def forward(self, inputs):
x, t = inputs
if chainer.is_debug():
if not ((0 <= t).all() and
(t < x.shape[1]).all()):
msg = 'Each label `t` need to satisfty `0 <= t < x.shape[1]`'
raise ValueError(msg)
xp = cuda.get_array_module(x)
if xp is numpy:
# This code is equivalent to `t.choose(x.T)`, but `numpy.choose`
# does not work when `x.shape[1] > 32`.
return x[six.moves.range(t.size), t],
else:
y = cuda.elementwise(
'S t, raw T x',
'T y',
'int ind[] = {i, t}; y = x[ind];',
'getitem_fwd'
)(t, x)
return y,
python类is_debug()的实例源码
def forward(self, inputs):
x, W = inputs
if chainer.is_debug():
if not ((0 <= x).all() and
(x < len(W)).all()):
msg = 'Each `x` value need to satisfty `0 <= x < len(W)`'
raise ValueError(msg)
if self.ignore_label is not None:
xp = cuda.get_array_module(*inputs)
mask = (x == self.ignore_label)
return xp.where(
mask[..., None], 0, W.take(xp.where(mask, 0, x), axis=0)),
return W.take(x, axis=0),
def forward_cpu(self, inputs):
x, t = inputs
if chainer.is_debug():
self._check_input_values(x, t)
log_y = softmax_log(x, False)
if self.cache_score:
self.y = numpy.exp(log_y)
log_yd = numpy.rollaxis(log_y, 1)
log_yd = log_yd.reshape(len(log_yd), -1)
log_p = log_yd[numpy.maximum(t.ravel(), 0), six.moves.range(t.size)]
# deal with the case where the SoftmaxCrossEntropy is
# unpickled from the old version
if getattr(self, 'normalize', True):
count = (t != self.ignore_label).sum()
else:
count = len(x)
self._coeff = 1.0 / max(count, 1)
y = (log_p * (t.ravel() != self.ignore_label)).sum(keepdims=True) \
* (-self._coeff)
return y.reshape(()),
def forward_gpu(self, inputs):
cupy = cuda.cupy
x, t = inputs
if chainer.is_debug():
self._check_input_values(x, t)
log_y = softmax_log(x, self.use_cudnn)
if self.cache_score:
self.y = cupy.exp(log_y)
if getattr(self, 'normalize', True):
coeff = cupy.maximum(1, (t != self.ignore_label).sum())
else:
coeff = max(1, len(t))
self._coeff = cupy.divide(1.0, coeff, dtype=x.dtype)
log_y = cupy.rollaxis(log_y, 1, log_y.ndim)
ret = cuda.reduce(
'S t, raw T log_y, int32 n_channel, raw T coeff', 'T out',
't == -1 ? T(0) : log_y[_j * n_channel + t]',
'a + b', 'out = a * -coeff[0]', '0', 'crossent_fwd'
)(t, log_y.reduced_view(), log_y.shape[-1], self._coeff)
return ret,
def forward_cpu(self, inputs):
x, t = inputs
if chainer.is_debug():
self._check_input_values(x, t)
log_y = numpy.log(x)
if self.cache_score:
self.y = x
log_yd = numpy.rollaxis(log_y, 1)
log_yd = log_yd.reshape(len(log_yd), -1)
log_p = log_yd[numpy.maximum(t.ravel(), 0), six.moves.range(t.size)]
if getattr(self, 'normalize', True):
count = (t != self.ignore_label).sum()
else:
count = len(x)
self._coeff = 1.0 / max(count, 1)
y = (log_p * (t.ravel() != self.ignore_label)).sum(keepdims=True) * (-self._coeff)
return y.reshape(()),
def forward_gpu(self, inputs):
cupy = cuda.cupy
x, t = inputs
if chainer.is_debug():
self._check_input_values(x, t)
log_y = cupy.log(x)
if self.cache_score:
self.y = x
if getattr(self, 'normalize', True):
coeff = cupy.maximum(1, (t != self.ignore_label).sum())
else:
coeff = max(1, len(t))
self._coeff = cupy.divide(1.0, coeff, dtype=x.dtype)
log_y = cupy.rollaxis(log_y, 1, log_y.ndim)
ret = cuda.reduce(
'S t, raw T log_y, int32 n_channel, raw T coeff', 'T out',
't == -1 ? 0 : log_y[_j * n_channel + t]',
'a + b', 'out = a * -coeff[0]', '0', 'crossent_fwd'
)(t, log_y.reduced_view(), log_y.shape[-1], self._coeff)
return ret,
def forward_cpu(self, inputs):
x, t, w = inputs
if chainer.is_debug():
self._check_input_values(x, t)
log_y = softmax_log(x, False)
if self.cache_score:
self.y = numpy.exp(log_y)
log_yd = numpy.rollaxis(log_y, 1)
log_yd = log_yd.reshape(len(log_yd), -1)
log_p = log_yd[numpy.maximum(t.ravel(), 0), six.moves.range(t.size)]
# deal with the case where the SoftmaxCrossEntropy is
# unpickled from the old version
if getattr(self, 'normalize', True):
count = (t != self.ignore_label).sum()
else:
count = len(x)
self._coeff = 1.0 / max(count, 1)
y = (log_p * (t.ravel() != self.ignore_label) * w.ravel()).sum(keepdims=True) \
* (-self._coeff)
return y.reshape(()),
def forward_gpu(self, inputs):
cupy = cuda.cupy
x, t, w = inputs
if chainer.is_debug():
self._check_input_values(x, t)
log_y = softmax_log(x, self.use_cudnn)
if self.cache_score:
self.y = cupy.exp(log_y)
if getattr(self, 'normalize', True):
coeff = cupy.maximum(1, (t != self.ignore_label).sum())
else:
coeff = max(1, len(t))
self._coeff = cupy.divide(1.0, coeff, dtype=x.dtype)
log_y = cupy.rollaxis(log_y, 1, log_y.ndim)
ret = cuda.reduce(
'S t, T w, raw T log_y, int32 n_channel, raw T coeff', 'T out',
't == -1 ? T(0) : log_y[_j * n_channel + t] * w',
'a + b', 'out = a * -coeff[0]', '0', 'crossent_fwd'
)(t, w, log_y.reduced_view(), log_y.shape[-1], self._coeff)
return ret,
def forward_cpu(self, inputs):
x, t = inputs
if chainer.is_debug():
self._check_input_values(x, t)
log_y = log_softmax._log_softmax(x, self.use_cudnn)
if self.cache_score:
self.y = numpy.exp(log_y)
if self.class_weight is not None:
if self.class_weight.shape != x.shape:
shape = [1 if d != 1 else -1 for d in six.moves.range(x.ndim)]
self.class_weight = numpy.broadcast_to(
self.class_weight.reshape(shape), x.shape)
log_y *= self.class_weight
log_yd = numpy.rollaxis(log_y, 1)
log_yd = log_yd.reshape(len(log_yd), -1)
log_p = log_yd[numpy.maximum(t.ravel(), 0), numpy.arange(t.size)]
# deal with the case where the SoftmaxCrossEntropy is
# unpickled from the old version
if self.normalize:
count = (t != self.ignore_label).sum()
else:
count = len(x)
self._coeff = 1.0 / max(count, 1)
y = (log_p * (t.ravel() != self.ignore_label)).sum(keepdims=True) \
* (-self._coeff)
return y.reshape(()),
def forward_gpu(self, inputs):
cupy = cuda.cupy
x, t = inputs
if chainer.is_debug():
self._check_input_values(x, t)
log_y = log_softmax._log_softmax(x, self.use_cudnn)
if self.cache_score:
self.y = cupy.exp(log_y)
if self.class_weight is not None:
shape = [1 if d != 1 else -1 for d in six.moves.range(x.ndim)]
log_y *= cupy.broadcast_to(
self.class_weight.reshape(shape), x.shape)
if self.normalize:
coeff = cupy.maximum(1, (t != self.ignore_label).sum())
else:
coeff = max(1, len(t))
self._coeff = cupy.divide(1.0, coeff, dtype=x.dtype)
log_y = cupy.rollaxis(log_y, 1, log_y.ndim)
ret = cuda.reduce(
'S t, raw T log_y, int32 n_channel, raw T coeff', 'T out',
't == -1 ? T(0) : log_y[_j * n_channel + t]',
'a + b', 'out = a * -coeff[0]', '0', 'crossent_fwd'
)(t, log_y.reduced_view(), log_y.shape[-1], self._coeff)
return ret,
test_softmax_cross_entropy.py 文件源码
项目：chainer-segnet
作者: pfnet-research
项目源码
文件源码
阅读 22
收藏 0
点赞 0
评论 0
def setUp(self):
self.x = numpy.random.uniform(-1, 1, (2, 2)).astype(numpy.float32)
# `0` is required to avoid NaN
self.t = numpy.array([self.t_value, 0], dtype=numpy.int32)
self.original_debug = chainer.is_debug()
chainer.set_debug(True)
def forward(self, inputs):
xp = cuda.get_array_module(inputs[0])
self.input_length = inputs[0]
length_unigram = inputs[1]
label_unigram = inputs[2]
label_bigram = inputs[3]
length_bigram = length_unigram
xs = inputs[4:]
if chainer.is_debug():
# batch size check.
assert len(xs[0]) == len(label_unigram)
assert len(xs[0]) == len(self.input_length)
assert len(xs[0]) == len(length_unigram)
# length check.
assert len(xs) >= xp.max(self.input_length)
assert len(label_unigram[0]) >= xp.max(length_unigram)
# unit check
assert xs[0].shape[1] > xp.max(label_unigram)
assert xs[0].shape[1] > xp.max(label_bigram)
assert xs[0].shape[1] > self.blank_symbol
self.path_length = length_unigram * 3 + 1
yseq_shape = (len(xs),) + xs[0].shape
self.yseq = _softmax(xp.vstack(xs).reshape(yseq_shape), xp)
log_yseq = _log_matrix(self.yseq, xp, self.zero_padding)
self.path = _label_to_path(label_unigram, label_bigram, self.blank_symbol, xp)
self.prob_trans = _compute_transition_probability(log_yseq, self.input_length,
label_unigram, length_unigram, label_bigram, length_bigram, self.path, self.path_length, xp, self.zero_padding)
loss = -_logsumexp(self.prob_trans[0], xp, axis=1)
if self.reduce == 'mean':
loss = utils.force_array(xp.mean(loss))
return loss,
def forward_cpu(self, inputs):
x, t = inputs
if chainer.is_debug():
self._check_input_values(x, t)
log_y = log_softmax._log_softmax(x)
if self.cache_score:
self.y = np.exp(log_y)
if self.class_weight is not None:
shape = [1 if d != 1 else -1 for d in six.moves.range(x.ndim)]
log_y *= _broadcast_to(self.class_weight.reshape(shape), x.shape)
log_yd = np.rollaxis(log_y, 1)
log_yd = log_yd.reshape(len(log_yd), -1)
log_p = log_yd[np.maximum(t.ravel(), 0), np.arange(t.size)]
log_p *= (t.ravel() != self.ignore_label)
if self.reduce == 'mean':
# deal with the case where the SoftmaxCrossEntropy is
# unpickled from the old version
if self.normalize:
count = (t != self.ignore_label).sum()
else:
count = len(x)
self._coeff = 1.0 / max(count, 1)
y = log_p.sum(keepdims=True) * (-self._coeff)
return y.reshape(()),
else:
return -log_p.reshape(t.shape),
def setUp(self):
self.link = links.EmbedID(2, 2)
self.t = numpy.array([self.t_value], dtype=numpy.int32)
self.original_debug = chainer.is_debug()
chainer.set_debug(True)
def setUp(self):
self.original_debug = chainer.is_debug()
chainer.set_debug(True)
self.one = numpy.array([1], numpy.float32)
self.f = chainer.Function()
def setUp(self):
self.x = numpy.random.uniform(-1, 1, (2, 2)).astype(numpy.float32)
# `0` is required to avoid NaN
self.t = numpy.array([self.t_value, 0], dtype=numpy.int32)
self.original_debug = chainer.is_debug()
chainer.set_debug(True)
def setUp(self):
self.x = numpy.random.uniform(-1, 1, (1, 2)).astype(numpy.float32)
self.t = numpy.array([self.t_value], dtype=numpy.int32)
self.original_debug = chainer.is_debug()
chainer.set_debug(True)
def setUp(self):
self.x = numpy.random.uniform(-1, 1, (2, 2)).astype(numpy.float32)
# `0` is required to avoid NaN
self.t = numpy.array([self.t_value, 0], dtype=numpy.int32)
self.original_debug = chainer.is_debug()
chainer.set_debug(True)
def forward_cpu(self, inputs):
x, t = inputs
if chainer.is_debug():
self._check_input_values(x, t)
log_y = log_softmax._log_softmax(x)
if self.cache_score:
self.y = np.exp(log_y)
if self.class_weight is not None:
shape = [1 if d != 1 else -1 for d in xrange(x.ndim)]
log_y *= _broadcast_to(self.class_weight.reshape(shape), x.shape)
log_yd = np.rollaxis(log_y, 1)
log_yd = log_yd.reshape(len(log_yd), -1)
log_p = log_yd[np.maximum(t.ravel(), 0), np.arange(t.size)]
log_p *= (t.ravel() != self.ignore_label)
if self.reduce == 'mean':
# deal with the case where the SoftmaxCrossEntropy is
# unpickled from the old version
if self.normalize:
count = (t != self.ignore_label).sum()
else:
count = len(x)
self._coeff = 1.0 / max(count, 1)
y = log_p.sum(keepdims=True) * (-self._coeff)
return y.reshape(()),
else:
return -log_p.reshape(t.shape),
def forward_gpu(self, inputs):
cupy = cuda.cupy
x, t = inputs
if chainer.is_debug():
self._check_input_values(x, t)
log_y = log_softmax._log_softmax(x)
if self.cache_score:
self.y = cupy.exp(log_y)
if self.class_weight is not None:
shape = [1 if d != 1 else -1 for d in six.moves.range(x.ndim)]
log_y *= cupy.broadcast_to(
self.class_weight.reshape(shape), x.shape)
if self.normalize:
coeff = cupy.maximum(1, (t != self.ignore_label).sum())
else:
coeff = max(1, len(t))
self._coeff = cupy.divide(1.0, coeff, dtype=x.dtype)
log_y = cupy.rollaxis(log_y, 1, log_y.ndim)
if self.reduce == 'mean':
ret = cuda.reduce(
'S t, raw T log_y, int32 n_channel, raw T coeff, '
'S ignore_label',
'T out',
't == ignore_label ? T(0) : log_y[_j * n_channel + t]',
'a + b', 'out = a * -coeff[0]', '0', 'crossent_fwd'
)(t, log_y.reduced_view(), log_y.shape[-1],
self._coeff, self.ignore_label)
else:
ret = cuda.elementwise(
'S t, raw T log_y, int32 n_channel, T ignore', 'T out',
'''
if (t == ignore) {
out = 0;
} else {
out = -log_y[i * n_channel + t];
}
''',
'softmax_crossent_no_reduce_fwd'
)(t, log_y.reduced_view(), log_y.shape[-1], self.ignore_label)
ret = ret.reshape(t.shape)
return ret,
def forward_gpu(self, inputs):
cupy = cuda.cupy
x, t = inputs
if chainer.is_debug():
self._check_input_values(x, t)
log_y = log_softmax._log_softmax(x)
if self.cache_score:
self.y = cupy.exp(log_y)
if self.class_weight is not None:
shape = [1 if d != 1 else -1 for d in xrange(x.ndim)]
log_y *= cupy.broadcast_to(
self.class_weight.reshape(shape), x.shape)
if self.normalize:
coeff = cupy.maximum(1, (t != self.ignore_label).sum())
else:
coeff = max(1, len(t))
self._coeff = cupy.divide(1.0, coeff, dtype=x.dtype)
log_y = cupy.rollaxis(log_y, 1, log_y.ndim)
if self.reduce == 'mean':
ret = cuda.reduce(
'S t, raw T log_y, int32 n_channel, raw T coeff, '
'S ignore_label',
'T out',
't == ignore_label ? T(0) : log_y[_j * n_channel + t]',
'a + b', 'out = a * -coeff[0]', '0', 'crossent_fwd'
)(t, log_y.reduced_view(), log_y.shape[-1],
self._coeff, self.ignore_label)
else:
ret = cuda.elementwise(
'S t, raw T log_y, int32 n_channel, T ignore', 'T out',
'''
if (t == ignore) {
out = 0;
} else {
out = -log_y[i * n_channel + t];
}
''',
'softmax_crossent_no_reduce_fwd'
)(t, log_y.reduced_view(), log_y.shape[-1], self.ignore_label)
ret = ret.reshape(t.shape)
return ret,
