def accumulate_grads(self, grads):
"""Accumulates gradients from other source.
This method just adds given gradient arrays to gradients that this
optimizer holds. It is typically used in data-parallel optimization,
where gradients for different shards are computed in parallel and
aggregated by this method. This method correctly treats multiple GPU
devices.
Args:
grads (Iterable): Iterable of gradient arrays to be accumulated.
.. deprecated:: v1.5
Use the :meth:`chainer.Link.addgrads` method of the target link
instead.
"""
for param, g_src in zip(self.target.params(), grads):
g_dst = param.grad
if isinstance(g_dst, numpy.ndarray):
g_dst += cuda.to_cpu(g_src)
continue
with cuda.get_device(g_dst):
if (isinstance(g_src, cuda.ndarray)
and g_dst.device != g_src.device):
g_dst += cuda.copy(g_src, out_device=g_dst.device)
else:
g_dst += cuda.to_gpu(g_src)
def accumulate_grads(self, grads):
"""Accumulates gradients from other source.
This method just adds given gradient arrays to gradients that this
optimizer holds. It is typically used in data-parallel optimization,
where gradients for different shards are computed in parallel and
aggregated by this method. This method correctly treats multiple GPU
devices.
Args:
grads (Iterable): Iterable of gradient arrays to be accumulated.
.. deprecated:: v1.5
Use the :meth:`chainer.Link.addgrads` method of the target link
instead.
"""
for param, g_src in zip(self.target.params(), grads):
g_dst = param.grad
if isinstance(g_dst, numpy.ndarray):
g_dst += cuda.to_cpu(g_src)
continue
with cuda.get_device(g_dst):
if (isinstance(g_src, cuda.ndarray) and
g_dst.device != g_src.device):
g_dst += cuda.copy(g_src, out_device=g_dst.device)
else:
g_dst += cuda.to_gpu(g_src)
def forward_gpu(self, x):
self.retain_inputs(())
self._in_device = cuda.get_device_from_array(x[0])
if self.out_device == -1:
return cuda.to_cpu(x[0]),
else:
return cuda.copy(x[0], out_device=self.out_device),
def forward(self, xs):
# self.retain_inputs(())
completed = xs[0]
original = self.x_org
out_ary = cuda.copy(original)
self.slices = []
self._in_shape = original.shape
self._in_dtype = original.dtype
for i in range(0, self.batchsize):
mask_ind_h, mask_ind_w = self.mask_info['index'][i]
mask_h, mask_w = self.mask_info['size'][i]
# マスクされた部分をスライス表記して保存しておく
self.slices.append(
(slice(0, 3), slice(mask_ind_h, mask_ind_h + mask_h),
slice(mask_ind_w, mask_ind_w + mask_w)))
# 元画像のマスクされる領域のみを補完器の出力に置き換える
# <--> 補完器の出力のマスクされた領域以外を元の画像で置きかえる
out_ary[i][self.slices[i]] = completed[i][self.slices[i]]
return out_ary,
def forward_gpu(self, x):
self.retain_inputs(())
self._in_device = cuda.get_device_from_array(x[0])
if self.out_device == -1:
return cuda.to_cpu(x[0]),
else:
return cuda.copy(x[0], out_device=self.out_device),
def copy_parameters_from(self, params):
"""Copies parameters from another source without reallocation.
Args:
params (Iterable): Iterable of parameter arrays.
"""
for dst, src in zip(self.parameters, params):
if isinstance(dst, numpy.ndarray):
if isinstance(src, numpy.ndarray):
numpy.copyto(dst, src)
else:
dst[:] = src.get()
elif isinstance(src, numpy.ndarray):
dst.set(src)
else:
cuda.copy(src, out=dst)
def copy_parameters_from(self, params):
"""Copies parameters from another source without reallocation.
Args:
params (Iterable): Iterable of parameter arrays.
"""
for dst, src in zip(self.parameters, params):
if isinstance(dst, numpy.ndarray):
if isinstance(src, numpy.ndarray):
numpy.copyto(dst, src)
else:
dst[:] = src.get()
elif isinstance(src, numpy.ndarray):
dst.set(src)
else:
cuda.copy(src, out=dst)
def to_gpu(self, device=None):
"""Migrates the function to GPU and returns self.
The default implementation moves all fields of type
:class:`~numpy.ndarray` onto GPU.
Args:
device (int or :class:`pycuda.driver.Device` or ``None``): Device
ID of GPU that the function will be migrated on. If this is
``None``, the current device is used.
Returns:
self.
"""
with cuda.using_device(device):
for k, v in six.iteritems(self.__dict__):
if isinstance(v, numpy.ndarray):
setattr(self, k, cuda.to_gpu(v))
elif (isinstance(v, cuda.GPUArray) and
v.gpudata.device != device):
setattr(self, k, cuda.copy(v, out_device=device))
return self
def to_gpu(self, device=None):
"""Migrates the function to GPU and returns self.
The default implementation moves all fields of type
:class:`~numpy.ndarray` onto GPU.
Args:
device (int or :class:`pycuda.driver.Device` or ``None``): Device
ID of GPU that the function will be migrated on. If this is
``None``, the current device is used.
Returns:
self.
"""
with cuda.using_device(device):
for k, v in six.iteritems(self.__dict__):
if isinstance(v, numpy.ndarray):
setattr(self, k, cuda.to_gpu(v))
elif (isinstance(v, cuda.GPUArray)
and v.gpudata.device != device):
setattr(self, k, cuda.copy(v, out_device=device))
return self
def accumulate_grads(self, grads):
"""Accumulates gradients from other source.
This method just adds given gradient arrays to gradients that this
optimizer holds. It is typically used in data-parallel optimization,
where gradients for different shards are computed in parallel and
aggregated by this method. This method correctly treats multiple GPU
devices.
Args:
grads (Iterable): Iterable of gradient arrays to be accumulated.
"""
for (_, g_dst, _), g_src in zip(self.tuples, grads):
if isinstance(g_dst, numpy.ndarray):
g_dst += cuda.to_cpu(g_src)
continue
with cuda.get_device(g_dst):
if (isinstance(g_src, cuda.ndarray)
and g_dst.gpudata.device != g_src.gpudata.device):
g_dst += cuda.copy(g_src, out_device=g_dst.gpudata.device)
else:
g_dst += cuda.to_gpu(g_src)
def accumulate_grads(self, grads):
"""Accumulates gradients from other source.
This method just adds given gradient arrays to gradients that this
optimizer holds. It is typically used in data-parallel optimization,
where gradients for different shards are computed in parallel and
aggregated by this method. This method correctly treats multiple GPU
devices.
Args:
grads (Iterable): Iterable of gradient arrays to be accumulated.
"""
for (_, g_dst, _), g_src in zip(self.tuples, grads):
if isinstance(g_dst, numpy.ndarray):
g_dst += cuda.to_cpu(g_src)
continue
with cuda.using_device(g_dst):
if (isinstance(g_src, cuda.GPUArray) and
g_dst.gpudata.device != g_src.gpudata.device):
g_dst += cuda.copy(g_src, out_device=g_src.gpudata.device)
else:
g_dst += cuda.to_gpu(g_src)
def forward_gpu(self, x):
return cuda.copy(x[0], out_device=self.out_device),
def backward_gpu(self, x, gy):
if self.out_device == -1:
return cuda.to_gpu(gy[0], device=self._in_device),
else:
return cuda.copy(gy[0], out_device=self._in_device),
def backward_gpu(self, x, gy):
return cuda.copy(gy[0], out_device=cuda.get_device(x[0])),
def forward_gpu(self, x):
if self.out_device == -1:
return cuda.to_cpu(x[0]),
else:
return cuda.copy(x[0], out_device=self.out_device),
def forward_gpu(self, x):
return cuda.copy(x[0], out_device=self.out_device),
def backward_gpu(self, x, gy):
if self.out_device == -1:
return cuda.to_gpu(gy[0], device=cuda.get_device(x[0])),
else:
return cuda.copy(gy[0], out_device=cuda.get_device(x[0])),
def forward_gpu(self, x):
if self.out_device == -1:
return cuda.to_cpu(x[0]),
else:
return cuda.copy(x[0], out_device=self.out_device),
def backward_gpu(self, x, gy):
return cuda.copy(gy[0], out_device=cuda.get_device(x[0])),
def backward_gpu(self, x, gy):
if self.out_device == -1:
return cuda.to_gpu(gy[0], device=self._in_device),
else:
return cuda.copy(gy[0], out_device=self._in_device),
def backward_gpu(self, x, gy):
if self.out_device == -1:
return cuda.to_gpu(gy[0], device=cuda.get_device(x[0])),
else:
return cuda.copy(gy[0], out_device=cuda.get_device(x[0])),
