Skip to content

P
pytensor
提交 c7db795f authored 作者: bergstrj@iro.umontreal.ca's avatar bergstrj@iro.umontreal.ca
浏览文件
操作

moved scalar_ops to scalar, removed old elemwise.py.

上级 f1153866
显示空白字符变更
内嵌 并排
正在显示 包含 193 行增加 和 856 行删除
_test_elemwise2.py
浏览文件 @ c7db795f
...@@ -5,7 +5,7 @@ import unittest ...@@ -5,7 +5,7 @@ import unittest
from gof import ResultBase, Op, Env, modes from gof import ResultBase, Op, Env, modes
import gof import gof
from scalar_ops import * from scalar import *
from elemwise2 import * from elemwise2 import *
......
_test_gradient.py
浏览文件 @ c7db795f
...@@ -22,7 +22,7 @@ class _test_grad_sources_inputs(unittest.TestCase): ...@@ -22,7 +22,7 @@ class _test_grad_sources_inputs(unittest.TestCase):
try: try:
grad_sources_inputs([(a.out, 1)], None) grad_sources_inputs([(a.out, 1)], None)
except ValueError, e: except ValueError, e:
self.failUnless(e[0] is gradient._msg_retNone) self.failUnless(e[0] is gradient._msg_retType)
return return
self.fail() self.fail()
def test_retNone1_b(self): def test_retNone1_b(self):
......
_test_scalar_ops.py → _test_scalar.py
浏览文件 @ c7db795f
...@@ -4,7 +4,7 @@ import unittest ...@@ -4,7 +4,7 @@ import unittest
from gof import ResultBase, Op, Env, modes from gof import ResultBase, Op, Env, modes
import gof import gof
from scalar_ops import * from scalar import *
def inputs(): def inputs():
......
elemwise.py deleted 100644 → 0
浏览文件 @ f1153866
from copy import copy
from gof import Op, Destroyer
from gof.utils import AbstractFunctionError
class Elemwise(Op):
def var_desc(self):
raise AbstractFunctionError()
def c_var_names(self):
idesc, odesc = self.var_desc()
return [[i[0] for i in idesc],
[o[0] for o in odesc]]
def loop_variables(self):
idesc, odesc = self.var_desc()
return [[i[0] for i in idesc if i[1]],
[o[0] for o in odesc if o[1]]]
def c_code_init(self):
raise AbstractFunctionError()
def c_code_foreach(self):
raise AbstractFunctionError()
def c_code_finalize(self):
raise AbstractFunctionError()
def __c_code(self):
try: before = self.c_code_init()
except AbstractFunctionError: before = ""
during = self.c_code_foreach()
try: after = self.c_code_finalize()
except AbstractFunctionError: after = ""
idesc, odesc = self.var_desc()
input_loop_vars = [x[0] for x in idesc if x[1]]
output_loop_vars = [x[0] for x in odesc if x[1]]
aliases = {}
try:
dmap = self.destroy_map()
except:
dmap = {}
if dmap != {}:
for (oname, isloopvar), output in zip(odesc, self.outputs):
if isloopvar:
for input in dmap.get(output, []):
iname = idesc[self.inputs.index(input)][0]
aliases["%("+oname+")s"] = "%("+iname+")s"
output_loop_vars.remove(oname)
ret = elemwise_wrap(before, during, after,
["%("+v+")s" for v in input_loop_vars],
["%("+v+")s" for v in output_loop_vars],
aliases)
return ret
def c_validate_update(self, input_names, output_names, sub):
sub = dict(sub)
icvn, ocvn = self.c_var_names()
for real, tosub in zip(input_names + output_names, icvn + ocvn):
sub[tosub] = real
(valupd, valupd_cleanup), (code, code_cleanup) = self.__c_code()
return valupd % sub
def c_validate_update_cleanup(self, input_names, output_names, sub):
sub = dict(sub)
icvn, ocvn = self.c_var_names()
for real, tosub in zip(input_names + output_names, icvn + ocvn):
sub[tosub] = real
(valupd, valupd_cleanup), (code, code_cleanup) = self.__c_code()
return valupd_cleanup % sub
def c_code(self, input_names, output_names, sub):
sub = dict(sub)
icvn, ocvn = self.c_var_names()
for real, tosub in zip(input_names + output_names, icvn + ocvn):
sub[tosub] = real
(valupd, valupd_cleanup), (code, code_cleanup) = self.__c_code()
return code % sub
def c_code_cleanup(self, input_names, output_names, sub):
sub = dict(sub)
icvn, ocvn = self.c_var_names()
for real, tosub in zip(input_names + output_names, icvn + ocvn):
sub[tosub] = real
(valupd, valupd_cleanup), (code, code_cleanup) = self.__c_code()
return code_cleanup % sub
@classmethod
def inplace_version(cls, dmap = {0:0}):
class Ret(cls, Destroyer):
def destroy_map(self):
return {self.outputs[0]: [self.inputs[0]]}
return Ret
class Looper:
def __init__(self, declare, init, acquire, next, cleanup):
self.declare = declare
self.init = init
self.acquire = acquire
self.next = next
self.cleanup = cleanup
def fill(self, **template):
ret = copy(self)
ret.declare %= template
ret.init %= template
ret.acquire %= template
ret.next %= template
ret.cleanup %= template
return ret
general_looper = Looper(declare = """
PyArrayIterObject* %(iter)s_%(name)s = NULL;
//dtype_%(name)s& %(name)s_i;
""",
init = """
%(iter)s_%(name)s = (PyArrayIterObject*)PyArray_IterNew((PyObject*)%(name)s);
if (%(iter)s_%(name)s == NULL) {
PyErr_SetString(PyExc_ValueError, "Could not make an iterator over variable %(name)s.");
%%(fail)s;
}
""",
acquire = "dtype_%(name)s& %(name)s_i = *((dtype_%(name)s*)%(iter)s_%(name)s->dataptr);",
next = "PyArray_ITER_NEXT(%(iter)s_%(name)s);",
cleanup = "if (%(iter)s_%(name)s) Py_DECREF(%(iter)s_%(name)s);")
contiguous_looper = Looper(declare = """
dtype_%(name)s* %(iter)s_%(name)s = NULL;
//dtype_%(name)s %(name)s_i;
""",
init = "%(iter)s_%(name)s = (dtype_%(name)s*)PyArray_DATA(%(name)s);",
acquire = "dtype_%(name)s& %(name)s_i = *%(iter)s_%(name)s;",
next = "%(iter)s_%(name)s++;",
cleanup = "")
alias_looper = Looper(declare = "", #dtype_%(name)s%(type_suffix)s %(name)s_i;",
init = "",
acquire = "dtype_%(name)s& %(name)s_i = %(other)s_i;",
next = "",
cleanup = "")
def elemwise_loopcode(loopcode, loopers):
def make_block(loopers, type):
return "\n".join([getattr(looper, type) for looper in loopers])
template = dict(loopcode = loopcode)
for block_type in ['declare', 'init', 'next', 'acquire', 'cleanup']:
template[block_type] = make_block(loopers, block_type)
code = """
%(init)s
while (__SIZE--) {
%(acquire)s
%(loopcode)s
%(next)s
}
""" % template
cleanup = """
%(cleanup)s
""" % template
return template['declare'], code, cleanup
def elemwise_wrap(beforeloop, inloop, afterloop, input_loop_vars, output_loop_vars, aliases):
if len(input_loop_vars) == 0:
raise Exception("You need at least one input loop variable.")
validate = """
npy_intp nd = %(first_loop_var)s->nd;
npy_intp* dims = %(first_loop_var)s->dimensions;
npy_intp* dims2;
""" % dict(first_loop_var = input_loop_vars[0])
for other_loop_var in input_loop_vars[1:]:
validate += """
if (%(other_loop_var)s->nd != nd) {
PyErr_SetString(PyExc_ValueError, \"The number of dimensions of the inputs do not match.\");
%%(fail)s
}
dims2 = %(other_loop_var)s->dimensions;
for (int i = 0; i < nd; i++) {
if (dims2[i] != dims[i]) {
PyErr_SetString(PyExc_ValueError, \"The dimensions of the inputs do not match.\");
%%(fail)s;
}
}
""" % dict(other_loop_var = other_loop_var)
update = ""
for output_loop_var in output_loop_vars:
update += """
if (!%(output_loop_var)s) {
%(output_loop_var)s = (PyArrayObject*)PyArray_EMPTY(nd, dims, type_num_%(output_loop_var)s, 0);
}
else {
PyArray_Dims new_dims;
new_dims.len = nd;
new_dims.ptr = dims;
PyObject* success = PyArray_Resize(%(output_loop_var)s, &new_dims, 0, PyArray_CORDER);
if (!success) {
// If we can't resize the ndarray we have we can allocate a new one.
PyErr_Clear();
Py_XDECREF(%(output_loop_var)s);
%(output_loop_var)s = (PyArrayObject*)PyArray_EMPTY(nd, dims, type_num_%(output_loop_var)s, 0);
}
}
if (!%(output_loop_var)s) {
%%(fail)s
}
""" % dict(output_loop_var = output_loop_var)
validate_update = validate + update
validate_update_cleanup = ""
general_loopers = []
general_loopers += [general_looper.fill(name = loop_var, iter = "iter_general_") for loop_var in input_loop_vars + output_loop_vars]
general_loopers += [alias_looper.fill(name = v1, other = v2, iter = "iter_general_") for v1, v2 in aliases.items()]
contig_loopers = []
contig_loopers += [contiguous_looper.fill(name = loop_var, iter = "iter_contig_") for loop_var in input_loop_vars + output_loop_vars]
contig_loopers += [alias_looper.fill(name = v1, other = v2, iter = "iter_contig_") for v1, v2 in aliases.items()]
general_decl, general_loop, general_loop_cleanup = elemwise_loopcode(inloop, general_loopers)
contig_decl, contig_loop, contig_loop_cleanup = elemwise_loopcode(inloop, contig_loopers)
contiguity_check = "".join(["all_c_contiguous &= PyArray_ISCARRAY(%(loop_var)s);\n" \
"all_f_contiguous &= PyArray_ISFARRAY(%(loop_var)s);\n" \
% dict(loop_var = loop_var)
for loop_var in input_loop_vars + output_loop_vars])
v1 = input_loop_vars[0]
code = """
%(general_decl)s
%(contig_decl)s
npy_intp __SIZE = PyArray_SIZE(%(v1)s);
bool all_c_contiguous = 1;
bool all_f_contiguous = 1;
%(beforeloop)s
%(contiguity_check)s
if (all_c_contiguous || all_f_contiguous) {
%(contig_loop)s
}
else {
%(general_loop)s
}
%(afterloop)s
""" % locals()
cleanup = """
if (all_c_contiguous || all_f_contiguous) {
%(contig_loop_cleanup)s
}
else {
%(general_loop_cleanup)s
}
""" % locals()
# print """
# VALIDATE_UPDATE
# %(validate_update)s
# CODE
# %(code)s
# CLEANUP
# %(cleanup)s
# VU_CLEANUP
# %(validate_update_cleanup)s
# """ % locals()
return (validate_update, validate_update_cleanup), (code, cleanup)
# import core
# def elemwise_loopcode(loopcode, init_template, next_template, acquire_template, cleanup_template, loop_vars, writable_loop_vars, aliases):
# all_loop_vars = loop_vars + writable_loop_vars
# template = dict(
# init = "".join([init_template % dict(loop_var = loop_var) for loop_var in all_loop_vars]),
# next = "".join([next_template % dict(loop_var = loop_var) for loop_var in all_loop_vars]),
# cleanup = "".join([cleanup_template % dict(loop_var = loop_var) for loop_var in all_loop_vars]),
# idefs = "".join([("%(loop_var)s_dtype %(loop_var)s_i = " + acquire_template + ";\n")
# % dict(loop_var = loop_var) for loop_var in loop_vars]),
# odefs = "".join([("%(loop_var)s_dtype& %(loop_var)s_i = " + acquire_template + ";\n")
# % dict(loop_var = loop_var) for loop_var in writable_loop_vars]),
# aliasdefs = "".join(["%(v1)s_dtype %(v1)s_i = %(v2)s_i;\n" % dict(v1=v1, v2=v2)
# for v1, v2 in aliases.items()]),
# loopcode = loopcode
# )
# code = """
# %(init)s
# while (__elemwise_size--) {
# %(idefs)s
# %(odefs)s
# %(aliasdefs)s
# %(loopcode)s
# %(next)s
# }
# %(cleanup)s
# """ % template
# return code
# def elemwise_wrap(beforeloop, inloop, afterloop, loop_vars, writable_loop_vars, aliases):
# check_init = """
# npy_intp nd = %(loop_var)s->nd;
# npy_intp* dims = %(loop_var)s->dimensions;
# npy_intp* dims2;
# """
# check = """
# if (%(loop_var)s->nd != nd) {
# PyErr_SetString(PyExc_ValueError, \"The number of dimensions of the inputs do not match.\");
# }
# dims2 = %(loop_var)s->dimensions;
# for (int i = 0; i < nd; i++) {
# if (dims2[i] != dims[i]) {
# PyErr_SetString(PyExc_ValueError, \"The dimensions of the inputs do not match.\");
# return 1;
# }
# }
# """
# general_init = "PyArrayIterObject* %(loop_var)s_iter = (PyArrayIterObject*)PyArray_IterNew((PyObject*)%(loop_var)s);\n"
# # "if (%(loop_var)s_iter == NULL) {\n" \
# # " PyErr_SetString(PyExc_ValueError, \"Could not make an iterator over variable %(loop_var)s.\");\n" \
# # " return 1;\n" \
# # "}\n"
# general_next = "PyArray_ITER_NEXT(%(loop_var)s_iter);\n"
# general_acquire = "*((%(loop_var)s_dtype*)%(loop_var)s_iter->dataptr)";
# general_cleanup = "if (%(loop_var)s_iter) Py_DECREF(%(loop_var)s_iter);\n";
# contiguous_init = "%(loop_var)s_dtype* __restrict__ %(loop_var)s_iter = (%(loop_var)s_dtype*)PyArray_DATA(%(loop_var)s);\n"
# contiguous_next = "%(loop_var)s_iter++;\n"
# contiguous_acquire = "*%(loop_var)s_iter"
# contiguous_cleanup = ""
# all_loop_vars = loop_vars + writable_loop_vars
# v1 = (loop_vars + writable_loop_vars)[0]
# template = dict(
# v1 = v1,
# check_init = check_init % dict(loop_var = v1),
# check = "\n".join([check % dict(loop_var = loop_var) for loop_var in loop_vars + writable_loop_vars if loop_var is not v1]),
# beforeloop = beforeloop,
# general_loop = elemwise_loopcode(
# inloop,
# general_init, general_next, general_acquire, general_cleanup,
# loop_vars, writable_loop_vars, aliases),
# contiguous_loop = elemwise_loopcode(
# inloop,
# contiguous_init, contiguous_next, contiguous_acquire, contiguous_cleanup,
# loop_vars, writable_loop_vars, aliases),
# contiguity_check = "".join(["all_c_contiguous &= PyArray_ISCARRAY(%(loop_var)s);\n" \
# "all_f_contiguous &= PyArray_ISFARRAY(%(loop_var)s);\n" \
# % dict(loop_var = loop_var)
# for loop_var in all_loop_vars]),
# afterloop = afterloop)
# code = """
# {
# %(check_init)s
# %(check)s
# }
# npy_intp __elemwise_size = PyArray_SIZE(%(v1)s);
# %(beforeloop)s
# bool all_c_contiguous = 1;
# bool all_f_contiguous = 1;
# %(contiguity_check)s
# if (all_c_contiguous || all_f_contiguous) {
# %(contiguous_loop)s
# }
# else {
# %(general_loop)s
# }
# %(afterloop)s
# """ % template
# return code
# class elemwise(omega_op):
# @staticmethod
# def __clsinit__(cls, name, bases, dct):
# for fname in ['c_init', 'c_foreach', 'c_finalize']:
# gof.make_static(cls, fname)
# # make impl, grad, etc. static methods
# omega_op.__clsinit__(cls, name, bases, dct)
# def TOGO_specs(self):
# try:
# return self.specs(*[input.spec for input in self.inputs])
# except NotImplementedError:
# inames, onames = self.variable_names()
# linames, lonames = self.loop_variables()
# for oname in onames:
# if oname not in lonames:
# raise Exception("cannot infer a specification automatically for variable " \
# "%s.%s because it is not part of the elementwise loop - "\
# "please override the specs method" % (self.__class__.__name__, oname))
# shape, dtype = None, None
# for iname, input in zip(inames, self.inputs):
# if iname in linames:
# if input.spec:
# shape = input.spec[2]
# if shape is None:
# raise Exception("cannot infer a specification automatically for output variables " \
# "because there is no input variable in the loop from which to get the shape, "\
# "or their shape is unknown")
# try:
# dtype = core.upcast(*[input.spec[1]
# for iname, input in zip(inames, self.inputs)
# if input.spec[0] is numpy.ndarray])
# except IndexError:
# raise Exception("not all numpy inputs are specified")
# dmap = self.destroy_map()
# res = []
# for output in self.outputs:
# inplace_inputs = dmap.get(output, [])
# if inplace_inputs:
# assert len(inplace_inputs) == 1
# res.append(inplace_inputs[0].spec)
# else:
# res.append((numpy.ndarray, dtype, shape))
# if self.nout == 1:
# return res[0]
# else:
# return res
# def TOGO_alloc(self, except_list = []):
# dmap = self.destroy_map()
# vmap = self.view_map()
# gof.PythonOp.alloc(self, except_list = except_list + dmap.keys())
# for output, (input, ) in dmap.items():
# if output not in except_list:
# output.set_value(input.data)
# def refresh_shape(self):
# """Make the output have the right stuff"""
# if len(self.outputs) > 1:
# raise NotImplementedError('multiple outputs')
# dmap = self.destroy_map()
# vmap = self.view_map()
# if dmap != {} or vmap != {}:
# raise NotImplementedError('destroys or views confuse things',
# self.__class__, dmap, vmap)
# # take the shape of the leftmost loop_variable input
# inames, onames = self.variable_names()
# linames, lonames = self.loop_variables()
# unknown_output_names = [n for n in onames if n not in lonames]
# if len(unknown_output_names):
# raise Exception("cannot infer a specification automatically for variables " \
# "%s.{%s} because it is not part of the elementwise loop - "\
# "please override the specs method" %
# (self.__class__.__name__, str(unknown_output_names)))
# # shape is leftmost loop-variable input
# input_loop_shapes = [i.shape for n,i in zip(inames, self.inputs) if n in linames]
# if len(input_loop_shapes) == 0:
# raise Exception("cannot infer a specification automatically for output variables " \
# "because there is no input loop variable ")
# for i in xrange(1,len(input_loop_shapes)):
# if input_loop_shapes[i] != input_loop_shapes[0]:
# raise Exception("Input loop variables have different shapes", self.__class__)
# return input_loop_shapes[0]
# def refresh_dtype(self):
# return core.upcast(*[i.dtype for i in self.inputs if hasattr(i, 'dtype')])
# @classmethod
# def set_impl(cls, impl):
# gof.lib.make_static(cls, 'impl')
# @staticmethod
# def is_loop_var(name):
# return name.endswith("_i")
# @staticmethod
# def extract_name(name):
# if name.endswith("_i"):
# return name[:-2]
# else:
# return name
# @classmethod
# def variable_names(cls):
# (inames, onames), _1, _2, _3 = inspect.getargspec(cls.c_foreach)
# spec = ([cls.extract_name(name) for name in inames],
# [cls.extract_name(name) for name in onames])
# if cls.c_init is not elemwise.c_init:
# (inames, onames), _1, _2, _3 = inspect.getargspec(cls.c_init)
# assert spec == (list(inames), list(onames))
# if cls.c_finalize is not elemwise.c_finalize:
# (inames, onames), _1, _2, _3 = inspect.getargspec(cls.c_finalize)
# assert spec == (list(inames), list(onames))
# return spec
# @classmethod
# def loop_variables(cls):
# (inames, onames), _1, _2, _3 = inspect.getargspec(cls.c_foreach)
# return ([cls.extract_name(name) for name in inames if cls.is_loop_var(name)],
# [cls.extract_name(name) for name in onames if cls.is_loop_var(name)])
# def _c_init(self):
# return self.c_init(self.inputs, self.outputs)
# def c_init(inputs, outputs):
# return ""
# def _c_foreach(self):
# return self.c_foreach(self.inputs, self.outputs)
# def c_foreach(inputs, outputs):
# raise NotImplementedError()
# def _c_finalize(self):
# return self.c_finalize(self.inputs, self.outputs)
# def c_finalize(inputs, outputs):
# return ""
# def c_code(self, converters = None, elemwise_wrap = elemwise_wrap):
# def mangle(name):
# if name.endswith("_i"):
# return name[:-2]
# else:
# return name
# try:
# self._c_impl()
# raise Exception("c_impl is not used by elemwise ops - define behavior in c_foreach instead")
# except NotImplementedError:
# pass
# before = self._c_init()
# during = self._c_foreach()
# after = self._c_finalize()
# (inames, onames) = self.variable_names()
# (linames, lonames) = self.loop_variables()
# aliases = {}
# dmap = self.destroy_map()
# if dmap != {}:
# for oname, output in zip(onames, self.outputs):
# if oname in lonames:
# for input in dmap.get(output, []):
# aliases[inames[self.inputs.index(input)]] = oname
# behavior = elemwise_wrap(before, during, after,
# [name for name in linames if name not in aliases],
# lonames,
# aliases)
# return cgen(self.__class__.__name__, behavior, inames + onames, self.inputs + self.outputs, converters)
# @classmethod
# def inplace_version(cls, dmap = {0: 0}):
# inames, onames = cls.variable_names()
# linames, lonames = cls.loop_variables()
# for i, oname in enumerate(onames):
# if i in dmap:
# assert oname in lonames
# class C(cls):
# def destroy_map(self):
# assert cls.destroy_map(self) == {}
# ret = {}
# for output, input in dmap.items():
# ret[self.outputs[output]] = [self.inputs[input]]
# return ret
# def _impl(self):
# if self.impl is not cls.impl:
# # If the user sets his own inplace operation, we use it
# return cls._impl(self)
# else:
# res = cls._impl(self)
# if isinstance(res, (list, tuple)):
# res = copy.copy(res)
# else:
# res = [res]
# for output, input in dmap.items():
# # The default implementation returned a copy, so we just
# # overwrite the original input with the contents of that copy
# # This is not meant to be efficient, only correct.
# #
# # TODO: change this to use set_value_inplace
# a = self.inputs[input].data
# a[:] = res[output]
# res[output] = a
# if len(res) == 1:
# return res[0]
# else:
# return res
# if dmap == {0:0}:
# C.__name__ = cls.__name__ + "_inplace" % dmap
# else:
# C.__name__ = cls.__name__ + "_inplace%s" % dmap
# return C
elemwise2.py
浏览文件 @ c7db795f
...@@ -4,8 +4,8 @@ import elemwise_cgen as cgen ...@@ -4,8 +4,8 @@ import elemwise_cgen as cgen
import numpy import numpy
from gof import Op, Viewer, Destroyer from gof import Op, Viewer, Destroyer
from base_tensor import BaseTensor as Tensor from base_tensor import BaseTensor as Tensor
import scalar
from scalar import upcast, Scalar from scalar import upcast, Scalar
import scalar_ops
import gof import gof
from gof.python25 import all from gof.python25 import all
...@@ -424,7 +424,7 @@ class CAReduce(Op): ...@@ -424,7 +424,7 @@ class CAReduce(Op):
tosum = self.dimensions_to_reduce tosum = self.dimensions_to_reduce
if tosum == (): if tosum == ():
return Broadcast(scalar_ops.Identity, (input, ))._c_all(inames, onames, sub) return Broadcast(scalar.Identity, (input, ))._c_all(inames, onames, sub)
order1 = [i for i in xrange(len(input.broadcastable)) if i not in tosum] order1 = [i for i in xrange(len(input.broadcastable)) if i not in tosum]
order = order1 + list(tosum) order = order1 + list(tosum)
...@@ -522,7 +522,7 @@ def make_reduce(scalar_opclass, name = None): ...@@ -522,7 +522,7 @@ def make_reduce(scalar_opclass, name = None):
New.__name__ = "Reduce" + scalar_opclass.__name__ New.__name__ = "Reduce" + scalar_opclass.__name__
return New return New
class Sum(make_reduce(scalar_ops.Add)): class Sum(make_reduce(scalar.Add)):
def grad(self, (x, ), (gz, )): def grad(self, (x, ), (gz, )):
if self.dimensions_to_reduce == (): if self.dimensions_to_reduce == ():
return gz, return gz,
...@@ -534,7 +534,7 @@ class Sum(make_reduce(scalar_ops.Add)): ...@@ -534,7 +534,7 @@ class Sum(make_reduce(scalar_ops.Add)):
else: else:
new_dims.append(i) new_dims.append(i)
i += 1 i += 1
return Broadcast(scalar_ops.Second, (x, DimShuffle(gz, new_dims).out)).out, return Broadcast(scalar.Second, (x, DimShuffle(gz, new_dims).out)).out,
def reduce(op): def reduce(op):
......
gof/op.py
浏览文件 @ c7db795f
...@@ -58,6 +58,8 @@ class Op(object): ...@@ -58,6 +58,8 @@ class Op(object):
# #
# Python stdlib compatibility # Python stdlib compatibility
# #
# These are defined so that sets of Ops, Results will have a consistent
# ordering
def __cmp__(self, other): def __cmp__(self, other):
return cmp(id(self), id(other)) return cmp(id(self), id(other))
......
gradient.py
浏览文件 @ c7db795f
...@@ -3,7 +3,7 @@ import numpy #for numeric_grad ...@@ -3,7 +3,7 @@ import numpy #for numeric_grad
from gof.python25 import all from gof.python25 import all
_msg_retNone = 'op.grad(...) returned None, consider returning [None]' _msg_retType = 'op.grad(...) returned a non-list'
_msg_badlen = 'op.grad(...) returned wrong number of gradients' _msg_badlen = 'op.grad(...) returned wrong number of gradients'
def _unpack_result(lst): def _unpack_result(lst):
...@@ -82,10 +82,8 @@ def grad_sources_inputs(sources, graph_inputs): ...@@ -82,10 +82,8 @@ def grad_sources_inputs(sources, graph_inputs):
output_arg = g_outputs output_arg = g_outputs
input_arg = op.inputs input_arg = op.inputs
op_grad = op.grad(input_arg, output_arg) op_grad = op.grad(input_arg, output_arg)
if op_grad is None: if not isinstance(op_grad, (list,tuple)):
raise ValueError(_msg_retNone, op.__class__) raise ValueError(_msg_retType, op.__class__)
if isinstance(op_grad, float):
raise TypeError('wtf!!!!!!!!', op)
g_inputs = op_grad #_pack_result(op_grad) g_inputs = op_grad #_pack_result(op_grad)
assert isinstance(g_inputs, (list, tuple)) assert isinstance(g_inputs, (list, tuple))
if len(g_inputs) != len(op.inputs): if len(g_inputs) != len(op.inputs):
......
scalar.py
浏览文件 @ c7db795f
import numpy import numpy
import math
from copy import copy from copy import copy
import inspect import inspect
...@@ -143,6 +144,35 @@ class Scalar(ResultBase): ...@@ -143,6 +144,35 @@ class Scalar(ResultBase):
cpy.data = self.data cpy.data = self.data
return cpy return cpy
#UNARY
def __abs__(self): return Abs(self).out
def __neg__(self): return Neg(self).out
#CASTS
def __int__(self): return AsInt(self).out
def __float__(self): return AsInt(self).out
def __complex__(self): return AsComplex(self).out
#COMPARISONS
def __lt__(self,other): return lt(self, other)
def __le__(self,other): return le(self, other)
def __gt__(self,other): return gt(self, other)
def __ge__(self,other): return ge(self, other)
#ARITHMETIC - NORMAL
def __add__(self,other): return add(self,other)
def __sub__(self,other): return sub(self,other)
def __mul__(self,other): return mul(self,other)
def __div__(self,other): return div(self,other)
def __pow__(self,other): return pow(self,other)
#ARITHMETIC - RIGHT-OPERAND
def __radd__(self,other): return add(other,self)
def __rsub__(self,other): return sub(other,self)
def __rmul__(self,other): return mul(other,self)
def __rdiv__(self,other): return div(other,self)
def __rpow__(self,other): return pow(other,self)
class ScalarMixedOp(GuardedOp): class ScalarMixedOp(GuardedOp):
...@@ -201,4 +231,154 @@ class BinaryScalarOp(PureScalarOp): ...@@ -201,4 +231,154 @@ class BinaryScalarOp(PureScalarOp):
class Add(BinaryScalarOp):
identity = 0
def impl(self, x, y):
return x + y
def c_code(self, (x, y), (z, ), sub):
return "%(z)s = %(x)s + %(y)s;" % locals()
def grad(self, (x, y), (gz, )):
return gz, gz
class Sub(BinaryScalarOp):
def impl(self, x, y):
return x - y
def c_code(self, (x, y), (z, ), sub):
return "%(z)s = %(x)s - %(y)s;" % locals()
def grad(self, (x, y), (gz, )):
return gz, neg(gz)
class Mul(BinaryScalarOp):
def impl(self, x, y):
return x * y
def c_code(self, (x, y), (z, ), sub):
return "%(z)s = %(x)s * %(y)s;" % locals()
def grad(self, (x, y), (gz, )):
return mul(y, gz), mul(x, gz)
class Div(BinaryScalarOp):
def impl(self, x, y):
return x / y
def c_code(self, (x, y), (z, ), sub):
return "%(z)s = %(x)s / %(y)s;" % locals()
def grad(self, (x, y), (gz, )):
return div(gz, y), neg(div(mul(x, gz), mul(y, y)))
class Pow(BinaryScalarOp):
def impl(self, x, y):
return x ** y
def c_code(self, (x, y), (z, ), sub):
return "%(z)s = pow(%(x)s, %(y)s);" % locals()
def grad(self, (x, y), (gz, )):
return mul(gz, mul(y, pow(x, sub(y, as_scalar(1))))), mul(gz, mul(log(x), pow(x, y)))
class First(BinaryScalarOp):
def impl(self, x, y):
return x
def c_code(self, (x, y), (z, ), sub):
return "%(z)s = %(x)s;" % locals()
def grad(self, (x, y), (gz, )):
return gz, None
class Second(BinaryScalarOp):
def impl(self, x, y):
return y
def c_code(self, (x, y), (z, ), sub):
return "%(z)s = %(y)s;" % locals()
def grad(self, (x, y), (gz, )):
return None, gz
class Identity(UnaryScalarOp):
def impl(self, x):
return x
def c_code(self, (x, ), (z, ), sub):
return "%(z)s = %(x)s;" % locals()
def grad(self, (x, y), (gz, )):
return gz,
class Neg(UnaryScalarOp):
def impl(self, x):
return -x
def grad(self, (x, ), (gz, )):
return neg(gz),
def c_code(self, (x, ), (z, ), sub):
return "%(z)s = -%(x)s;" % locals()
class Abs(UnaryScalarOp):
def impl(self, x):
return numpy.abs(x)
def grad(self, (x, ), (gz, )):
return mul(gz, sgn(x)),
def c_code(self, (x, ), (z, ), sub):
return "%(z)s = abs(%(x)s);" % locals()
class Sgn(UnaryScalarOp):
def impl(self, x):
return numpy.abs(x) / x
def grad(self, (x, ), (gz, )):
return None,
def c_code(self, (x, ), (z, ), sub):
return "%(z)s = %(x)s/abs(%(x)s);" % locals() # TODO: C use copysign
class Inv(UnaryScalarOp):
def impl(self, x):
return 1 / x
def grad(self, (x, ), (gz, )):
return div(neg(gz), mul(x, x)),
def c_code(self, (x, ), (z, ), sub):
return "%(z)s = 1 / %(x)s;" % locals()
class Log(UnaryScalarOp):
def impl(self, x):
return math.log(x)
def grad(self, (x, ), (gz, )):
return div(gz, x),
def c_code(self, (x, ), (z, ), sub):
return "%(z)s = log(%(x)s);" % locals()
class Log2(UnaryScalarOp):
def impl(self, x):
return numpy.log2(x)
def grad(self, (x, ), (gz, )):
return div(gz, mul(x, as_scalar(math.log(2.0)))),
def c_code(self, (x, ), (z, ), sub):
return "%(z)s = log2(%(x)s);" % locals()
class Exp(UnaryScalarOp):
def impl(self, x):
return math.exp(x)
def grad(self, (x, ), (gz, )):
return mul(gz, exp(x)),
def c_code(self, (x, ), (z, ), sub):
return "%(z)s = exp(%(x)s);" % locals()
class Sqr(UnaryScalarOp):
def impl(self, x):
return x*x
def grad(self, (x, ), (gz, )):
return mul(gz, mul(x, as_scalar(2))),
def c_code(self, (x, ), (z, ), sub):
return "%(z)s = %(x)s * %(x)s;" % locals()
class Sqrt(UnaryScalarOp):
def impl(self, x):
return math.sqrt(x)
def grad(self, (x, ), (gz, )):
return div(mul(gz, as_scalar(0.5)), sqrt(x)),
def c_code(self, (x, ), (z, ), sub):
return "%(z)s = sqrt(%(x)s);" % locals()
#NOTE WELL!!!
# The following adds functions to this module automatically.
# For every scalar op class, a lower-case symbol is added which is a constructor
# for that class.
from gof import modes
modes.make_constructors(globals())
scalar_ops.py deleted 100644 → 0
浏览文件 @ f1153866
from scalar import *
import math
class Add(BinaryScalarOp):
identity = 0
def impl(self, x, y):
return x + y
def c_code(self, (x, y), (z, ), sub):
return "%(z)s = %(x)s + %(y)s;" % locals()
def grad(self, (x, y), (gz, )):
return gz, gz
class Sub(BinaryScalarOp):
def impl(self, x, y):
return x - y
def c_code(self, (x, y), (z, ), sub):
return "%(z)s = %(x)s - %(y)s;" % locals()
def grad(self, (x, y), (gz, )):
return gz, neg(gz)
class Mul(BinaryScalarOp):
def impl(self, x, y):
return x * y
def c_code(self, (x, y), (z, ), sub):
return "%(z)s = %(x)s * %(y)s;" % locals()
def grad(self, (x, y), (gz, )):
return mul(y, gz), mul(x, gz)
class Div(BinaryScalarOp):
def impl(self, x, y):
return x / y
def c_code(self, (x, y), (z, ), sub):
return "%(z)s = %(x)s / %(y)s;" % locals()
def grad(self, (x, y), (gz, )):
return div(gz, y), neg(div(mul(x, gz), mul(y, y)))
class Pow(BinaryScalarOp):
def impl(self, x, y):
return x ** y
def c_code(self, (x, y), (z, ), sub):
return "%(z)s = pow(%(x)s, %(y)s);" % locals()
def grad(self, (x, y), (gz, )):
return mul(gz, mul(y, pow(x, sub(y, as_scalar(1))))), mul(gz, mul(log(x), pow(x, y)))
class First(BinaryScalarOp):
def impl(self, x, y):
return x
def c_code(self, (x, y), (z, ), sub):
return "%(z)s = %(x)s;" % locals()
def grad(self, (x, y), (gz, )):
return gz, None
class Second(BinaryScalarOp):
def impl(self, x, y):
return y
def c_code(self, (x, y), (z, ), sub):
return "%(z)s = %(y)s;" % locals()
def grad(self, (x, y), (gz, )):
return None, gz
class Identity(UnaryScalarOp):
def impl(self, x):
return x
def c_code(self, (x, ), (z, ), sub):
return "%(z)s = %(x)s;" % locals()
def grad(self, (x, y), (gz, )):
return gz,
class Neg(UnaryScalarOp):
def impl(self, x):
return -x
def grad(self, (x, ), (gz, )):
return neg(gz),
def c_code(self, (x, ), (z, ), sub):
return "%(z)s = -%(x)s;" % locals()
class Abs(UnaryScalarOp):
def impl(self, x):
return numpy.abs(x)
def grad(self, (x, ), (gz, )):
return mul(gz, sgn(x)),
def c_code(self, (x, ), (z, ), sub):
return "%(z)s = abs(%(x)s);" % locals()
class Sgn(UnaryScalarOp):
def impl(self, x):
return numpy.abs(x) / x
def grad(self, (x, ), (gz, )):
return None,
def c_code(self, (x, ), (z, ), sub):
return "%(z)s = %(x)s/abs(%(x)s);" % locals() # TODO: C use copysign
class Inv(UnaryScalarOp):
def impl(self, x):
return 1 / x
def grad(self, (x, ), (gz, )):
return div(neg(gz), mul(x, x)),
def c_code(self, (x, ), (z, ), sub):
return "%(z)s = 1 / %(x)s;" % locals()
class Log(UnaryScalarOp):
def impl(self, x):
return math.log(x)
def grad(self, (x, ), (gz, )):
return div(gz, x),
def c_code(self, (x, ), (z, ), sub):
return "%(z)s = log(%(x)s);" % locals()
class Log2(UnaryScalarOp):
def impl(self, x):
return numpy.log2(x)
def grad(self, (x, ), (gz, )):
return div(gz, mul(x, as_scalar(math.log(2.0)))),
def c_code(self, (x, ), (z, ), sub):
return "%(z)s = log2(%(x)s);" % locals()
class Exp(UnaryScalarOp):
def impl(self, x):
return math.exp(x)
def grad(self, (x, ), (gz, )):
return mul(gz, exp(x)),
def c_code(self, (x, ), (z, ), sub):
return "%(z)s = exp(%(x)s);" % locals()
class Sqr(UnaryScalarOp):
def impl(self, x):
return x*x
def grad(self, (x, ), (gz, )):
return mul(gz, mul(x, as_scalar(2))),
def c_code(self, (x, ), (z, ), sub):
return "%(z)s = %(x)s * %(x)s;" % locals()
class Sqrt(UnaryScalarOp):
def impl(self, x):
return math.sqrt(x)
def grad(self, (x, ), (gz, )):
return div(mul(gz, as_scalar(0.5)), sqrt(x)),
def c_code(self, (x, ), (z, ), sub):
return "%(z)s = sqrt(%(x)s);" % locals()
# class Sigmoid(UnaryComposite):
# def expand_impl(self, x):
# return 1.0 / (1.0 + exp(-x))
from gof import modes
modes.make_constructors(globals())
tensor.py
浏览文件 @ c7db795f
...@@ -12,7 +12,7 @@ from base_tensor import BaseTensor, BaseTensorOp ...@@ -12,7 +12,7 @@ from base_tensor import BaseTensor, BaseTensorOp
import blas # for gemm, dot import blas # for gemm, dot
import elemwise2 as s2t import elemwise2 as s2t
import scalar_ops as scal import scalar as scal
class Tensor(BaseTensor): class Tensor(BaseTensor):
......
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