未验证 提交 bbb9b28a 编写于 作者: zhouweiwei2014's avatar zhouweiwei2014 提交者: GitHub

add new API paddle.nn.initializer.Dirac (#37389)

* add new API paddle.nn.initializer.Dirac

* fix doc
上级 e64829e2
...@@ -1039,9 +1039,10 @@ def calculate_gain(nonlinearity, param=None): ...@@ -1039,9 +1039,10 @@ def calculate_gain(nonlinearity, param=None):
Get the recommended gain value of some nonlinearity function. Get the recommended gain value of some nonlinearity function.
Args: Args:
nonlinearity(str): nonlinearity function. nonlinearity(str): name of nonlinearity activation function. If it is a linear function, which is one of
param(bool|int|float, optional): optional parameter for somme nonlinearity function. Now, it only applies to 'leaky_relu'. Default: None, "linear/conv1d/conv2d/conv3d/conv1d_transpose/conv2d_transpose/conv3d_transpose" , will return 1.0
it will be calculated as 0.01 in the formula. param(bool|int|float, optional): optional parameter for somme nonlinearity function. Now, it only applies to
'leaky_relu'. Default: None, it will be calculated as 0.01 in the formula.
Returns: Returns:
The recommended gain value for nonlinearity function. The recommended gain value for nonlinearity function.
...@@ -1065,9 +1066,9 @@ def calculate_gain(nonlinearity, param=None): ...@@ -1065,9 +1066,9 @@ def calculate_gain(nonlinearity, param=None):
'conv1d': 1, 'conv1d': 1,
'conv2d': 1, 'conv2d': 1,
'conv3d': 1, 'conv3d': 1,
'conv_transpose1d': 1, 'conv1d_transpose': 1,
'conv_transpose2d': 1, 'conv2d_transpose': 1,
'conv_transpose3d': 1, 'conv3d_transpose': 1,
'tanh': 5.0 / 3, 'tanh': 5.0 / 3,
'relu': math.sqrt(2.0), 'relu': math.sqrt(2.0),
'leaky_relu': math.sqrt(2.0 / (1 + param**2)), 'leaky_relu': math.sqrt(2.0 / (1 + param**2)),
......
...@@ -915,5 +915,118 @@ class TestOrthogonalInitializer6(TestOrthogonalInitializer4): ...@@ -915,5 +915,118 @@ class TestOrthogonalInitializer6(TestOrthogonalInitializer4):
self.assertTrue(np.allclose(np.matmul(a, a.T), np.eye(36), atol=1.e-6)) self.assertTrue(np.allclose(np.matmul(a, a.T), np.eye(36), atol=1.e-6))
# initialize Conv1D weight
class TestDiracInitializer1(unittest.TestCase):
def config(self):
self.weight_attr = paddle.ParamAttr(
initializer=paddle.nn.initializer.Dirac())
self.dtype = "float64"
self.in_channels = 3
self.out_channels = 2
self.kernel_size = 3
self.input_shape = [8, self.in_channels, 10]
self.conv_layer = paddle.nn.Conv1D
self.num_ops = 8 #fill_constant*2, reshape*2, assign_value*2, scatter, cast
def check_result(self, w_dygraph, w_static, conv_in, conv_out):
self.assertTrue(np.array_equal(w_dygraph, w_static))
self.assertTrue(np.array_equal(conv_out, conv_in[:, 0:2, 1:9]))
def test_dirac(self):
self.config()
paddle.set_default_dtype(self.dtype)
paddle.disable_static()
conv = self.conv_layer(
self.in_channels,
self.out_channels,
self.kernel_size,
weight_attr=self.weight_attr)
weight_dygraph = conv.weight.numpy()
paddle.enable_static()
start_prog = paddle.static.Program()
main_prog = paddle.static.Program()
with paddle.static.program_guard(main_prog, start_prog):
inp = paddle.rand(self.input_shape)
conv = self.conv_layer(
self.in_channels,
self.out_channels,
self.kernel_size,
weight_attr=self.weight_attr)
output = conv(inp)
block = start_prog.global_block()
self.assertEqual(len(block.ops), self.num_ops)
self.assertEqual(block.ops[0].type, 'fill_constant')
self.assertEqual(block.ops[1].type, 'reshape')
self.assertEqual(block.ops[2].type, 'assign_value')
self.assertEqual(block.ops[3].type, 'assign_value')
self.assertEqual(block.ops[4].type, 'scatter')
self.assertEqual(block.ops[5].type, 'reshape')
exe = paddle.static.Executor()
exe.run(start_prog)
fetch = exe.run(main_prog, fetch_list=[inp, output, conv.weight])
conv_input = fetch[0]
conv_output = fetch[1]
weight_static = fetch[2]
self.check_result(weight_dygraph, weight_static, conv_input,
conv_output)
# initialize Conv2D weight
class TestDiracInitializer2(TestDiracInitializer1):
def config(self):
self.weight_attr = paddle.ParamAttr(
initializer=paddle.nn.initializer.Dirac(groups=1))
self.dtype = "float64"
self.in_channels = 4
self.out_channels = 8
self.kernel_size = (3, 3)
self.input_shape = [8, self.in_channels, 10, 10]
self.conv_layer = paddle.nn.Conv2D
self.num_ops = 8
def check_result(self, w_dygraph, w_static, conv_in, conv_out):
self.assertTrue(np.array_equal(w_dygraph, w_static))
self.assertTrue(
np.array_equal(conv_out[:, 0:4, :, :], conv_in[:, :, 1:9, 1:9]))
self.assertTrue(
np.array_equal(conv_out[:, 4:8, :, :], np.zeros([8, 4, 8, 8])))
# initialize Conv3D weight
class TestDiracInitializer3(TestDiracInitializer1):
def config(self):
self.weight_attr = paddle.ParamAttr(
initializer=paddle.nn.initializer.Dirac(groups=2))
self.dtype = "float32"
self.in_channels = 5
self.out_channels = 10
self.kernel_size = (3, 3, 3)
self.input_shape = [8, self.in_channels, 10, 10, 10]
self.conv_layer = paddle.nn.Conv3D
self.num_ops = 7
def check_result(self, w_dygraph, w_static, conv_in, conv_out):
self.assertTrue(np.array_equal(w_dygraph, w_static))
self.assertTrue(
np.array_equal(conv_out[:, 0:5, :, :, :], conv_in[:, :, 1:9, 1:9, 1:
9]))
self.assertTrue(
np.array_equal(conv_out[:, 5:10, :, :, :], conv_in[:, :, 1:9, 1:9,
1:9]))
def test_error(self):
self.config()
with self.assertRaises(AssertionError):
paddle.nn.Linear(10, 10, weight_attr=self.weight_attr)
with self.assertRaises(AssertionError):
paddle.nn.Conv2D(5, 9, (3, 3), weight_attr=self.weight_attr)
if __name__ == '__main__': if __name__ == '__main__':
unittest.main() unittest.main()
...@@ -34,6 +34,8 @@ from .uniform import Uniform # noqa: F401 ...@@ -34,6 +34,8 @@ from .uniform import Uniform # noqa: F401
from .orthogonal import Orthogonal # noqa: F401 from .orthogonal import Orthogonal # noqa: F401
from .dirac import Dirac # noqa: F401
__all__ = [ #noqa __all__ = [ #noqa
'Bilinear', 'Bilinear',
'Constant', 'Constant',
...@@ -46,6 +48,7 @@ __all__ = [ #noqa ...@@ -46,6 +48,7 @@ __all__ = [ #noqa
'TruncatedNormal', 'TruncatedNormal',
'Uniform', 'Uniform',
'Orthogonal', 'Orthogonal',
'Dirac',
'set_global_initializer', 'set_global_initializer',
'calculate_gain' 'calculate_gain'
] ]
# Copyright (c) 2021 PaddlePaddle Authors. All Rights Reserved.
#
# Licensed under the Apache License, Version 2.0 (the "License");
# you may not use this file except in compliance with the License.
# You may obtain a copy of the License at
#
# http://www.apache.org/licenses/LICENSE-2.0
#
# Unless required by applicable law or agreed to in writing, software
# distributed under the License is distributed on an "AS IS" BASIS,
# WITHOUT WARRANTIES OR CONDITIONS OF ANY KIND, either express or implied.
# See the License for the specific language governing permissions and
# limitations under the License.
from ...fluid.initializer import Initializer
from ...fluid.data_feeder import check_variable_and_dtype
from ...fluid.core import VarDesc
from ...fluid import unique_name, framework
__all__ = []
class Dirac(Initializer):
"""Initialize the 3D/4D/5D Tensor with Dirac delta function.
It can reserve the feature of convolution layer input, which means that
as many channels are reserved as possible.
In this initialize method, elements in the middle of convolution kernels will
be set to 1 . The formula can be described as:
$ Assuming: N=min(in\_channels, out\_channels)$
$ X[d, d, shape[2]//2, shape[3]//2, ...]=1, \ d=0,1...N$
Args:
groups(int): 0-dimension of the Tensor will be divided by groups, each group has the same value.
name(str, optional): The default value is None. Normally there is no need for user to set this
property. For more information, please refer to :ref:`api_guide_Name`.
Returns:
Dirac initializer instance objects.
Examples:
.. code-block:: python
import paddle
#1.For kernel_size is uneven number:
attr = paddle.ParamAttr(initializer=paddle.nn.initializer.Dirac())
conv = paddle.nn.Conv1D(3, 2, 3, weight_attr=attr)
conv.weight
# Tensor(shape=[2, 3, 3], dtype=float32, place=CPUPlace, stop_gradient=False,
# [[[0., 1., 0.],
# [0., 0., 0.],
# [0., 0., 0.]],
#
# [[0., 0., 0.],
# [0., 1., 0.],
# [0., 0., 0.]]])
input = paddle.rand([8, 3, 10])
output = conv(input)
output == input[:, 0:2, 1:9]
# output.shape is [8, 2, 8], It means output is almost the same with input, 2 channels are reserved
#2. For kernel_size is even number:
attr = paddle.ParamAttr(initializer=paddle.nn.initializer.Dirac())
conv = paddle.nn.Conv1D(3, 2, 4, weight_attr=attr)
conv.weight
# Tensor(shape=[2, 3, 4], dtype=float32, place=CPUPlace, stop_gradient=False,
# [[[0., 0., 1., 0.],
# [0., 0., 0., 0.],
# [0., 0., 0., 0.]],
#
# [[0., 0., 0., 0.],
# [0., 0., 1., 0.],
# [0., 0., 0., 0.]]])
"""
def __init__(self, groups=1, name=None):
assert groups > 0 and isinstance(
groups, int), " 'groups' must be a positive integer. "
super(Dirac, self).__init__()
self._groups = groups
def __call__(self, var, block=None):
"""Initialize the input tensor with dirac initializer.
Args:
var(Tensor): Tensor that needs to be initialized.
block(Block, optional): The block in which initialization ops
should be added. Used in static graph only, default None.
Returns:
The most critical OP(scatter) in this initializer, which contains 7~8 ops in total.
"""
block = self._check_block(block)
assert isinstance(var, framework.Parameter)
assert isinstance(block, framework.Block)
check_variable_and_dtype(
var, "Out", ['float16', 'bfloat16', 'float32', 'float64'], 'Dirac')
assert len(var.shape) in [
3, 4, 5
], "Only Tensor with 3/4/5 dimensions can be initialized by Dirac"
assert (var.shape[0] % self._groups
) == 0, "Tensor 0-dimension must be divisible by groups"
if var.dtype != VarDesc.VarType.FP32:
out_var = block.create_var(
name=unique_name.generate(".".join(['dirac', var.name, 'tmp'])),
shape=var.shape,
dtype=VarDesc.VarType.FP32,
type=VarDesc.VarType.LOD_TENSOR,
persistable=False)
else:
out_var = var
block.append_op(
type='fill_constant',
inputs={},
outputs={'Out': out_var},
attrs={
'value': float(0),
'dtype': out_var.dtype,
'shape': out_var.shape,
},
stop_gradient=True)
origin_shape = var.shape
num_per_group = origin_shape[0] // self._groups
min_shape = min(num_per_group, origin_shape[1])
idx_list = []
value_list = []
strides = []
prod = 1
for dim in reversed(origin_shape):
strides.insert(0, prod)
prod *= dim
for i in range(self._groups):
for j in range(min_shape):
value_list.append(1.0)
offset = 0
for (k, stride) in enumerate(strides):
if (k == 0):
offset += (j + i * num_per_group) * stride
elif (k == 1):
offset += j * stride
else:
offset += origin_shape[k] // 2 * stride
idx_list.append(offset)
block.append_op(
type="reshape",
inputs={"X": out_var},
attrs={'shape': [-1]},
outputs={"Out": out_var},
stop_gradient=True)
index_tensor = block.create_var(
name=unique_name.generate('scatter_index'),
persistable=False,
stop_gradient=True)
block.append_op(
type='assign_value',
outputs={'Out': index_tensor},
attrs={
'dtype': VarDesc.VarType.INT64,
'shape': [len(idx_list)],
'int64_values': idx_list
},
stop_gradient=True)
value_tensor = block.create_var(
name=unique_name.generate('scatter_value'),
persistable=False,
stop_gradient=True)
block.append_op(
type='assign_value',
outputs={'Out': value_tensor},
attrs={
'dtype': VarDesc.VarType.FP32,
'shape': [len(value_list)],
'fp32_values': value_list
},
stop_gradient=True)
op = block.append_op(
type="scatter",
inputs={
"X": out_var,
"Ids": index_tensor,
"Updates": value_tensor
},
attrs={'overwrite': True},
outputs={"Out": out_var},
stop_gradient=True)
block.append_op(
type="reshape",
inputs={"X": out_var},
attrs={'shape': origin_shape},
outputs={"Out": out_var},
stop_gradient=True)
if var.dtype != VarDesc.VarType.FP32:
block.append_op(
type="cast",
inputs={"X": out_var},
outputs={"Out": var},
attrs={"in_dtype": out_var.dtype,
"out_dtype": var.dtype},
stop_gradient=True)
if not framework.in_dygraph_mode():
var.op = op
return op
Markdown is supported
0% .
You are about to add 0 people to the discussion. Proceed with caution.
先完成此消息的编辑!
想要评论请 注册