Skip to content

P
pytensor
提交 60b5ccc2 authored 作者: abergeron's avatar abergeron
浏览文件
操作

Merge pull request #1972 from nouiz/cumfct

Add check on the output of NumPy.
上级 8127a8aa 0a6ebc1f
隐藏空白字符变更
内嵌 并排
正在显示 包含 70 行增加 和 48 行删除
theano/tensor/extra_ops.py
浏览文件 @ 60b5ccc2
......@@ -10,7 +10,6 @@ tensor = basic
from theano.gradient import DisconnectedType
class CumsumOp(theano.Op):
# See function cumsum for docstring
def __init__(self, axis=None):
......@@ -45,8 +44,8 @@ class CumsumOp(theano.Op):
# We need to reverse the gradients along ``self.axis``,
# compute cumsum, then reverse again
reverse_slicing = [slice(None,None,None)] * gi.ndim
reverse_slicing[self.axis] = slice(None,None,-1)
reverse_slicing = [slice(None, None, None)] * gi.ndim
reverse_slicing[self.axis] = slice(None, None, -1)
reverse_slicing = tuple(reverse_slicing)
return [cumsum(gi[reverse_slicing], self.axis)[reverse_slicing]]
......@@ -74,13 +73,19 @@ class CumsumOp(theano.Op):
if (!%(z)s)
%(fail)s;
{
PyArray_CumSum(%(x)s, NPY_MAXDIMS, PyArray_TYPE((PyArrayObject*) py_%(x)s), %(z)s);
Py_XDECREF(%(z)s); // Because PyArray_CumSum returns a newly created reference on %(z)s.
PyObject * t = PyArray_CumSum(
%(x)s, NPY_MAXDIMS,
PyArray_TYPE((PyArrayObject*) py_%(x)s), %(z)s);
if (!t){
%(fail)s;
}
// Because PyArray_CumSum returns a newly created reference on t.
Py_XDECREF(t);
}
""" % locals()
else:
code = """
if(!(%(z)s && PyArray_CompareLists(PyArray_DIMS(%(z)s), PyArray_DIMS(%(x)s), PyArray_NDIM(%(x)s)) ))
if(!(%(z)s && PyArray_CompareLists(PyArray_DIMS(%(z)s), PyArray_DIMS(%(x)s), PyArray_NDIM(%(x)s))))
{
Py_XDECREF(%(z)s);
%(z)s = (PyArrayObject*) PyArray_SimpleNew(PyArray_NDIM(%(x)s), PyArray_DIMS(%(x)s), PyArray_TYPE((PyArrayObject*) py_%(x)s));
......@@ -89,15 +94,22 @@ class CumsumOp(theano.Op):
if (!%(z)s)
%(fail)s;
{
PyArray_CumSum(%(x)s, %(axis)s, PyArray_TYPE((PyArrayObject*) py_%(x)s), %(z)s);
Py_XDECREF(%(z)s); // Because PyArray_CumSum returns a newly created reference on %(z)s.
PyObject * t = PyArray_CumSum(
%(x)s, %(axis)s,
PyArray_TYPE((PyArrayObject*) py_%(x)s), %(z)s);
if (!t){
%(fail)s;
}
// Because PyArray_CumSum returns a newly created reference on t.
Py_XDECREF(t);
}
""" % locals()
return code
def c_code_cache_version(self):
return (3,)
return (6,)
def __str__(self):
return "%s{%s}" % (self.__class__.__name__, self.axis)
......@@ -183,8 +195,14 @@ class CumprodOp(theano.Op):
if (!%(z)s)
%(fail)s;
{
PyArray_CumProd(%(x)s, NPY_MAXDIMS, PyArray_TYPE((PyArrayObject*) py_%(x)s), %(z)s);
Py_XDECREF(%(z)s); // Because PyArray_CumSum returns a newly created reference on %(z)s.
PyObject * t = PyArray_CumProd(
%(x)s, NPY_MAXDIMS,
PyArray_TYPE((PyArrayObject*) py_%(x)s), %(z)s);
if (!t){
%(fail)s;
}
// Because PyArray_CumSum returns a newly created reference on t.
Py_XDECREF(t);
}
""" % locals()
else:
......@@ -198,15 +216,21 @@ class CumprodOp(theano.Op):
if (!%(z)s)
%(fail)s;
{
PyArray_CumProd(%(x)s, %(axis)s, PyArray_TYPE((PyArrayObject*) py_%(x)s), %(z)s);
Py_XDECREF(%(z)s); // Because PyArray_CumSum returns a newly created reference on %(z)s.
PyObject * t = PyArray_CumProd(
%(x)s, %(axis)s,
PyArray_TYPE((PyArrayObject*) py_%(x)s), %(z)s);
if (!t){
%(fail)s;
}
// Because PyArray_CumSum returns a newly created reference on t.
Py_XDECREF(t);
}
""" % locals()
return code
def c_code_cache_version(self):
return (2,)
return (4,)
def __str__(self):
return "%s{%s}" % (self.__class__.__name__, self.axis)
......@@ -319,7 +343,7 @@ class BinCountOp(theano.Op):
def __eq__(self, other):
return (type(self) == type(other) and
self.minlength == other.minlength)
self.minlength == other.minlength)
def __hash__(self):
return hash(type(self)) ^ hash(self.minlength)
......@@ -346,8 +370,8 @@ class BinCountOp(theano.Op):
if x.dtype in numpy_unsupported_dtypes:
raise TypeError(
("Input dtypes %s are not supported by numpy.bincount, "
% numpy_unsupported_dtypes), x.dtype)
("Input dtypes %s are not supported by numpy.bincount, "
% numpy_unsupported_dtypes), x.dtype)
if x.ndim != 1:
raise TypeError("Inputs must be of dimension 1.")
......@@ -473,9 +497,9 @@ class RepeatOp(theano.Op):
if repeats.dtype in numpy_unsupported_dtypes:
raise TypeError(
("dtypes %s are not supported by numpy.repeat "
"for the 'repeats' parameter, "
% str(numpy_unsupported_dtypes)), repeats.dtype)
("dtypes %s are not supported by numpy.repeat "
"for the 'repeats' parameter, "
% str(numpy_unsupported_dtypes)), repeats.dtype)
if self.axis is None:
broadcastable = [False]
......@@ -532,7 +556,7 @@ class RepeatOp(theano.Op):
repeats = node.inputs[1]
out_shape = list(i0_shapes)
#uint64 shape are not supported.
# uint64 shape are not supported.
dtype = None
if repeats.dtype in ['uint8', 'uint16', 'uint32']:
dtype = 'int64'
......@@ -597,9 +621,9 @@ class Bartlett(gof.Op):
if M.ndim != 0:
raise TypeError('%s only works on scalar input'
% self.__class__.__name__)
elif (not M.dtype.startswith('int')) and \
(not M.dtype.startswith('uint')):
# dtype is a theano attribute here
elif (not M.dtype.startswith('int') and
not M.dtype.startswith('uint')):
# dtype is a theano attribute here
raise TypeError('%s only works on integer input'
% self.__class__.__name__)
return gof.Apply(self, [M], [tensor.dvector()])
......@@ -612,7 +636,8 @@ class Bartlett(gof.Op):
def infer_shape(self, node, in_shapes):
temp = node.inputs[0]
M = tensor.switch(tensor.lt(temp, 0),
tensor.cast(0, temp.dtype), temp)
tensor.cast(0, temp.dtype),
temp)
return [[M]]
def grad(self, inputs, output_grads):
......@@ -620,7 +645,7 @@ class Bartlett(gof.Op):
bartlett_ = Bartlett()
#I create a function only to have the doc show well.
# I create a function only to have the doc show well.
def bartlett(M):
"""An instance of this class returns the Bartlett spectral window in the
time-domain. The Bartlett window is very similar to a triangular window,
......@@ -668,7 +693,7 @@ class FillDiagonal(gof.Op):
val = tensor.cast(val, dtype=scalar.upcast(a.dtype, val.dtype))
if val.dtype != a.dtype:
raise TypeError('%s: type of second parameter must be the same as'
' the first\'s' % self.__class__.__name__)
' the first\'s' % self.__class__.__name__)
return gof.Apply(self, [a, val], [a.type()])
def perform(self, node, inputs, output_storage):
......@@ -698,7 +723,8 @@ class FillDiagonal(gof.Op):
return [None, None]
elif a.ndim > 2:
raise NotImplementedError('%s: gradient is currently implemented'
' for matrices only' % self.__class__.__name__)
' for matrices only' %
self.__class__.__name__)
wr_a = fill_diagonal(grad, 0) # valid for any number of dimensions
# diag is only valid for matrices
import theano.sandbox.linalg
......@@ -707,7 +733,7 @@ class FillDiagonal(gof.Op):
fill_diagonal_ = FillDiagonal()
#I create a function only to have the doc show well.
# I create a function only to have the doc show well.
def fill_diagonal(a, val):
""" Returns a copy of an array with all
elements of the main diagonal set to a specified scalar value.
......@@ -730,7 +756,6 @@ def fill_diagonal(a, val):
return fill_diagonal_(a, val)
class FillDiagonalOffset(gof.Op):
# See function fill_diagonal_offset for docstring
def __eq__(self, other):
......@@ -753,10 +778,10 @@ class FillDiagonalOffset(gof.Op):
raise TypeError('%s: first parameter must have exactly'
' two dimensions' % self.__class__.__name__)
elif val.ndim != 0:
raise TypeError('%s: second parameter must be a scalar'\
raise TypeError('%s: second parameter must be a scalar'
% self.__class__.__name__)
elif offset.ndim != 0:
raise TypeError('%s: third parameter must be a scalar'\
raise TypeError('%s: third parameter must be a scalar'
% self.__class__.__name__)
val = tensor.cast(val, dtype=scalar.upcast(a.dtype, val.dtype))
if val.dtype != a.dtype:
......@@ -764,11 +789,9 @@ class FillDiagonalOffset(gof.Op):
' as the first\'s' % self.__class__.__name__)
elif offset.dtype[:3] != 'int':
raise TypeError('%s: type of third parameter must be as integer'
' use theano.tensor.cast( input, \'int32/int64\')' \
' use theano.tensor.cast( input, \'int32/int64\')'
% self.__class__.__name__)
return gof.Apply(self, [a, val, offset], [a.type()])
def perform(self, node, inputs, output_storage):
......@@ -780,24 +803,23 @@ class FillDiagonalOffset(gof.Op):
"""
Note: The fill_diagonal only support rectangular matrix. The output
of tall matrix is "wrapped", which is an option in numpy 1.9.0
but was regarded as a bug in numpy 1.6.2. Here I implement the
but was regarded as a bug in numpy 1.6.2. Here I implement the
fill_diagonal_offset with unwrapped output, so fill_diagonal_offset
supports tall matrix.(This make a little difference between the output
of fill_diagonal and fill_diagonal_offset only in the case of tall
of fill_diagonal and fill_diagonal_offset only in the case of tall
matrix)
"""
if offset >= 0:
start = offset
num_of_step = min( min(width,height), width - offset)
num_of_step = min(min(width, height), width - offset)
else:
start = - offset * a.shape[1]
num_of_step = min( min(width,height), height + offset)
num_of_step = min(min(width, height), height + offset)
step = a.shape[1] + 1
end = start + step * num_of_step
# Write the value out into the diagonal.
a.flat[start:end:step] = val
output_storage[0][0] = a
def grad(self, inp, cost_grad):
......@@ -812,19 +834,19 @@ class FillDiagonalOffset(gof.Op):
if (a.dtype.startswith('complex')):
return [None, None]
# only valid for matrices
wr_a = fill_diagonal_offset(grad, 0, offset)
offset_abs = basic.abs_( offset )
pos_offset_flag = basic.ge( offset, 0 )
neg_offset_flag = basic.lt( offset, 0 )
min_wh = basic.minimum(width,height)
# only valid for matrices
wr_a = fill_diagonal_offset(grad, 0, offset)
offset_abs = basic.abs_(offset)
pos_offset_flag = basic.ge(offset, 0)
neg_offset_flag = basic.lt(offset, 0)
min_wh = basic.minimum(width, height)
start = offset * pos_offset_flag + offset_abs * width \
* neg_offset_flag
num_of_step = basic.minimum( min_wh, width * pos_offset_flag
+ height * neg_offset_flag - offset_abs )
step = a.shape[1] + 1
end = start + step * num_of_step
......
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