
elif isinstance(entry, slice): elif isinstance(entry, slice):
helper(entry.start) helper(entry.start)
helper(entry.stop) helper(entry.stop)
helper(entry.step) helper( entry.step)
for idx in idxs: for idx in idxs:
helper(idx) helper(idx)
return ret return ret
...@@ -2312,7 +2312,7 @@ class Subtensor(Op): ...@@ -2312,7 +2312,7 @@ class Subtensor(Op):
def make_node(self, x, *inputs): def make_node(self, x, *inputs):
x = as_tensor_variable(x) x = as_tensor_variable(x)
inputs = tuple(self.my_as_scalar(a) for a in inputs) inputs = tuple(self.my_as_scalar(a) for a in inputs)
idx_list = list(self.idx_list) idx_list = list(self.idx_list)
if len(idx_list) > x.type.ndim: if len(idx_list) > x.type.ndim:
raise ValueError(Subtensor.e_invalid, raise ValueError(Subtensor.e_invalid,
...@@ -2412,7 +2412,7 @@ class Subtensor(Op): ...@@ -2412,7 +2412,7 @@ class Subtensor(Op):
msg += [(entry.start, entry.stop, entry.step)] msg += [(entry.start, entry.stop, entry.step)]
else: else:
msg += [entry] msg += [entry]
idx_list = tuple(msg) idx_list = tuple(msg)
#backport #backport
#idx_list = tuple((entry.start, entry.stop, entry.step) #idx_list = tuple((entry.start, entry.stop, entry.step)
...@@ -2472,7 +2472,7 @@ class SubtensorPrinter: ...@@ -2472,7 +2472,7 @@ class SubtensorPrinter:
msg3 = "" msg3 = ""
else: else:
msg3 = ":%s" % entry.step msg3 = ":%s" % entry.step
sidxs.append("%s:%s%s" % (msg1, msg2, msg3)) sidxs.append("%s:%s%s" % (msg1, msg2, msg3))
#backport #backport
#sidxs.append("%s:%s%s" % ("" if entry.start is None or entry.start == 0 else entry.start, #sidxs.append("%s:%s%s" % ("" if entry.start is None or entry.start == 0 else entry.start,
...@@ -2531,10 +2531,10 @@ def inc_subtensor(x, y, inplace=False, set_instead_of_inc=False): ...@@ -2531,10 +2531,10 @@ def inc_subtensor(x, y, inplace=False, set_instead_of_inc=False):
class IncSubtensor(Op): class IncSubtensor(Op):
"""Increment a subtensor. """Increment a subtensor.
This is like numpy's This is like numpy's
x[i,j,k] += y x[i,j,k] += y
It is used internally to implement the gradient on SubTensor. It is used internally to implement the gradient on SubTensor.
:param set_instead_of_inc: if True set the subtensor to the value instead :param set_instead_of_inc: if True set the subtensor to the value instead
...@@ -2592,7 +2592,7 @@ class IncSubtensor(Op): ...@@ -2592,7 +2592,7 @@ class IncSubtensor(Op):
def make_node(self, x, y, *inputs): def make_node(self, x, y, *inputs):
x, y = map(as_tensor_variable, [x, y]) x, y = map(as_tensor_variable, [x, y])
inputs = tuple(map(Subtensor.my_as_scalar, inputs)) inputs = tuple(map(Subtensor.my_as_scalar, inputs))
idx_list = list(self.idx_list) idx_list = list(self.idx_list)
if len(idx_list) > x.type.ndim: if len(idx_list) > x.type.ndim:
raise ValueError(Subtensor.e_invalid, raise ValueError(Subtensor.e_invalid,
...@@ -2671,11 +2671,11 @@ class Split(Op): ...@@ -2671,11 +2671,11 @@ class Split(Op):
"""Partition a `TensorVariable` along some axis. """Partition a `TensorVariable` along some axis.
.. python:: .. python::
x = vector() x = vector()
splits = lvector() splits = lvector()
# you have to declare right away how many split_points there will be. # you have to declare right away how many split_points there will be.
ra, rb, rc = split(x, splits, n_splits = 3, axis = 0) ra, rb, rc = split(x, splits, n_splits = 3, axis = 0)
f = function([x, splits], [ra, rb, rc]) f = function([x, splits], [ra, rb, rc])
...@@ -2709,16 +2709,16 @@ class Split(Op): ...@@ -2709,16 +2709,16 @@ class Split(Op):
node = self.make_node(*inputs, **kwargs) node = self.make_node(*inputs, **kwargs)
node.tag.trace = traceback.extract_stack()[:-1] node.tag.trace = traceback.extract_stack()[:-1]
return node.outputs return node.outputs
def make_node(self, x, axis, splits): def make_node(self, x, axis, splits):
"""WRITEME""" """WRITEME"""
x = as_tensor_variable(x) x = as_tensor_variable(x)
axis = as_tensor_variable(axis) axis = as_tensor_variable(axis)
splits = as_tensor_variable(splits) splits = as_tensor_variable(splits)
if splits.type not in int_vector_types: if splits.type not in int_vector_types:
raise TypeError('splits must have type tensor.lvector', splits.type) raise TypeError('splits must have type tensor.lvector', splits.type)
if axis.type not in int_types: if axis.type not in int_types:
raise TypeError('axis must have type lscalar', axis.type) raise TypeError('axis must have type lscalar', axis.type)
# # The following lines are necessary if we allow splits of zero # # The following lines are necessary if we allow splits of zero
...@@ -2738,21 +2738,21 @@ class Split(Op): ...@@ -2738,21 +2738,21 @@ class Split(Op):
#in python 2.4, x.shape[numpy.asarray(1)] don't work. #in python 2.4, x.shape[numpy.asarray(1)] don't work.
if sys.version_info[0:2]==(2, 4) and axis.size==1: if sys.version_info[0:2]==(2, 4) and axis.size==1:
axis=int(axis) axis=int(axis)
try: try:
len_along_axis = x.shape[axis] len_along_axis = x.shape[axis]
except : except :
raise ValueError('Split.perform() with axis=(%s) is invalid for x.shape==(%s)' raise ValueError('Split.perform() with axis=(%s) is invalid for x.shape==(%s)'
%(axis, x.shape)) %(axis, x.shape))
if len(splits) != self.len_splits: if len(splits) != self.len_splits:
raise ValueError('In Split.perform(), len(splits) != len_splits.', raise ValueError('In Split.perform(), len(splits) != len_splits.',
(len(splits), self.len_splits)) (len(splits), self.len_splits))
if numpy.sum(splits) != len_along_axis: if numpy.sum(splits) != len_along_axis:
raise ValueError('The splits sum to %s, expected %s' % (numpy.sum(splits), len_along_axis)) raise ValueError('The splits sum to %s, expected %s' % (numpy.sum(splits), len_along_axis))
if not all(splits): if not all(splits):
raise ValueError('Cannot have a split of zero.') raise ValueError('Cannot have a split of zero.')
# Checking is done, let's roll the splitting algorithm! # Checking is done, let's roll the splitting algorithm!
# Basically we step along the given axis of x, extracting subtensors of size splits[i] # Basically we step along the given axis of x, extracting subtensors of size splits[i]
# as we go along. # as we go along.
...@@ -2826,7 +2826,7 @@ def addbroadcast(x, *axes): ...@@ -2826,7 +2826,7 @@ def addbroadcast(x, *axes):
def unbroadcast(x, *axes): def unbroadcast(x, *axes):
""" """
Make the input impossible to broadcast in the specified axes. Make the input impossible to broadcast in the specified axes.
We apply the opt here to don't pollute the graph especially during the gpu optimization We apply the opt here to don't pollute the graph especially during the gpu optimization
""" """
rval = Rebroadcast(*[(axis, False) for axis in axes])(x) rval = Rebroadcast(*[(axis, False) for axis in axes])(x)
...@@ -2835,7 +2835,7 @@ def unbroadcast(x, *axes): ...@@ -2835,7 +2835,7 @@ def unbroadcast(x, *axes):
def patternbroadcast(x, broadcastable): def patternbroadcast(x, broadcastable):
""" """
Make the input impossible to broadcast in the specified axes. Make the input impossible to broadcast in the specified axes.
We apply the opt here to don't pollute the graph especially during the gpu optimization We apply the opt here to don't 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 range(len(broadcastable))])(x)
...@@ -2853,7 +2853,7 @@ class Join(Op): ...@@ -2853,7 +2853,7 @@ class Join(Op):
For joins involving scalar values, see @stack. For joins involving scalar values, see @stack.
.. python:: .. python::
x, y, z = tensor.matrix(), tensor.matrix(), tensor.matrix() x, y, z = tensor.matrix(), tensor.matrix(), tensor.matrix()
u = tensor.vector() u = tensor.vector()
...@@ -2952,7 +2952,7 @@ class Join(Op): ...@@ -2952,7 +2952,7 @@ class Join(Op):
return [None] + split_gz return [None] + split_gz
else: else:
# assume that this isn't differentiable # assume that this isn't differentiable
return [None] * (1 + len(tensors)) return [None] * (1 + len(tensors))
def _native_grad(self, axis_and_tensors, (gz,)): def _native_grad(self, axis_and_tensors, (gz,)):
"""WRITEME""" """WRITEME"""
...@@ -3006,7 +3006,7 @@ pprint.assign(lambda pstate, r: r.owner and isinstance(r.owner.op, Join), ...@@ -3006,7 +3006,7 @@ pprint.assign(lambda pstate, r: r.owner and isinstance(r.owner.op, Join),
@constructor @constructor
def shape_padleft(t, n_ones=1): def shape_padleft(t, n_ones=1):
"""Reshape `t` by left-padding the shape with `n_ones` 1s """Reshape `t` by left-padding the shape with `n_ones` 1s
See also: `shape_padright` and `Dimshuffle` See also: `shape_padright` and `Dimshuffle`
""" """
_t = as_tensor_variable(t) _t = as_tensor_variable(t)
...@@ -3017,7 +3017,7 @@ def shape_padleft(t, n_ones=1): ...@@ -3017,7 +3017,7 @@ def shape_padleft(t, n_ones=1):
@constructor @constructor
def shape_padright(t, n_ones=1): def shape_padright(t, n_ones=1):
"""Reshape `t` by right-padding the shape with `n_ones` 1s """Reshape `t` by right-padding the shape with `n_ones` 1s
See also: `shape_padleft` and `Dimshuffle` See also: `shape_padleft` and `Dimshuffle`
""" """
_t = as_tensor_variable(t) _t = as_tensor_variable(t)
...@@ -3045,10 +3045,10 @@ def stack(*tensors): ...@@ -3045,10 +3045,10 @@ def stack(*tensors):
@constructor @constructor
def concatenate(tensor_list, axis=0): def concatenate(tensor_list, axis=0):
"""Alias for `join`(axis, *tensor_list). """Alias for `join`(axis, *tensor_list).
This function is similar to `join`, but uses the signature of numpy's concatenate function. This function is similar to `join`, but uses the signature of numpy's concatenate function.
This function This function
:Exceptions: :Exceptions:
- `TypeError` : the tensor_list must be a tuple or list - `TypeError` : the tensor_list must be a tuple or list
...@@ -3072,7 +3072,7 @@ def get_vector_length(v): ...@@ -3072,7 +3072,7 @@ def get_vector_length(v):
:Exceptions: :Exceptions:
- `TypeError` : `v` hasn't the proper type. - `TypeError` : `v` hasn't the proper type.
- `ValueError` : No special case applies, the length is not known. - `ValueError` : No special case applies, the length is not known.
In general this is not possible, but for a number of special cases the length can be In general this is not possible, but for a number of special cases the length can be
determined at compile / graph-construction time. This function implements these special determined at compile / graph-construction time. This function implements these special
cases. cases.
...@@ -3165,7 +3165,7 @@ else: ...@@ -3165,7 +3165,7 @@ else:
class Reshape(Op): class Reshape(Op):
"""Perform a reshape operation of the input x to the new shape shp. """Perform a reshape operation of the input x to the new shape shp.
The number of dimensions to which to reshape to (ndim) must be known at graph The number of dimensions to which to reshape to (ndim) must be known at graph
build time.""" build time."""
view_map = {0: [0]} #output 0 is potentially aliased to inputs [0] view_map = {0: [0]} #output 0 is potentially aliased to inputs [0]
def __init__(self, ndim, name = None): def __init__(self, ndim, name = None):
...@@ -3248,7 +3248,7 @@ class Flatten(Op): ...@@ -3248,7 +3248,7 @@ class Flatten(Op):
def grad(self, (x,), (g_out,)): def grad(self, (x,), (g_out,)):
return [reshape(g_out, shape(x), x.ndim)] return [reshape(g_out, shape(x), x.ndim)]
def flatten(x, outdim=1): def flatten(x, outdim=1):
return Flatten(outdim)(x) return Flatten(outdim)(x)
class TileGrad(Op): class TileGrad(Op):
...@@ -3634,7 +3634,7 @@ class AdvancedSubtensor(Op): ...@@ -3634,7 +3634,7 @@ class AdvancedSubtensor(Op):
# TODO: in general, we need to re-pack the inputs into a valid index, just like # TODO: in general, we need to re-pack the inputs into a valid index, just like
# subtensor # subtensor
out[0] = inputs[0].__getitem__(inputs[1:]) out[0] = inputs[0].__getitem__(inputs[1:])
#return #return
#raise NotImplementedError() #raise NotImplementedError()
def grad(self, inputs, (gz,)): def grad(self, inputs, (gz,)):
...@@ -3703,7 +3703,7 @@ class Dot(Op): ...@@ -3703,7 +3703,7 @@ class Dot(Op):
return hash(type(self)) return hash(type(self))
# the rationale for Dot22 is related to getting GEMM Ops into the graph. See Dot22 in tensor.blas for details. # the rationale for Dot22 is related to getting GEMM Ops into the graph. See Dot22 in tensor.blas for details.
def make_node(self, *inputs): def make_node(self, *inputs):
inputs = map(as_tensor_variable, inputs) inputs = map(as_tensor_variable, inputs)
...@@ -3764,7 +3764,7 @@ class Dot(Op): ...@@ -3764,7 +3764,7 @@ class Dot(Op):
elif x.type.ndim == 1 and y.type.ndim > 1: elif x.type.ndim == 1 and y.type.ndim > 1:
rval = dot(gz, y.T), outer(x.T, gz) rval = dot(gz, y.T), outer(x.T, gz)
elif x.type.ndim > 1 and y.type.ndim == 1: elif x.type.ndim > 1 and y.type.ndim == 1:
rval = outer(gz, y.T), dot(x.T, gz) rval = outer(gz, y.T), dot(x.T, gz)
else: else:
rval = dot(gz, y.T), dot(x.T, gz) rval = dot(gz, y.T), dot(x.T, gz)
return cast(rval[0], x.dtype), cast(rval[1], y.dtype) return cast(rval[0], x.dtype), cast(rval[1], y.dtype)
...@@ -3865,7 +3865,7 @@ class TensorDot(Op): ...@@ -3865,7 +3865,7 @@ class TensorDot(Op):
if len(axes[0])!=len(axes[1]): if len(axes[0])!=len(axes[1]):
raise ValueError("We need that the axes 2 sub list of axes are of the same size") raise ValueError("We need that the axes 2 sub list of axes are of the same size")
assert len(axes[0])==len(axes[1]) assert len(axes[0])==len(axes[1])
self.axes = axes self.axes = axes
def __eq__(self, other): def __eq__(self, other):
...@@ -3887,7 +3887,7 @@ class TensorDot(Op): ...@@ -3887,7 +3887,7 @@ class TensorDot(Op):
if axesdim > x.type.ndim or axesdim > y.type.ndim: if axesdim > x.type.ndim or axesdim > y.type.ndim:
raise TypeError('Cannot sum over more dimensions than input. %i > %i,%i' % raise TypeError('Cannot sum over more dimensions than input. %i > %i,%i' %
axesdim, x.type.ndim, y.type.ndim) axesdim, x.type.ndim, y.type.ndim)
outdim = x.type.ndim + y.type.ndim - 2*axesdim outdim = x.type.ndim + y.type.ndim - 2*axesdim
output = tensor(dtype=scal.upcast(x.dtype, y.dtype), output = tensor(dtype=scal.upcast(x.dtype, y.dtype),
broadcastable=[False]*outdim); broadcastable=[False]*outdim);
...@@ -3904,7 +3904,7 @@ class TensorDot(Op): ...@@ -3904,7 +3904,7 @@ class TensorDot(Op):
def grad(self, (x, y), (gz,)): def grad(self, (x, y), (gz,)):
gx, gy = tensordot_grad(self.axes)(x, y, gz) gx, gy = tensordot_grad(self.axes)(x, y, gz)
return [gx, gy] return [gx, gy]
def __str__(self): def __str__(self):
return "tensordot" return "tensordot"
tensordot = TensorDot tensordot = TensorDot
...@@ -3923,7 +3923,7 @@ class Outer(Op): ...@@ -3923,7 +3923,7 @@ class Outer(Op):
if nx != 1: raise TypeError('non-vector arg0 to outer()', x) if nx != 1: raise TypeError('non-vector arg0 to outer()', x)
if ny != 1: raise TypeError('not-vector arg1 to outer()', y) if ny != 1: raise TypeError('not-vector arg1 to outer()', y)
bz = [x.type.broadcastable[0], y.type.broadcastable[0]] bz = [x.type.broadcastable[0], y.type.broadcastable[0]]
i_dtypes = [input.type.dtype for input in inputs] i_dtypes = [input.type.dtype for input in inputs]
...@@ -3997,8 +3997,8 @@ class numeric_grad: ...@@ -3997,8 +3997,8 @@ class numeric_grad:
# #
# There is a relationship between the step size and the function value and the measurement # There is a relationship between the step size and the function value and the measurement
# error that is incurred due to rounding. The finite difference we measure is # error that is incurred due to rounding. The finite difference we measure is
# delta = f(x0) - f(x0+eps) # delta = f(x0) - f(x0+eps)
# #
# For maximum precision, f should be close to zero. # For maximum precision, f should be close to zero.
# For every power of 2 that f departs from zero, we lose a bit of precision in delta. # For every power of 2 that f departs from zero, we lose a bit of precision in delta.
# #
...@@ -4009,7 +4009,7 @@ class numeric_grad: ...@@ -4009,7 +4009,7 @@ class numeric_grad:
# bias into our measurement in general for non-linear functions. # bias into our measurement in general for non-linear functions.
# #
# It would be interesting to have a version of numeric grad that used an adaptive stepsize. # It would be interesting to have a version of numeric grad that used an adaptive stepsize.
# #
# For now, we use a heuristic that catches very bad gradients, but is not perfectly # For now, we use a heuristic that catches very bad gradients, but is not perfectly
# accurate. # accurate.
type_eps = {'float64': 1e-7, type_eps = {'float64': 1e-7,
...@@ -4161,7 +4161,7 @@ def verify_grad(fun, pt, n_tests=2, rng=None, eps=None, abs_tol=None, rel_tol=No ...@@ -4161,7 +4161,7 @@ def verify_grad(fun, pt, n_tests=2, rng=None, eps=None, abs_tol=None, rel_tol=No
mode=None, cast_to_output_type=False): mode=None, cast_to_output_type=False):
""" Test a gradient by Finite Difference Method. Raise error on failure. """ Test a gradient by Finite Difference Method. Raise error on failure.
Example: Example:
>>> verify_grad(theano.tensor.tanh, >>> verify_grad(theano.tensor.tanh,
(numpy.asarray([[2,3,4], [-1, 3.3, 9.9]]),), (numpy.asarray([[2,3,4], [-1, 3.3, 9.9]]),),
rng=numpy.random) rng=numpy.random)
...@@ -4187,8 +4187,8 @@ def verify_grad(fun, pt, n_tests=2, rng=None, eps=None, abs_tol=None, rel_tol=No ...@@ -4187,8 +4187,8 @@ def verify_grad(fun, pt, n_tests=2, rng=None, eps=None, abs_tol=None, rel_tol=No
debug mode, which can be very slow if it has to verify a lot debug mode, which can be very slow if it has to verify a lot
of intermediate computations. of intermediate computations.
:note: This op does not support multiple outputs. In tests/test_scan.py there is :note: This op does not support multiple outputs. In tests/test_scan.py there is
an experimental verify_grad that covers that case as well by using random an experimental verify_grad that covers that case as well by using random
projections. projections.
""" """
assert isinstance(pt, (list,tuple)) assert isinstance(pt, (list,tuple))
...@@ -4244,7 +4244,7 @@ def verify_grad(fun, pt, n_tests=2, rng=None, eps=None, abs_tol=None, rel_tol=No ...@@ -4244,7 +4244,7 @@ def verify_grad(fun, pt, n_tests=2, rng=None, eps=None, abs_tol=None, rel_tol=No
t_r = shared(random_projection()) t_r = shared(random_projection())
#random projection of o onto t_r #random projection of o onto t_r
cost = sum(t_r * o_output) #This sum() is defined above, it's not the builtin sum. cost = sum(t_r * o_output) #This sum() is defined above, it's not the builtin sum.
cost_fn = function(tensor_pt, cost) cost_fn = function(tensor_pt, cost)
#todo-- determine if this is actually needed #todo-- determine if this is actually needed
......
theano/tensor/opt.py
浏览文件 @ cfc588cc
...@@ -119,11 +119,11 @@ def insert_inplace_optimizer_op(OP): ...@@ -119,11 +119,11 @@ def insert_inplace_optimizer_op(OP):
""" """
#we should not validate too often as this take too much time to execute! #we should not validate too often as this take too much time to execute!
#It is the _dfs_toposort() fct in theano/gof/destroyhandler.py #It is the _dfs_toposort() fct in theano/gof/destroyhandler.py
#that take so much time. #that take so much time.
#Should we try to use another lib that do toposort? #Should we try to use another lib that do toposort?
# igraph: http://igraph.sourceforge.net/ # igraph: http://igraph.sourceforge.net/
# networkx: https://networkx.lanl.gov/ # networkx: https://networkx.lanl.gov/
#Should we try to use cython? #Should we try to use cython?
# compiling only that fct is not enought, should we try to add the deque class too? # compiling only that fct is not enought, should we try to add the deque class too?
# and init the deque and other list to an upper bound number of element? # and init the deque and other list to an upper bound number of element?
#Should Theano do online toposort as in http://code.google.com/p/acyclic/? #Should Theano do online toposort as in http://code.google.com/p/acyclic/?
...@@ -213,7 +213,7 @@ def insert_inplace_optimizer_op(OP): ...@@ -213,7 +213,7 @@ def insert_inplace_optimizer_op(OP):
insert_inplace_optimizer = insert_inplace_optimizer_op(T.Elemwise) insert_inplace_optimizer = insert_inplace_optimizer_op(T.Elemwise)
compile.optdb.register('inplace_opt', insert_inplace_optimizer, 75, 'fast_run', 'inplace') compile.optdb.register('inplace_opt', insert_inplace_optimizer, 75, 'fast_run', 'inplace')
def register_canonicalize(lopt, *tags, **kwargs): def register_canonicalize(lopt, *tags, **kwargs):
name = (kwargs and kwargs.pop('name')) or lopt.__name__ name = (kwargs and kwargs.pop('name')) or lopt.__name__
...@@ -304,7 +304,7 @@ class MakeVector(T.Op): ...@@ -304,7 +304,7 @@ class MakeVector(T.Op):
"""Concatenate a number of scalars together into a vector """Concatenate a number of scalars together into a vector
This is a simple version of stack() that introduces far less cruft into the graph. This is a simple version of stack() that introduces far less cruft into the graph.
Should work with 0 inputs. The constant_folding optimization will remove it. Should work with 0 inputs. The constant_folding optimization will remove it.
""" """
def __init__(self, dtype='int64'): def __init__(self, dtype='int64'):
...@@ -398,7 +398,7 @@ class Shape_i(T.Op): ...@@ -398,7 +398,7 @@ class Shape_i(T.Op):
%(out)s=(PyArrayObject*)PyArray_ZEROS(0, NULL, PyArray_INT64, 0); %(out)s=(PyArrayObject*)PyArray_ZEROS(0, NULL, PyArray_INT64, 0);
((npy_int64*)PyArray_DATA(%(out)s))[0]=%(x)s->dimensions[%(i)s]; ((npy_int64*)PyArray_DATA(%(out)s))[0]=%(x)s->dimensions[%(i)s];
"""%locals() """%locals()
elif node.inputs[0].type.__class__.__name__=="CudaNdarrayType": elif node.inputs[0].type.__class__.__name__=="CudaNdarrayType":
#Don't want to import cuda stuff here. #Don't want to import cuda stuff here.
return """ return """
...@@ -413,12 +413,12 @@ class Shape_i(T.Op): ...@@ -413,12 +413,12 @@ class Shape_i(T.Op):
class ShapeFeature(object): class ShapeFeature(object):
"""Graph optimizer for removing all calls to shape() """Graph optimizer for removing all calls to shape()
This optimizer replaces all Shapes and Subtensors of Shapes with Shape_i and MakeVector This optimizer replaces all Shapes and Subtensors of Shapes with Shape_i and MakeVector
Ops. Ops.
This optimizer has several goals: This optimizer has several goals:
1. to 'lift' Shapes to as close to the inputs as possible. 1. to 'lift' Shapes to as close to the inputs as possible.
2. to infer the shape of every node in the graph in terms of the input shapes. 2. to infer the shape of every node in the graph in terms of the input shapes.
3. remove all fills (T.second, T.fill) from the graph 3. remove all fills (T.second, T.fill) from the graph
...@@ -430,7 +430,7 @@ class ShapeFeature(object): ...@@ -430,7 +430,7 @@ class ShapeFeature(object):
Many optimizations refuse to work on nodes with multiple clients. Many optimizations refuse to work on nodes with multiple clients.
Lifting is done by using an `<Op>.infer_shape` function if one is present, or else using a Lifting is done by using an `<Op>.infer_shape` function if one is present, or else using a
conservative default. An Op that supports shape-lifting should define a conservative default. An Op that supports shape-lifting should define a
infer_shape(self, node, input_shapes) function. The argument input_shapes is a tuple infer_shape(self, node, input_shapes) function. The argument input_shapes is a tuple
of tuples... there is an interior tuple for each input to the node. The tuple has as many of tuples... there is an interior tuple for each input to the node. The tuple has as many
elements as dimensions. The element in position i of tuple j represents the i'th shape elements as dimensions. The element in position i of tuple j represents the i'th shape
...@@ -439,9 +439,9 @@ class ShapeFeature(object): ...@@ -439,9 +439,9 @@ class ShapeFeature(object):
the output[j].shape[i] of the function. If an output is not a TensorType, then None should the output[j].shape[i] of the function. If an output is not a TensorType, then None should
be returned instead of a tuple for that output. be returned instead of a tuple for that output.
For example the infer_shape for a matrix-matrix product would accept For example the infer_shape for a matrix-matrix product would accept
input_shapes=((x0,x1), (y0,y1)) and return ((x0, y1),). input_shapes=((x0,x1), (y0,y1)) and return ((x0, y1),).
Inferring the shape of internal nodes in the graph is important for doing size-driven Inferring the shape of internal nodes in the graph is important for doing size-driven
optimizations. If we know how big various intermediate results will be, we can estimate optimizations. If we know how big various intermediate results will be, we can estimate
...@@ -495,7 +495,7 @@ class ShapeFeature(object): ...@@ -495,7 +495,7 @@ class ShapeFeature(object):
return T.constant(s_i, dtype='int64') return T.constant(s_i, dtype='int64')
if type(s_i) in (tuple,list): if type(s_i) in (tuple,list):
# this dimension is the same as many of the inputs # this dimension is the same as many of the inputs
# which tells us that if one of the inputs is known, # which tells us that if one of the inputs is known,
# the others all become known. # the others all become known.
# TODO: should be implemented in Elemwise, and Dot # TODO: should be implemented in Elemwise, and Dot
# #
...@@ -506,7 +506,7 @@ class ShapeFeature(object): ...@@ -506,7 +506,7 @@ class ShapeFeature(object):
raise TypeError('Shape element must be scalar', s_i) raise TypeError('Shape element must be scalar', s_i)
return s_i return s_i
else: else:
raise TypeError('Unsupported shape element', raise TypeError('Unsupported shape element',
s_i, type(s_i), getattr(s_i, 'type', None)) s_i, type(s_i), getattr(s_i, 'type', None))
def set_shape(self, r, s): def set_shape(self, r, s):
...@@ -534,7 +534,7 @@ class ShapeFeature(object): ...@@ -534,7 +534,7 @@ class ShapeFeature(object):
assert not hasattr(env, 'shape_feature') assert not hasattr(env, 'shape_feature')
env.shape_feature = self env.shape_feature = self
self.shape_of = {} # Variable -> tuple(scalars) or None (All tensor vars map to tuple) self.shape_of = {} # Variable -> tuple(scalars) or None (All tensor vars map to tuple)
self.scheduled = {} # Variable -> self.scheduled = {} # Variable ->
self.lscalar_one = T.constant(1, dtype='int64') self.lscalar_one = T.constant(1, dtype='int64')
assert self.lscalar_one.type == T.lscalar assert self.lscalar_one.type == T.lscalar
for node in env.toposort(): for node in env.toposort():
...@@ -622,7 +622,7 @@ def local_fill_to_alloc(node): ...@@ -622,7 +622,7 @@ def local_fill_to_alloc(node):
This is an important optimization because with the shape_to_shape_i optimization, the This is an important optimization because with the shape_to_shape_i optimization, the
dependency on 's' is often removed. dependency on 's' is often removed.
""" """
if node.op == T.fill: if node.op == T.fill:
r, v = node.inputs r, v = node.inputs
...@@ -637,7 +637,7 @@ def local_fill_to_alloc(node): ...@@ -637,7 +637,7 @@ def local_fill_to_alloc(node):
shape_of = node.env.shape_feature.shape_of shape_of = node.env.shape_feature.shape_of
# TODO: cut out un-necessary dimshuffles of v # TODO: cut out un-necessary dimshuffles of v
rval = [T.alloc(T.cast(v, node.outputs[0].dtype), *shape_of[node.outputs[0]])] rval = [T.alloc(T.cast(v, node.outputs[0].dtype), *shape_of[node.outputs[0]])]
#if rval[0].type != node.outputs[0].type: #if rval[0].type != node.outputs[0].type:
#print >> sys.stderr, theano.printing.debugprint(node.outputs[0], file='str') #print >> sys.stderr, theano.printing.debugprint(node.outputs[0], file='str')
...@@ -700,7 +700,7 @@ def local_subtensor_make_vector(node): ...@@ -700,7 +700,7 @@ def local_subtensor_make_vector(node):
raise raise
if isinstance(idx, (scalar.Scalar, T.TensorType)): if isinstance(idx, (scalar.Scalar, T.TensorType)):
# The idx is a Scalar, ie a Type. This means the actual index # The idx is a Scalar, ie a Type. This means the actual index
# is contained in node.inputs[1] # is contained in node.inputs[1]
old_idx, idx = idx, node.inputs[1] old_idx, idx = idx, node.inputs[1]
assert idx.type == old_idx assert idx.type == old_idx
...@@ -773,7 +773,7 @@ class Assert(T.Op): ...@@ -773,7 +773,7 @@ class Assert(T.Op):
cond = [T.as_tensor_variable(c) for c in conds] cond = [T.as_tensor_variable(c) for c in conds]
assert numpy.all([c.type.ndim == 0 for c in cond]) assert numpy.all([c.type.ndim == 0 for c in cond])
return gof.Apply(self, [value]+cond, [value.type()]) return gof.Apply(self, [value]+cond, [value.type()])
def __str__(self): def __str__(self):
return self.__class__.__name__ return self.__class__.__name__
def perform(self, node, inputs, (out,)): def perform(self, node, inputs, (out,)):
...@@ -807,7 +807,7 @@ class Assert(T.Op): ...@@ -807,7 +807,7 @@ class Assert(T.Op):
def infer_shape(self, node, input_shapes): def infer_shape(self, node, input_shapes):
return [input_shapes[0]] return [input_shapes[0]]
assert_ = Assert() assert_ = Assert()
@register_specialize @register_specialize
...@@ -818,13 +818,13 @@ def local_remove_useless_assert(node): ...@@ -818,13 +818,13 @@ def local_remove_useless_assert(node):
for c in node.inputs[1:]: for c in node.inputs[1:]:
try: try:
const = get_constant_value(c) const = get_constant_value(c)
if 0!=const.ndim or const==0: if 0!=const.ndim or const==0:
#Should we raise an error here? How to be sure it is not catched? #Should we raise an error here? How to be sure it is not catched?
cond.append(c) cond.append(c)
except TypeError: except TypeError:
cond.append(c) cond.append(c)
if len(cond)==0: if len(cond)==0:
return [node.inputs[0]] return [node.inputs[0]]
if len(cond)!=len(node.inputs)-1: if len(cond)!=len(node.inputs)-1:
...@@ -873,12 +873,12 @@ def local_alloc_elemwise(node): ...@@ -873,12 +873,12 @@ def local_alloc_elemwise(node):
isinstance(i.owner.inputs[0].owner.op,T.Alloc)): isinstance(i.owner.inputs[0].owner.op,T.Alloc)):
no_broad_idx = idx no_broad_idx = idx
break break
assert no_broad_idx>=0 assert no_broad_idx>=0
assert_op = node.inputs[no_broad_idx] assert_op = node.inputs[no_broad_idx]
cmp_op = assert_op cmp_op = assert_op
new = [] new = []
for i in node.inputs: for i in node.inputs:
if i.owner and isinstance(i.owner.op,T.Alloc) and i.owner.inputs[0].type != i.owner.outputs[0].type: if i.owner and isinstance(i.owner.op,T.Alloc) and i.owner.inputs[0].type != i.owner.outputs[0].type:
#when i.owner.inputs[0].type == i.owner.outputs[0].type we will remove that alloc later #when i.owner.inputs[0].type == i.owner.outputs[0].type we will remove that alloc later
...@@ -1017,8 +1017,8 @@ def local_IncSubtensor_serialize(node): ...@@ -1017,8 +1017,8 @@ def local_IncSubtensor_serialize(node):
IncSubtensor(Elemwise{second}(a, 0), g(f(a[2])), [2]) IncSubtensor(Elemwise{second}(a, 0), g(f(a[2])), [2])
This is much worse because this time we have to produce 3 matrices the size of 'a', just so This is much worse because this time we have to produce 3 matrices the size of 'a', just so
we can add them together. we can add them together.
This Op rearranges IncSubtensor's that all work on the same initial argument (here, This Op rearranges IncSubtensor's that all work on the same initial argument (here,
Elemwise{second}(a,0)) into a chain. The advantage of the chain structure is that each one Elemwise{second}(a,0)) into a chain. The advantage of the chain structure is that each one
can be optimized later in the pipeline to operate inplace. can be optimized later in the pipeline to operate inplace.
...@@ -1028,7 +1028,7 @@ def local_IncSubtensor_serialize(node): ...@@ -1028,7 +1028,7 @@ def local_IncSubtensor_serialize(node):
# #
# add(x, incsubtensor(b, c), incsubtensor(b, d)) # add(x, incsubtensor(b, c), incsubtensor(b, d))
# -> incsubtensor(incsubtensor(add(x,b,b), c), d) # -> incsubtensor(incsubtensor(add(x,b,b), c), d)
""" """
def movable(i): def movable(i):
# Return True iff this is a incsubtensor that we can move # Return True iff this is a incsubtensor that we can move
...@@ -1138,7 +1138,7 @@ def local_rebroadcast_lift(node): ...@@ -1138,7 +1138,7 @@ def local_rebroadcast_lift(node):
def apply_rebroadcast_opt(rval): def apply_rebroadcast_opt(rval):
""" """
Apply as many times as required the optimization local_useless_rebroadcast Apply as many times as required the optimization local_useless_rebroadcast
and local_rebroadcast_lift. and local_rebroadcast_lift.
:param rval: a Variable :param rval: a Variable
...@@ -1149,7 +1149,7 @@ def apply_rebroadcast_opt(rval): ...@@ -1149,7 +1149,7 @@ def apply_rebroadcast_opt(rval):
while changed and rval.owner: while changed and rval.owner:
changed = False changed = False
rval2 = theano.tensor.opt.local_useless_rebroadcast.transform(rval.owner) rval2 = theano.tensor.opt.local_useless_rebroadcast.transform(rval.owner)
if rval2: if rval2:
assert len(rval2)==1 assert len(rval2)==1
rval = rval2[0] rval = rval2[0]
changed = True changed = True
...@@ -1216,7 +1216,7 @@ def local_mul_switch_sink(node): ...@@ -1216,7 +1216,7 @@ def local_mul_switch_sink(node):
fct[0].values_eq_approx = fct[0].type.values_eq_approx_remove_nan fct[0].values_eq_approx = fct[0].type.values_eq_approx_remove_nan
return fct return fct
except TypeError: except TypeError:
pass pass
try: try:
if get_constant_value(switch.inputs[2]) == 0.: if get_constant_value(switch.inputs[2]) == 0.:
listmul = node.inputs[:idx] + node.inputs[idx+1:] listmul = node.inputs[:idx] + node.inputs[idx+1:]
...@@ -1274,7 +1274,7 @@ def local_reshape_chain(node): ...@@ -1274,7 +1274,7 @@ def local_reshape_chain(node):
""" """
if not opt.check_chain(node, T.Reshape, T.Reshape): if not opt.check_chain(node, T.Reshape, T.Reshape):
return False return False
# TODO: this can permit a failing program to run by eliminating the the lower # TODO: this can permit a failing program to run by eliminating the the lower
# reshape # reshape
return [node.op(node.inputs[0].owner.inputs[0], node.inputs[1])] return [node.op(node.inputs[0].owner.inputs[0], node.inputs[1])]
...@@ -1304,7 +1304,7 @@ if 0: ...@@ -1304,7 +1304,7 @@ if 0:
y_shape = node.env.shape_feature.shape_of[y] y_shape = node.env.shape_feature.shape_of[y]
def tmp(thing): def tmp(thing):
try: try:
return T.get_constant_value(thing) return T.get_constant_value(thing)
except (TypeError, ValueError), e: except (TypeError, ValueError), e:
print e, thing.owner.inputs[0] print e, thing.owner.inputs[0]
...@@ -1322,15 +1322,15 @@ def local_fill_cut(node): ...@@ -1322,15 +1322,15 @@ def local_fill_cut(node):
If c.type == a.type. If c.type == a.type.
""" """
# this optimization is essentially for getting broadcasting to replace fill. # this optimization is essentially for getting broadcasting to replace fill.
# This is always possible when using a Compound Elemwise operation, # This is always possible when using a Compound Elemwise operation,
# but it is not always possible without one (consider filling a large matrix with a scalar, # but it is not always possible without one (consider filling a large matrix with a scalar,
# and then adding another scalar. The only numbers that count are the two scalars, but we # and then adding another scalar. The only numbers that count are the two scalars, but we
# can't ignore the large matrix because it gives the shape of the result. # can't ignore the large matrix because it gives the shape of the result.
if not opt.check_chain(node, T.Elemwise): if not opt.check_chain(node, T.Elemwise):
return False return False
output = node.outputs[0] output = node.outputs[0]
try: try:
#reference is some input with the same type as the input but that is not produced by a fill #reference is some input with the same type as the input but that is not produced by a fill
...@@ -1397,7 +1397,7 @@ class Canonizer(gof.LocalOptimizer): ...@@ -1397,7 +1397,7 @@ class Canonizer(gof.LocalOptimizer):
Simplification tool. Simplification tool.
Usage: Canonizer(main, inverse, reciprocal, calculate) Usage: Canonizer(main, inverse, reciprocal, calculate)
* main: a suitable Op class that is commutative, associative and * main: a suitable Op class that is commutative, associative and
takes one to an arbitrary number of inputs, e.g. add or takes one to an arbitrary number of inputs, e.g. add or
mul mul
...@@ -1421,7 +1421,7 @@ class Canonizer(gof.LocalOptimizer): ...@@ -1421,7 +1421,7 @@ class Canonizer(gof.LocalOptimizer):
T = theano.tensor T = theano.tensor
add_canonizer = Canonizer(T.add, T.sub, T.neg, lambda n, d: sum(n) - sum(d)) add_canonizer = Canonizer(T.add, T.sub, T.neg, lambda n, d: sum(n) - sum(d))
mul_canonizer = Canonizer(T.mul, T.true_div, T.inv, lambda n, d: prod(n) / prod(d)) mul_canonizer = Canonizer(T.mul, T.true_div, T.inv, lambda n, d: prod(n) / prod(d))
Examples of optimizations mul_canonizer can perform: Examples of optimizations mul_canonizer can perform:
x / x -> 1 x / x -> 1
(x * y) / x -> y (x * y) / x -> y
...@@ -1659,7 +1659,7 @@ class Canonizer(gof.LocalOptimizer): ...@@ -1659,7 +1659,7 @@ class Canonizer(gof.LocalOptimizer):
# Lists representing the *constant* elements of num and denum # Lists representing the *constant* elements of num and denum
numct, denumct = [], [] numct, denumct = [], []
for v in orig_num: for v in orig_num:
ct = self.get_constant(v) ct = self.get_constant(v)
if ct is not None: if ct is not None:
...@@ -1788,7 +1788,7 @@ register_canonicalize(local_mul_canonizer, name = 'local_mul_canonizer') ...@@ -1788,7 +1788,7 @@ register_canonicalize(local_mul_canonizer, name = 'local_mul_canonizer')
@gof.local_optimizer([T.neg]) @gof.local_optimizer([T.neg])
def local_neg_to_mul(node): def local_neg_to_mul(node):
if node.op == T.neg: if node.op == T.neg:
return [T.mul(numpy.array(-1, dtype = node.inputs[0].dtype), return [T.mul(numpy.array(-1, dtype = node.inputs[0].dtype),
node.inputs[0])] node.inputs[0])]
register_canonicalize(local_neg_to_mul) register_canonicalize(local_neg_to_mul)
...@@ -1797,7 +1797,7 @@ register_canonicalize(local_neg_to_mul) ...@@ -1797,7 +1797,7 @@ register_canonicalize(local_neg_to_mul)
def local_sum_mul_by_scalar(node): def local_sum_mul_by_scalar(node):
"""sum(scalar * smth) -> scalar * sum(smth) """sum(scalar * smth) -> scalar * sum(smth)
""" """
# TODO: if the the thing inside the Sum is a division, # TODO: if the the thing inside the Sum is a division,
# we should get at the numerator.... # we should get at the numerator....
if isinstance(node.op, T.Sum): if isinstance(node.op, T.Sum):
thing_summed, = node.inputs thing_summed, = node.inputs
...@@ -1935,7 +1935,7 @@ def local_sum_sum(node): ...@@ -1935,7 +1935,7 @@ def local_sum_sum(node):
# special case of local_cut_useless_reduce # special case of local_cut_useless_reduce
return [T.Sum(None)(summed.owner.inputs[0])] return [T.Sum(None)(summed.owner.inputs[0])]
if node.op.axis is None: if node.op.axis is None:
# we're summing up everything anyway so lets # we're summing up everything anyway so lets
# do it all at once # do it all at once
return [T.Sum(None)(summed.owner.inputs[0])] return [T.Sum(None)(summed.owner.inputs[0])]
...@@ -1983,7 +1983,6 @@ def local_sum_alloc(node): ...@@ -1983,7 +1983,6 @@ def local_sum_alloc(node):
if summed.owner and isinstance(summed.owner.op, T.Alloc): if summed.owner and isinstance(summed.owner.op, T.Alloc):
input = summed.owner.inputs[0] input = summed.owner.inputs[0]
shapes = summed.owner.inputs[1:] shapes = summed.owner.inputs[1:]
#import pdb;pdb.set_trace()
if node.op.axis is None or node.op.axis == tuple(range(input.ndim)): if node.op.axis is None or node.op.axis == tuple(range(input.ndim)):
try: try:
val = get_constant_value(input) val = get_constant_value(input)
...@@ -2019,7 +2018,7 @@ register_specialize(local_mul_to_neg) ...@@ -2019,7 +2018,7 @@ register_specialize(local_mul_to_neg)
@register_specialize @register_specialize
@gof.local_optimizer([T.neg]) @gof.local_optimizer([T.neg])
def local_neg_neg(node): def local_neg_neg(node):
# other specializations shouldn't put this in, # other specializations shouldn't put this in,
# but sometimes they do # but sometimes they do
if node.op == T.neg: if node.op == T.neg:
if node.inputs[0].owner and node.inputs[0].owner.op == T.neg: if node.inputs[0].owner and node.inputs[0].owner.op == T.neg:
...@@ -2177,11 +2176,11 @@ def local_pow_specialize_device(node): ...@@ -2177,11 +2176,11 @@ def local_pow_specialize_device(node):
rval1 = None rval1 = None
rval1_scal = None rval1_scal = None
while y_to_do>0: while y_to_do>0:
log_to_do = int(numpy.log2(y_to_do)) log_to_do = int(numpy.log2(y_to_do))
if rval1: if rval1:
rval1 *= pow2[log_to_do] rval1 *= pow2[log_to_do]
rval1_scal *= pow2_scal[log_to_do] rval1_scal *= pow2_scal[log_to_do]
else: else:
rval1 = pow2[log_to_do] rval1 = pow2[log_to_do]
rval1_scal = pow2_scal[log_to_do] rval1_scal = pow2_scal[log_to_do]
y_to_do -= 2**log_to_do y_to_do -= 2**log_to_do
...@@ -2197,7 +2196,7 @@ def local_pow_specialize_device(node): ...@@ -2197,7 +2196,7 @@ def local_pow_specialize_device(node):
rval[0] = T.cast(rval[0], odtype) rval[0] = T.cast(rval[0], odtype)
assert rval[0].type == node.outputs[0].type, (rval, node.outputs) assert rval[0].type == node.outputs[0].type, (rval, node.outputs)
return rval return rval
@gof.local_optimizer([T.mul]) @gof.local_optimizer([T.mul])
def local_mul_specialize(node): def local_mul_specialize(node):
"""Remove special-case constants from mul arguments """Remove special-case constants from mul arguments
...@@ -2210,7 +2209,7 @@ def local_mul_specialize(node): ...@@ -2210,7 +2209,7 @@ def local_mul_specialize(node):
neg = False neg = False
new_inputs = [] new_inputs = []
for input in node.inputs: for input in node.inputs:
# remove any neg arguments # remove any neg arguments
while input.owner and input.owner.op == T.neg: while input.owner and input.owner.op == T.neg:
neg ^= True neg ^= True
input = input.owner.inputs[0] input = input.owner.inputs[0]
...@@ -2303,8 +2302,8 @@ def check_for_x_over_absX(numerators, denominators): ...@@ -2303,8 +2302,8 @@ def check_for_x_over_absX(numerators, denominators):
if den.owner and den.owner.op == T.abs_ and den.owner.inputs[0] in numerators: if den.owner and den.owner.op == T.abs_ and den.owner.inputs[0] in numerators:
if den.owner.inputs[0].type.dtype.startswith('complex'): if den.owner.inputs[0].type.dtype.startswith('complex'):
#TODO: Make an Op that projects a complex number to have unit length #TODO: Make an Op that projects a complex number to have unit length
# but projects 0 to 0. That would be a weird Op, but consistent with the # but projects 0 to 0. That would be a weird Op, but consistent with the
# special case below. I heard there's some convention in Matlab that is # special case below. I heard there's some convention in Matlab that is
# similar to this... but not sure. # similar to this... but not sure.
pass pass
else: else:
...@@ -2319,7 +2318,7 @@ local_mul_canonizer.add_simplifier(check_for_x_over_absX, 'X_over_absX') ...@@ -2319,7 +2318,7 @@ local_mul_canonizer.add_simplifier(check_for_x_over_absX, 'X_over_absX')
def local_abs_lift(node): def local_abs_lift(node):
""" """
move the abs toward the input. This is needed for check_for_x_over_absX to apply in more case. move the abs toward the input. This is needed for check_for_x_over_absX to apply in more case.
""" """
if node.op == T.abs_ and node.inputs[0].owner: if node.op == T.abs_ and node.inputs[0].owner:
assert node.nin == 1 assert node.nin == 1
...@@ -2328,13 +2327,13 @@ def local_abs_lift(node): ...@@ -2328,13 +2327,13 @@ def local_abs_lift(node):
if node.inputs[0].owner.op == T.true_div: if node.inputs[0].owner.op == T.true_div:
i = node.inputs[0].owner.inputs i = node.inputs[0].owner.inputs
return [T.true_div(T.abs_(i[0]),T.abs_(i[1]))] return [T.true_div(T.abs_(i[0]),T.abs_(i[1]))]
@register_specialize @register_specialize
@gof.local_optimizer([]) @gof.local_optimizer([])
def local_abs_merge(node): def local_abs_merge(node):
""" """
merge abs generated by local_abs_lift when the canonizer don't need it anymore merge abs generated by local_abs_lift when the canonizer don't need it anymore
""" """
if node.op == T.mul and sum([i.owner.op == T.abs_ for i in node.inputs if i.owner])>1: if node.op == T.mul and sum([i.owner.op == T.abs_ for i in node.inputs if i.owner])>1:
inputs = [] inputs = []
...@@ -2570,7 +2569,7 @@ def constant_folding(node): ...@@ -2570,7 +2569,7 @@ def constant_folding(node):
return msg return msg
register_canonicalize(constant_folding, 'fast_compile') register_canonicalize(constant_folding, 'fast_compile')
register_stabilize(constant_folding) # because register_stabilize(constant_folding) # because
register_specialize(constant_folding) register_specialize(constant_folding)
...@@ -2598,7 +2597,7 @@ def _is_minus1(expr): ...@@ -2598,7 +2597,7 @@ def _is_minus1(expr):
return False return False
#1+erf(x)=>erfc(-x) #1+erf(x)=>erfc(-x)
local_one_plus_erf = gof.PatternSub((T.add, local_one_plus_erf = gof.PatternSub((T.add,
dict(pattern='y', constraint = _is_1), dict(pattern='y', constraint = _is_1),
(T.erf, 'x')), (T.erf, 'x')),
(T.erfc, (T.neg, 'x')), (T.erfc, (T.neg, 'x')),
...@@ -2608,7 +2607,7 @@ register_stabilize(local_one_plus_erf, name='local_one_plus_erf') ...@@ -2608,7 +2607,7 @@ register_stabilize(local_one_plus_erf, name='local_one_plus_erf')
register_specialize(local_one_plus_erf, name='local_one_plus_erf') register_specialize(local_one_plus_erf, name='local_one_plus_erf')
#1-erf(x)=>erfc(x) #1-erf(x)=>erfc(x)
local_one_minus_erf = gof.PatternSub((T.sub, local_one_minus_erf = gof.PatternSub((T.sub,
dict(pattern='y', constraint = _is_1), dict(pattern='y', constraint = _is_1),
(T.erf, 'x')), (T.erf, 'x')),
(T.erfc, 'x'), (T.erfc, 'x'),
...@@ -2629,7 +2628,7 @@ register_specialize(local_one_minus_erf2) ...@@ -2629,7 +2628,7 @@ register_specialize(local_one_minus_erf2)
#1+(-erf(x))=>erfc(x) #1+(-erf(x))=>erfc(x)
#This is a different graph then the previous as the canonicalize don't work completly #This is a different graph then the previous as the canonicalize don't work completly
local_one_plus_neg_erf = gof.PatternSub((T.add, local_one_plus_neg_erf = gof.PatternSub((T.add,
dict(pattern='y', constraint = _is_1), dict(pattern='y', constraint = _is_1),
(T.neg,(T.erf, 'x'))), (T.neg,(T.erf, 'x'))),
(T.erfc, 'x'), (T.erfc, 'x'),
...@@ -2640,7 +2639,7 @@ register_specialize(local_one_plus_neg_erf, name='local_one_plus_neg_erf') ...@@ -2640,7 +2639,7 @@ register_specialize(local_one_plus_neg_erf, name='local_one_plus_neg_erf')
#(-1)+erf(x) => -erfc(x) #(-1)+erf(x) => -erfc(x)
#don't need erf(x)+(-1) as the canonicalize will put the -1 as the first argument. #don't need erf(x)+(-1) as the canonicalize will put the -1 as the first argument.
local_erf_minus_one = gof.PatternSub((T.add, local_erf_minus_one = gof.PatternSub((T.add,
dict(pattern='y', constraint = _is_minus1), dict(pattern='y', constraint = _is_minus1),
(T.erf, 'x')), (T.erf, 'x')),
(T.neg,(T.erfc, 'x')), (T.neg,(T.erfc, 'x')),
...@@ -2650,7 +2649,7 @@ register_stabilize(local_erf_minus_one, name='local_erf_minus_one') ...@@ -2650,7 +2649,7 @@ register_stabilize(local_erf_minus_one, name='local_erf_minus_one')
register_specialize(local_erf_minus_one, name='local_erf_minus_one') register_specialize(local_erf_minus_one, name='local_erf_minus_one')
#1-erfc(x) => erf(x) #1-erfc(x) => erf(x)
local_one_minus_erfc = gof.PatternSub((T.sub, local_one_minus_erfc = gof.PatternSub((T.sub,
dict(pattern='y', constraint = _is_1), dict(pattern='y', constraint = _is_1),
(T.erfc, 'x')), (T.erfc, 'x')),
(T.erf, 'x'), (T.erf, 'x'),
...@@ -2665,7 +2664,7 @@ local_one_minus_erfc2 = gof.PatternSub((T.add, ...@@ -2665,7 +2664,7 @@ local_one_minus_erfc2 = gof.PatternSub((T.add,
(T.erf, 'x'), (T.erf, 'x'),
allow_multiple_clients = True, allow_multiple_clients = True,
name='local_one_minus_erfc2') name='local_one_minus_erfc2')
register_canonicalize(local_one_minus_erfc2) register_canonicalize(local_one_minus_erfc2)
register_stabilize(local_one_minus_erfc2) register_stabilize(local_one_minus_erfc2)
register_specialize(local_one_minus_erfc2) register_specialize(local_one_minus_erfc2)
...@@ -2675,13 +2674,13 @@ local_one_minus_erfc3 = gof.PatternSub((T.add, ...@@ -2675,13 +2674,13 @@ local_one_minus_erfc3 = gof.PatternSub((T.add,
(T.erf, 'x'), (T.erf, 'x'),
allow_multiple_clients = True, allow_multiple_clients = True,
name='local_one_minus_erfc3') name='local_one_minus_erfc3')
register_canonicalize(local_one_minus_erfc3) register_canonicalize(local_one_minus_erfc3)
register_stabilize(local_one_minus_erfc3) register_stabilize(local_one_minus_erfc3)
register_specialize(local_one_minus_erfc3) register_specialize(local_one_minus_erfc3)
#1+(-erfc(x)) => erf(x) #1+(-erfc(x)) => erf(x)
#This is a different graph then the previous as the canonicalize don't work completly #This is a different graph then the previous as the canonicalize don't work completly
local_one_add_neg_erfc = gof.PatternSub((T.add, local_one_add_neg_erfc = gof.PatternSub((T.add,
dict(pattern='y', constraint = _is_1), dict(pattern='y', constraint = _is_1),
(T.neg,(T.erfc, 'x'))), (T.neg,(T.erfc, 'x'))),
(T.erf, 'x'), (T.erf, 'x'),
...@@ -2691,7 +2690,7 @@ register_stabilize(local_one_add_neg_erfc, name='local_one_add_neg_erfc') ...@@ -2691,7 +2690,7 @@ register_stabilize(local_one_add_neg_erfc, name='local_one_add_neg_erfc')
register_specialize(local_one_add_neg_erfc, name='local_one_add_neg_erfc') register_specialize(local_one_add_neg_erfc, name='local_one_add_neg_erfc')
#(-1)+erfc(-x)=>erf(x) #(-1)+erfc(-x)=>erf(x)
local_erf_neg_minus_one = gof.PatternSub((T.add, local_erf_neg_minus_one = gof.PatternSub((T.add,
dict(pattern='y', constraint = _is_minus1), dict(pattern='y', constraint = _is_minus1),
(T.erfc, (T.neg,'x'))), (T.erfc, (T.neg,'x'))),
(T.erf, 'x'), (T.erf, 'x'),
...@@ -2701,7 +2700,7 @@ register_stabilize(local_erf_neg_minus_one, name='local_erf_neg_minus_one') ...@@ -2701,7 +2700,7 @@ register_stabilize(local_erf_neg_minus_one, name='local_erf_neg_minus_one')
register_specialize(local_erf_neg_minus_one, name='local_erf_neg_minus_one') register_specialize(local_erf_neg_minus_one, name='local_erf_neg_minus_one')
#(-1)+erfc(-1*x)=>erf(x) #(-1)+erfc(-1*x)=>erf(x)
local_erf_neg_minus_one2 = gof.PatternSub((T.add, local_erf_neg_minus_one2 = gof.PatternSub((T.add,
dict(pattern='y', constraint = _is_minus1), dict(pattern='y', constraint = _is_minus1),
(T.erfc, (T.mul,-1,'x'))), (T.erfc, (T.mul,-1,'x'))),
(T.erf, 'x'), (T.erf, 'x'),
...@@ -2732,7 +2731,7 @@ def local_log_erfc(node): ...@@ -2732,7 +2731,7 @@ def local_log_erfc(node):
x = node.inputs[0].owner.inputs[0] x = node.inputs[0].owner.inputs[0]
stab_value = -x**2-T.log(x)-.5*T.log(numpy.pi)+T.log(1-1/(2*x**2)+3/(4*x**4)-15/(8*x**6)) stab_value = -x**2-T.log(x)-.5*T.log(numpy.pi)+T.log(1-1/(2*x**2)+3/(4*x**4)-15/(8*x**6))
if node.outputs[0].dtype=='float32': if node.outputs[0].dtype=='float32':
threshold = 10.0541949 threshold = 10.0541949
elif node.outputs[0].dtype=='float64': elif node.outputs[0].dtype=='float64':
...@@ -2749,7 +2748,7 @@ def local_log_erfc(node): ...@@ -2749,7 +2748,7 @@ def local_log_erfc(node):
#for float64: threshold=26.63 see at the end of the fct for the explaination #for float64: threshold=26.63 see at the end of the fct for the explaination
#for float32: threshold=9.3 see at the end of the fct for the explaination #for float32: threshold=9.3 see at the end of the fct for the explaination
#TODO: remove the contraint that their is only 2 inputs to mul and the exp(x**2) is the second. #TODO: remove the contraint that their is only 2 inputs to mul and the exp(x**2) is the second.
#TODO: at the test point 10 in float32, their is instability in the original value. #TODO: at the test point 10 in float32, their is instability in the original value.
# the original give -30.0, the stab -20.1 and in float64 -18.1. # the original give -30.0, the stab -20.1 and in float64 -18.1.
# Make the test don't generate error in that case! # Make the test don't generate error in that case!
@register_stabilize @register_stabilize
...@@ -2809,7 +2808,7 @@ def local_grad_log_erfc_neg(node): ...@@ -2809,7 +2808,7 @@ def local_grad_log_erfc_neg(node):
new_inputs.append(i) new_inputs.append(i)
return new_inputs return new_inputs
mul_inputs = check_input(mul_neg.owner.inputs) mul_inputs = check_input(mul_neg.owner.inputs)
#put the constant first #put the constant first
for i in range(len(mul_inputs)): for i in range(len(mul_inputs)):
if isinstance(i, Constant): if isinstance(i, Constant):
...@@ -2821,7 +2820,7 @@ def local_grad_log_erfc_neg(node): ...@@ -2821,7 +2820,7 @@ def local_grad_log_erfc_neg(node):
mul_inputs[i]=tmp mul_inputs[i]=tmp
break break
mul_neg = T.mul(*mul_inputs) mul_neg = T.mul(*mul_inputs)
try: try:
cst2 = get_constant_value(mul_neg.owner.inputs[0]) cst2 = get_constant_value(mul_neg.owner.inputs[0])
except TypeError: except TypeError:
...@@ -2840,25 +2839,25 @@ def local_grad_log_erfc_neg(node): ...@@ -2840,25 +2839,25 @@ def local_grad_log_erfc_neg(node):
return False return False
if cst2!=-1: if cst2!=-1:
if (not erfc_x.owner or erfc_x.owner.op != T.mul if (not erfc_x.owner or erfc_x.owner.op != T.mul
or len(erfc_x.owner.inputs)!=2): or len(erfc_x.owner.inputs)!=2):
#todo implement that case #todo implement that case
return False return False
if erfc_x.owner.inputs[1] is not mul_neg.owner.inputs[1]: if erfc_x.owner.inputs[1] is not mul_neg.owner.inputs[1]:
return False return False
x = erfc_x x = erfc_x
try: try:
cst = get_constant_value(erfc_x.owner.inputs[0]) cst = get_constant_value(erfc_x.owner.inputs[0])
except TypeError: except TypeError:
return False return False
if cst2 != -cst*2: if cst2 != -cst*2:
return False return False
#The constant is valid. Must check that the #The constant is valid. Must check that the
elif erfc_x is not x: elif erfc_x is not x:
return False return False
else: else:
return False return False
...@@ -3014,7 +3013,7 @@ def local_elemwise_fusion_op(OP): ...@@ -3014,7 +3013,7 @@ def local_elemwise_fusion_op(OP):
try: try:
s_new_out.owner.op.c_code(s_new_out.owner, "test_presence_of_c_code", s_new_out.owner.op.c_code(s_new_out.owner, "test_presence_of_c_code",
["x" for x in s_g], ["x" for x in s_g],
"z",{}) "z",{})
except MethodNotDefined: except MethodNotDefined:
_logger.info("%s does not implement the c_code function. As well as being potentially slow, this disables loop fusion of this op." % str(s_new_out.owner.op)) _logger.info("%s does not implement the c_code function. As well as being potentially slow, this disables loop fusion of this op." % str(s_new_out.owner.op))
return False return False
...@@ -3046,19 +3045,18 @@ def local_elemwise_fusion_op(OP): ...@@ -3046,19 +3045,18 @@ def local_elemwise_fusion_op(OP):
return False return False
# print "local_elemwise_fusion: FUSED",nb_elemwise+1,"elemwise!" # print "local_elemwise_fusion: FUSED",nb_elemwise+1,"elemwise!"
#we fuse as many that we can at the same time to make debug mode faster #we fuse as many that we can at the same time to make debug mode faster
#debug mode will be faster as it won't test all intermediate step. #debug mode will be faster as it won't test all intermediate step.
while True: while True:
ret = local_fuse(n) ret = local_fuse(n)
if ret is not False and ret is not None: if ret is not False and ret is not None:
#print n,ret #print n,ret
#import pdb;pdb.set_trace()
assert len(ret)==len(n.outputs) assert len(ret)==len(n.outputs)
assert len(ret)==1 assert len(ret)==1
n = ret[0].owner n = ret[0].owner
else: break else: break
return n.outputs return n.outputs
return local_fuse return local_fuse
......
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