forked from mindspore-Ecosystem/mindspore
!40593 delete drop parameter type
Merge pull request !40593 from changzherui/del_drop_type
This commit is contained in:
Gitee
commit
c443ef283a
|
|
@ -11,6 +11,7 @@ mindspore.nn.Dropout
|
|||
|
||||
.. note::
|
||||
训练过程中每步对同一通道（或神经元）独立进行丢弃。
|
||||
`dtype` 参数会在未来版本删除，不建议使用这个参数。
|
||||
|
||||
参数：
|
||||
- **keep_prob** (float) - 输入神经元保留率，数值范围在0到1之间。例如，rate=0.9，删除10%的神经元。默认值：0.5。
|
||||
|
|
|
|||
|
|
@ -20,7 +20,7 @@ import math
|
|||
import numpy as np
|
||||
|
||||
import mindspore.common.dtype as mstype
|
||||
from mindspore import context
|
||||
from mindspore import context, log as logger
|
||||
from mindspore.ops.composite.multitype_ops import _constexpr_utils as const_utils
|
||||
from mindspore.common.seed import _get_graph_seed
|
||||
from mindspore.common.tensor import Tensor
|
||||
|
|
@ -119,6 +119,7 @@ class Dropout(Cell):
|
|||
|
||||
Note:
|
||||
Each channel will be zeroed out independently on every construct call.
|
||||
Parameter `dtype` will be removed in a future version. It is not recommended to define this parameter.
|
||||
|
||||
Args:
|
||||
keep_prob (float): The keep rate, greater than 0 and less equal than 1. E.g. rate=0.9,
|
||||
|
|
@ -159,6 +160,8 @@ class Dropout(Cell):
|
|||
raise ValueError(f"For '{self.cls_name}', the 'keep_prob' must be a number in range (0, 1], "
|
||||
f"but got {keep_prob}.")
|
||||
Validator.check_subclass("dtype", dtype, mstype.number_type, self.cls_name)
|
||||
if dtype != mstype.float32:
|
||||
logger.info("This parameter `dtype` will be deleted or invisible in the future. Please don't use it.")
|
||||
self.keep_prob = keep_prob
|
||||
seed0, seed1 = _get_graph_seed(0, "dropout")
|
||||
self.seed0 = seed0
|
||||
|
|
|
|||
|
|
@ -59,10 +59,10 @@ class _Conv(Cell):
|
|||
self.pad_mode = pad_mode
|
||||
self.weight_init = weight_init
|
||||
self.bias_init = bias_init
|
||||
self.format = Validator.check_string(data_format, ['NCHW', 'NHWC', 'NCDHW'], 'format', self.cls_name)
|
||||
if context.get_context("device_target") != "GPU" and self.format == "NHWC":
|
||||
self.data_format = Validator.check_string(data_format, ['NCHW', 'NHWC', 'NCDHW'], 'format', self.cls_name)
|
||||
if context.get_context("device_target") != "GPU" and self.data_format == "NHWC":
|
||||
raise ValueError(f"For '{self.cls_name}', the \"NHWC\" format only support in GPU target, "
|
||||
f"but got the 'format' is {self.format} and "
|
||||
f"but got the 'format' is {self.data_format} and "
|
||||
f"the platform is {context.get_context('device_target')}.")
|
||||
if isinstance(padding, int):
|
||||
Validator.check_non_negative_int(padding, 'padding', self.cls_name)
|
||||
|
|
@ -93,7 +93,7 @@ class _Conv(Cell):
|
|||
if transposed:
|
||||
shape = [in_channels, out_channels // group, *kernel_size]
|
||||
else:
|
||||
shape = [out_channels, *kernel_size, in_channels // group] if self.format == "NHWC" else \
|
||||
shape = [out_channels, *kernel_size, in_channels // group] if self.data_format == "NHWC" else \
|
||||
[out_channels, in_channels // group, *kernel_size]
|
||||
self.weight = Parameter(initializer(self.weight_init, shape), name='weight')
|
||||
|
||||
|
|
@ -108,6 +108,25 @@ class _Conv(Cell):
|
|||
"""Must be overridden by all subclasses."""
|
||||
raise NotImplementedError
|
||||
|
||||
def extend_repr(self):
|
||||
s = 'input_channels={}, output_channels={}, kernel_size={}, ' \
|
||||
'stride={}, pad_mode={}, padding={}, dilation={}, ' \
|
||||
'group={}, has_bias={}, ' \
|
||||
'weight_init={}, bias_init={}, format={}'.format(
|
||||
self.in_channels,
|
||||
self.out_channels,
|
||||
self.kernel_size,
|
||||
self.stride,
|
||||
self.pad_mode,
|
||||
self.padding,
|
||||
self.dilation,
|
||||
self.group,
|
||||
self.has_bias,
|
||||
self.weight_init,
|
||||
self.bias_init,
|
||||
self.data_format)
|
||||
return s
|
||||
|
||||
|
||||
class Conv2d(_Conv):
|
||||
r"""
|
||||
|
|
@ -283,8 +302,8 @@ class Conv2d(_Conv):
|
|||
stride=self.stride,
|
||||
dilation=self.dilation,
|
||||
group=self.group,
|
||||
data_format=self.format)
|
||||
self.bias_add = P.BiasAdd(data_format=self.format)
|
||||
data_format=self.data_format)
|
||||
self.bias_add = P.BiasAdd(data_format=self.data_format)
|
||||
|
||||
def construct(self, x):
|
||||
output = self.conv2d(x, self.weight)
|
||||
|
|
@ -292,25 +311,6 @@ class Conv2d(_Conv):
|
|||
output = self.bias_add(output, self.bias)
|
||||
return output
|
||||
|
||||
def extend_repr(self):
|
||||
s = 'input_channels={}, output_channels={}, kernel_size={}, ' \
|
||||
'stride={}, pad_mode={}, padding={}, dilation={}, ' \
|
||||
'group={}, has_bias={}, ' \
|
||||
'weight_init={}, bias_init={}, format={}'.format(
|
||||
self.in_channels,
|
||||
self.out_channels,
|
||||
self.kernel_size,
|
||||
self.stride,
|
||||
self.pad_mode,
|
||||
self.padding,
|
||||
self.dilation,
|
||||
self.group,
|
||||
self.has_bias,
|
||||
self.weight_init,
|
||||
self.bias_init,
|
||||
self.format)
|
||||
return s
|
||||
|
||||
|
||||
@constexpr
|
||||
def _check_input_3d(input_shape, op_name):
|
||||
|
|
@ -490,24 +490,6 @@ class Conv1d(_Conv):
|
|||
output = self.squeeze(output)
|
||||
return output
|
||||
|
||||
def extend_repr(self):
|
||||
s = 'input_channels={}, output_channels={}, kernel_size={}, ' \
|
||||
'stride={}, pad_mode={}, padding={}, dilation={}, ' \
|
||||
'group={}, has_bias={}, ' \
|
||||
'weight_init={}, bias_init={}'.format(
|
||||
self.in_channels,
|
||||
self.out_channels,
|
||||
self.kernel_size,
|
||||
self.stride,
|
||||
self.pad_mode,
|
||||
self.padding,
|
||||
self.dilation,
|
||||
self.group,
|
||||
self.has_bias,
|
||||
self.weight_init,
|
||||
self.bias_init)
|
||||
return s
|
||||
|
||||
|
||||
@constexpr
|
||||
def _check_input_5dims(input_shape, op_name):
|
||||
|
|
@ -696,8 +678,8 @@ class Conv3d(_Conv):
|
|||
stride=self.stride,
|
||||
dilation=self.dilation,
|
||||
group=self.group,
|
||||
data_format=self.format)
|
||||
self.bias_add = P.BiasAdd(data_format=self.format)
|
||||
data_format=self.data_format)
|
||||
self.bias_add = P.BiasAdd(data_format=self.data_format)
|
||||
self.shape = P.Shape()
|
||||
|
||||
def construct(self, x):
|
||||
|
|
@ -708,25 +690,6 @@ class Conv3d(_Conv):
|
|||
output = self.bias_add(output, self.bias)
|
||||
return output
|
||||
|
||||
def extend_repr(self):
|
||||
s = 'input_channels={}, output_channels={}, kernel_size={}, ' \
|
||||
'stride={}, pad_mode={}, padding={}, dilation={}, ' \
|
||||
'group={}, has_bias={}, ' \
|
||||
'weight_init={}, bias_init={}, format={}'.format(
|
||||
self.in_channels,
|
||||
self.out_channels,
|
||||
self.kernel_size,
|
||||
self.stride,
|
||||
self.pad_mode,
|
||||
self.padding,
|
||||
self.dilation,
|
||||
self.group,
|
||||
self.has_bias,
|
||||
self.weight_init,
|
||||
self.bias_init,
|
||||
self.format)
|
||||
return s
|
||||
|
||||
|
||||
class Conv3dTranspose(_Conv):
|
||||
r"""
|
||||
|
|
@ -910,8 +873,8 @@ class Conv3dTranspose(_Conv):
|
|||
dilation=self.dilation,
|
||||
group=self.group,
|
||||
output_padding=self.output_padding,
|
||||
data_format=self.format)
|
||||
self.bias_add = P.BiasAdd(data_format=self.format)
|
||||
data_format=self.data_format)
|
||||
self.bias_add = P.BiasAdd(data_format=self.data_format)
|
||||
self.shape = P.Shape()
|
||||
|
||||
def construct(self, x):
|
||||
|
|
@ -922,23 +885,6 @@ class Conv3dTranspose(_Conv):
|
|||
output = self.bias_add(output, self.bias)
|
||||
return output
|
||||
|
||||
def extend_repr(self):
|
||||
s = 'input_channels={}, output_channels={}, kernel_size={}, ' \
|
||||
'stride={}, pad_mode={}, padding={}, dilation={}, ' \
|
||||
'group={}, has_bias={}, ' \
|
||||
'weight_init={}, bias_init={}'.format(self.in_channels,
|
||||
self.out_channels,
|
||||
self.kernel_size,
|
||||
self.stride,
|
||||
self.pad_mode,
|
||||
self.padding,
|
||||
self.dilation,
|
||||
self.group,
|
||||
self.has_bias,
|
||||
self.weight_init,
|
||||
self.bias_init)
|
||||
return s
|
||||
|
||||
|
||||
def _deconv_output_length(is_valid, is_same, is_pad, input_length, filter_size, stride_size, dilation_size, padding):
|
||||
"""Calculate the width and height of output."""
|
||||
|
|
@ -1149,23 +1095,6 @@ class Conv2dTranspose(_Conv):
|
|||
self.bias)
|
||||
return self.conv2d_transpose(x, self.weight, (n, self.out_channels, h_out, w_out))
|
||||
|
||||
def extend_repr(self):
|
||||
s = 'input_channels={}, output_channels={}, kernel_size={}, ' \
|
||||
'stride={}, pad_mode={}, padding={}, dilation={}, ' \
|
||||
'group={}, has_bias={}, ' \
|
||||
'weight_init={}, bias_init={}'.format(self.in_channels,
|
||||
self.out_channels,
|
||||
self.kernel_size,
|
||||
self.stride,
|
||||
self.pad_mode,
|
||||
self.padding,
|
||||
self.dilation,
|
||||
self.group,
|
||||
self.has_bias,
|
||||
self.weight_init,
|
||||
self.bias_init)
|
||||
return s
|
||||
|
||||
|
||||
class Conv1dTranspose(_Conv):
|
||||
r"""
|
||||
|
|
@ -1351,20 +1280,3 @@ class Conv1dTranspose(_Conv):
|
|||
|
||||
output = self.squeeze(output)
|
||||
return output
|
||||
|
||||
def extend_repr(self):
|
||||
s = 'input_channels={}, output_channels={}, kernel_size={}, ' \
|
||||
'stride={}, pad_mode={}, padding={}, dilation={}, ' \
|
||||
'group={}, has_bias={}, ' \
|
||||
'weight_init={}, bias_init={}'.format(self.in_channels,
|
||||
self.out_channels,
|
||||
self.kernel_size,
|
||||
self.stride,
|
||||
self.pad_mode,
|
||||
self.padding,
|
||||
self.dilation,
|
||||
self.group,
|
||||
self.has_bias,
|
||||
self.weight_init,
|
||||
self.bias_init)
|
||||
return s
|
||||
|
|
|
|||
|
|
@ -3197,6 +3197,10 @@ class SGD(PrimitiveWithCheck):
|
|||
initialization and momentum in deep learning <http://proceedings.mlr.press/v28/sutskever13.html>`_.
|
||||
|
||||
Note:
|
||||
If parameters are not grouped, the `weight_decay` in optimizer will be applied on the network parameters without
|
||||
'beta' or 'gamma' in their names. Users can group parameters to change the strategy of decaying weight. When
|
||||
parameters are grouped, each group can set `weight_decay`. If not, the `weight_decay` in optimizer will be
|
||||
applied.
|
||||
For more details, please refer to :class:`mindspore.nn.SGD`.
|
||||
|
||||
Args:
|
||||
|
|
|
|||
Loading…
Reference in New Issue