Skip to content

P
Paddle

PaddlePaddle / Paddle
1 年多 前同步成功

通知 2302
Star 20931
Fork 5422
P
Paddle
未验证 提交 61953b90 编写于 作者: Z zhangkaihuo 提交者： GitHub
浏览文件
操作

[cherry-pick]Fix english documents of sparse api (#47496) 

Fix english documents of sparse api
上级 601626ac
显示空白变更内容
内联 并排
Showing with 558 addition and 534 deletion
python/paddle/sparse/binary.py
浏览文件 @ 61953b90
......@@ -241,11 +241,9 @@ def add(x, y, name=None):
.. code-block:: python
import paddle
from paddle.fluid.framework import _test_eager_guard
paddle.device.set_device("cpu")
with _test_eager_guard():
x = paddle.to_tensor([[0, -1, 0, 2], [0, 0, -3, 0], [4, 5, 0, 0]], 'float32')
y = paddle.to_tensor([[0, 0, 0, -2], [0, 2, -3, 0], [2, 3, 4, 8]], 'float32')
sparse_x = x.to_sparse_csr()
......@@ -268,10 +266,9 @@ def add(x, y, name=None):
inputs = {'x': x, 'y': y}
helper = LayerHelper(op_type)
out = helper.create_sparse_variable_for_type_inference(x.dtype)
helper.append_op(type=op_type,
inputs=inputs,
outputs={'out': out},
attrs={})
helper.append_op(
type=op_type, inputs=inputs, outputs={'out': out}, attrs={}
)
return out
......@@ -298,11 +295,9 @@ def subtract(x, y, name=None):
.. code-block:: python
import paddle
from paddle.fluid.framework import _test_eager_guard
paddle.device.set_device("cpu")
with _test_eager_guard():
x = paddle.to_tensor([[0, -1, 0, 2], [0, 0, -3, 0], [4, 5, 0, 0]], 'float32')
y = paddle.to_tensor([[0, 0, 0, -2], [0, 2, -3, 0], [2, 3, 4, 8]], 'float32')
sparse_x = x.to_sparse_csr()
......@@ -343,11 +338,9 @@ def multiply(x, y, name=None):
.. code-block:: python
import paddle
from paddle.fluid.framework import _test_eager_guard
paddle.device.set_device("cpu")
with _test_eager_guard():
x = paddle.to_tensor([[0, -1, 0, 2], [0, 0, -3, 0], [4, 5, 0, 0]], 'float32')
y = paddle.to_tensor([[0, 0, 0, -2], [0, 2, -3, 0], [2, 3, 4, 8]], 'float32')
sparse_x = x.to_sparse_csr()
......@@ -391,11 +384,9 @@ def divide(x, y, name=None):
.. code-block:: python
import paddle
from paddle.fluid.framework import _test_eager_guard
paddle.device.set_device("cpu")
with _test_eager_guard():
x = paddle.to_tensor([[0, -1, 0, 2], [0, 0, -3, 0], [4, 5, 0, 0]], 'float32')
y = paddle.to_tensor([[0, 0, 0, -2], [0, 2, -3, 0], [2, 3, 4, 8]], 'float32')
sparse_x = x.to_sparse_csr()
......
python/paddle/sparse/creation.py
浏览文件 @ 61953b90
......@@ -17,7 +17,12 @@ from paddle import _C_ops, _legacy_C_ops
from paddle.fluid.framework import core, dygraph_only
from paddle.fluid.framework import _current_expected_place, _get_paddle_place
from paddle.tensor import to_tensor, max
from paddle.fluid.data_feeder import check_variable_and_dtype, check_type, check_dtype, convert_dtype
from paddle.fluid.data_feeder import (
check_variable_and_dtype,
check_type,
check_dtype,
convert_dtype,
)
from paddle import in_dynamic_mode
from paddle.fluid.layer_helper import LayerHelper
......@@ -51,8 +56,8 @@ def _get_place(place):
if place is None:
place = _current_expected_place()
elif not isinstance(
place,
(core.Place, core.CPUPlace, core.CUDAPinnedPlace, core.CUDAPlace)):
place, (core.Place, core.CPUPlace, core.CUDAPinnedPlace, core.CUDAPlace)
):
raise ValueError(
"'place' must be any of paddle.Place, paddle.CPUPlace, paddle.CUDAPinnedPlace, paddle.CUDAPlace"
)
......@@ -66,12 +71,9 @@ def _check_indices_dtype(dtype):
)
def sparse_coo_tensor(indices,
values,
shape=None,
dtype=None,
place=None,
stop_gradient=True):
def sparse_coo_tensor(
indices, values, shape=None, dtype=None, place=None, stop_gradient=True
):
r"""
Constructs a sparse ``paddle.Tensor`` in coordinate format according to the indices
and values of the specified non-zero elements.
......@@ -107,9 +109,7 @@ def sparse_coo_tensor(indices,
.. code-block:: python
import paddle
from paddle.fluid.framework import _test_eager_guard
with _test_eager_guard():
indices = [[0, 1, 2], [1, 2, 0]]
values = [1.0, 2.0, 3.0]
dense_shape = [3, 3]
......@@ -125,10 +125,9 @@ def sparse_coo_tensor(indices,
place = _get_place(place)
if not isinstance(indices, core.eager.Tensor):
indices = to_tensor(indices,
dtype=None,
place=place,
stop_gradient=True)
indices = to_tensor(
indices, dtype=None, place=place, stop_gradient=True
)
if not isinstance(values, core.eager.Tensor):
values = to_tensor(values, dtype, place, stop_gradient)
if len(indices.shape) != 2:
......@@ -141,8 +140,10 @@ def sparse_coo_tensor(indices,
if nnz != values.shape[0]:
raise ValueError(
"the indices and values must have same number of non-zero, but get {} and {}"
.format(nnz, values.shape[0]))
"the indices and values must have same number of non-zero, but get {} and {}".format(
nnz, values.shape[0]
)
)
dense_dim = len(values.shape) - 1
......@@ -162,11 +163,15 @@ def sparse_coo_tensor(indices,
if shape < min_shape:
raise ValueError(
"the minimun shape required is {}, but get {}".format(
min_shape, shape))
min_shape, shape
)
)
if len(shape) != sparse_dim + dense_dim:
raise ValueError(
"the number of dimensions(len(shape) must be sparse_dim({}) + dense_dim({}), but get {}"
.format(sparse_dim, dense_dim, len(shape)))
"the number of dimensions(len(shape) must be sparse_dim({}) + dense_dim({}), but get {}".format(
sparse_dim, dense_dim, len(shape)
)
)
return _C_ops.sparse_sparse_coo_tensor(values, indices, shape)
......@@ -178,22 +183,17 @@ def sparse_coo_tensor(indices,
attrs = {'dense_shape': shape}
helper = LayerHelper(op_type)
out = helper.create_sparse_variable_for_type_inference(dtype)
helper.append_op(type=op_type,
inputs=inputs,
outputs={'out': out},
attrs=attrs)
helper.append_op(
type=op_type, inputs=inputs, outputs={'out': out}, attrs=attrs
)
return out
#TODO: need to support shape is None
# TODO: need to support shape is None
@dygraph_only
def sparse_csr_tensor(crows,
cols,
values,
shape,
dtype=None,
place=None,
stop_gradient=True):
def sparse_csr_tensor(
crows, cols, values, shape, dtype=None, place=None, stop_gradient=True
):
r"""
Constructs a sparse ``paddle.Tensor`` in CSR(Compressed Sparse Row) format according to the
``crows``, ``cols`` and ``values``.
......@@ -233,9 +233,7 @@ def sparse_csr_tensor(crows,
.. code-block:: python
import paddle
from paddle.fluid.framework import _test_eager_guard
with _test_eager_guard():
crows = [0, 2, 3, 5]
cols = [1, 3, 2, 0, 1]
values = [1, 2, 3, 4, 5]
......@@ -262,8 +260,10 @@ def sparse_csr_tensor(crows,
if len(shape) != 2 and len(shape) != 3:
raise ValueError(
"SparseCsrTensor only support 2-D or 3-D matrix. but get shape {}".
format(shape))
"SparseCsrTensor only support 2-D or 3-D matrix. but get shape {}".format(
shape
)
)
rows = shape[len(shape) - 2]
if not crows.place._equals(place):
......@@ -280,26 +280,32 @@ def sparse_csr_tensor(crows,
if len(crows.shape) != 1 or len(cols.shape) != 1 or len(values.shape) != 1:
raise ValueError("The 'crows', 'cols' and 'values' must be 1-D.")
if (len(cols) != len(values)):
if len(cols) != len(values):
raise ValueError("the length of cols must be same as length of values")
if len(shape) == 2:
if crows.shape[0] != rows + 1:
raise ValueError(
"The length({}) of crows must be equal to the rows({})+1 of matrix."
.format(crows.shape[0], rows))
"The length({}) of crows must be equal to the rows({})+1 of matrix.".format(
crows.shape[0], rows
)
)
if crows[0] != 0:
raise ValueError("the 0th value of crows must be 0")
if crows[-1] != values.shape[0]:
raise ValueError(
"the last value of crows must be equal the number of non-zero")
"the last value of crows must be equal the number of non-zero"
)
else:
if crows.shape[0] % (rows + 1) != 0:
raise ValueError(
"The length({}) of crows must be divisible the rows({})+1 of matrix."
.format(crows.shape[0], rows))
"The length({}) of crows must be divisible the rows({})+1 of matrix.".format(
crows.shape[0], rows
)
)
# TODO(zkh2016): check whether the value in crows and cols is legal
return core.eager.sparse_csr_tensor(crows, cols, values, shape,
stop_gradient)
return core.eager.sparse_csr_tensor(
crows, cols, values, shape, stop_gradient
)
python/paddle/sparse/nn/functional/conv.py
浏览文件 @ 61953b90
......@@ -23,7 +23,8 @@ from paddle.nn.functional.conv import _update_padding_nd
from paddle.fluid.layer_helper import LayerHelper
def _conv3d(x,
def _conv3d(
x,
weight,
bias=None,
stride=1,
......@@ -33,40 +34,55 @@ def _conv3d(x,
subm=False,
key=None,
data_format="NDHWC",
name=None):
name=None,
):
assert groups == 1, "Currently, only support groups=1"
dims = 3
# Currently, only support 'NDHWC'
if data_format not in ["NDHWC"]:
raise ValueError("Attr(data_format) should be 'NDHWC'. Received "
"Attr(data_format): {}.".format(data_format))
raise ValueError(
"Attr(data_format) should be 'NDHWC'. Received "
"Attr(data_format): {}.".format(data_format)
)
if len(x.shape) != 5:
raise ValueError(
"Input x should be 5D tensor, but received x with the shape of {}".
format(x.shape))
"Input x should be 5D tensor, but received x with the shape of {}".format(
x.shape
)
)
channel_last = (data_format == "NDHWC")
channel_last = data_format == "NDHWC"
channel_dim = -1 if channel_last else 1
if len(x.shape) != 5:
raise ValueError(
"Input x should be 5D tensor, but received x with the shape of {}".
format(x.shape))
"Input x should be 5D tensor, but received x with the shape of {}".format(
x.shape
)
)
num_channels = x.shape[channel_dim]
if num_channels < 0:
raise ValueError(
"The channel dimension of the input({}) should be defined. "
"Received: {}.".format(x.shape, num_channels))
"Received: {}.".format(x.shape, num_channels)
)
padding, padding_algorithm = _update_padding_nd(padding, channel_last, dims)
stride = convert_to_list(stride, dims, 'stride')
dilation = convert_to_list(dilation, dims, 'dilation')
if in_dynamic_mode():
pre_bias = _C_ops.sparse_conv3d(x, weight, padding, dilation, stride,
groups, subm,
key if key is not None else "")
pre_bias = _C_ops.sparse_conv3d(
x,
weight,
padding,
dilation,
stride,
groups,
subm,
key if key is not None else "",
)
if bias is not None:
return add(pre_bias, bias)
else:
......@@ -79,27 +95,29 @@ def _conv3d(x,
'strides': stride,
'groups': groups,
'subm': subm,
'key': key
'key': key,
}
op_type = 'sparse_conv3d'
helper = LayerHelper(op_type, **locals())
rulebook = helper.create_variable_for_type_inference(dtype='int32',
stop_gradient=True)
counter = helper.create_variable_for_type_inference(dtype='int32',
stop_gradient=True)
rulebook = helper.create_variable_for_type_inference(
dtype='int32', stop_gradient=True
)
counter = helper.create_variable_for_type_inference(
dtype='int32', stop_gradient=True
)
pre_bias = helper.create_sparse_variable_for_type_inference(x.dtype)
outputs = {"out": pre_bias, "rulebook": rulebook, "counter": counter}
helper.append_op(type=op_type,
inputs=inputs,
outputs=outputs,
attrs=attrs)
helper.append_op(
type=op_type, inputs=inputs, outputs=outputs, attrs=attrs
)
if bias is not None:
return add(pre_bias, bias)
else:
return pre_bias
def conv3d(x,
def conv3d(
x,
weight,
bias=None,
stride=1,
......@@ -107,7 +125,8 @@ def conv3d(x,
dilation=1,
groups=1,
data_format="NDHWC",
name=None):
name=None,
):
r"""
The sparse convolution3d functional calculates the output based on the input, filter
......@@ -132,36 +151,17 @@ def conv3d(x,
* :math:`b`: Bias value, a 1-D tensor with shape [M].
* :math:`Out`: Output value, the shape of :math:`Out` and :math:`X` may be different.
Example:
- Input:
Input shape: :math:`(N, D_{in}, H_{in}, W_{in}, C_{in})`
Filter shape: :math:`(D_f, H_f, W_f, C_{in}, C_{out})`
- Output:
Output shape: :math:`(N, D_{out}, H_{out}, W_{out}, C_{out})`
Where
.. math::
D_{out}&= \\frac{(D_{in} + 2 * paddings[0] - (dilations[0] * (D_f - 1) + 1))}{strides[0]} + 1 \\\\
H_{out}&= \\frac{(H_{in} + 2 * paddings[1] - (dilations[1] * (H_f - 1) + 1))}{strides[1]} + 1 \\\\
W_{out}&= \\frac{(W_{in} + 2 * paddings[2] - (dilations[2] * (W_f - 1) + 1))}{strides[2]} + 1
Args:
x (Tensor): The input is 5-D SparseCooTensor with shape [N, D, H, W, C], the data
type of input is float16 or float32 or float64.
weight (Tensor): The convolution kernel, a Tensor with shape [kD, kH, kW, C/g, M],
where M is the number of filters(output channels), g is the number of groups,
kD, kH, kW are the filter's depth, height and width respectively.
bias (Tensor, optional): The bias, a Tensor of shape [M, ], currently, only support bias is None.
stride (int|list|tuple): The stride size. It means the stride in convolution. If stride is a
bias (Tensor, optional): The bias, a Tensor of shape [M].
stride (int|list|tuple, optional): The stride size. It means the stride in convolution. If stride is a
list/tuple, it must contain three integers, (stride_depth, stride_height, stride_width).
Otherwise, stride_depth = stride_height = stride_width = stride. Default: stride = 1.
padding (string|int|list|tuple): The padding size. It means the number of zero-paddings
padding (string|int|list|tuple, optional): The padding size. It means the number of zero-paddings
on both sides for each dimension. If `padding` is a string, either 'VALID' or
'SAME' which is the padding algorithm. If padding size is a tuple or list,
it could be in three forms: `[pad_depth, pad_height, pad_width]` or
......@@ -171,11 +171,11 @@ def conv3d(x,
when `data_format` is `"NDHWC"`, `padding` can be in the form
`[[0,0], [pad_depth_front, pad_depth_back], [pad_height_top, pad_height_bottom], [pad_width_left, pad_width_right], [0,0]]`.
Default: padding = 0.
dilation (int|list|tuple): The dilation size. It means the spacing between the kernel points.
dilation (int|list|tuple, optional): The dilation size. It means the spacing between the kernel points.
If dilation is a list/tuple, it must contain three integers, (dilation_depth, dilation_height,
dilation_width). Otherwise, dilation_depth = dilation_height = dilation_width = dilation.
Default: dilation = 1.
groups (int): The groups number of the Conv3D Layer. According to grouped
groups (int, optional): The groups number of the Conv3D Layer. According to grouped
convolution in Alex Krizhevsky's Deep CNN paper: when group=2,
the first half of the filters is only connected to the first half
of the input channels, while the second half of the filters is only
......@@ -184,7 +184,7 @@ def conv3d(x,
will be consistent with that of the input. An optional string from: `"NCDHW"`, `"NDHWC"`.
The default is `"NDHWC"`. When it is `"NDHWC"`, the data is stored in the order of:
`[batch_size, input_depth, input_height, input_width, input_channels]`.
name(str|None): For detailed information, please refer
name(str, optional): For detailed information, please refer
to :ref:`api_guide_Name`. Usually name is no need to set and
None by default.
......@@ -195,9 +195,7 @@ def conv3d(x,
.. code-block:: python
import paddle
from paddle.fluid.framework import _test_eager_guard
with _test_eager_guard():
indices = [[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 1, 2], [1, 3, 2, 3]]
values = [[1], [2], [3], [4]]
indices = paddle.to_tensor(indices, dtype='int32')
......@@ -209,11 +207,23 @@ def conv3d(x,
print(y.shape)
# (1, 1, 1, 2, 1)
"""
return _conv3d(x, weight, bias, stride, padding, dilation, groups, False,
None, data_format, name)
def subm_conv3d(x,
return _conv3d(
x,
weight,
bias,
stride,
padding,
dilation,
groups,
False,
None,
data_format,
name,
)
def subm_conv3d(
x,
weight,
bias=None,
stride=1,
......@@ -222,7 +232,8 @@ def subm_conv3d(x,
groups=1,
data_format="NDHWC",
key=None,
name=None):
name=None,
):
r"""
The sparse submanifold convolution3d functional calculates the output based on the input, filter
......@@ -247,33 +258,14 @@ def subm_conv3d(x,
* :math:`b`: Bias value, a 1-D tensor with shape [M].
* :math:`Out`: Output value, the shape of :math:`Out` and :math:`X` may be different.
Example:
- Input:
Input shape: :math:`(N, D_{in}, H_{in}, W_{in}, C_{in})`
Filter shape: :math:`(D_f, H_f, W_f, C_{in}, C_{out})`
- Output:
Output shape: :math:`(N, D_{out}, H_{out}, W_{out}, C_{out})`
Where
.. math::
D_{out}&= \\frac{(D_{in} + 2 * paddings[0] - (dilations[0] * (D_f - 1) + 1))}{strides[0]} + 1 \\\\
H_{out}&= \\frac{(H_{in} + 2 * paddings[1] - (dilations[1] * (H_f - 1) + 1))}{strides[1]} + 1 \\\\
W_{out}&= \\frac{(W_{in} + 2 * paddings[2] - (dilations[2] * (W_f - 1) + 1))}{strides[2]} + 1
Args:
x (Tensor): The input is 5-D SparseCooTensor with shape [N, D, H, W, C], the data
type of input is float16 or float32 or float64.
weight (Tensor): The convolution kernel, a Tensor with shape [kD, kH, kW, C/g, M],
where M is the number of filters(output channels), g is the number of groups,
kD, kH, kW are the filter's depth, height and width respectively.
bias (Tensor, optional): The bias, a Tensor of shape [M, ], currently, only support bias is None.
stride (int|list|tuple): The stride size. It means the stride in convolution. If stride is a
bias (Tensor, optional): The bias, a Tensor of shape [M].
stride (int|list|tuple, optional): The stride size. It means the stride in convolution. If stride is a
list/tuple, it must contain three integers, (stride_depth, stride_height, stride_width).
Otherwise, stride_depth = stride_height = stride_width = stride. Default: stride = 1.
padding (string|int|list|tuple): The padding size. It means the number of zero-paddings
......@@ -286,11 +278,11 @@ def subm_conv3d(x,
when `data_format` is `"NHWC"`, `padding` can be in the form
`[[0,0], [pad_depth_front, pad_depth_back], [pad_height_top, pad_height_bottom], [pad_width_left, pad_width_right], [0,0]]`.
Default: padding = 0.
dilation (int|list|tuple): The dilation size. It means the spacing between the kernel points.
dilation (int|list|tuple, optional): The dilation size. It means the spacing between the kernel points.
If dilation is a list/tuple, it must contain three integers, (dilation_depth, dilation_height,
dilation_width). Otherwise, dilation_depth = dilation_height = dilation_width = dilation.
Default: dilation = 1.
groups (int): The groups number of the Conv3D Layer. According to grouped
groups (int, optional): The groups number of the Conv3D Layer. According to grouped
convolution in Alex Krizhevsky's Deep CNN paper: when group=2,
the first half of the filters is only connected to the first half
of the input channels, while the second half of the filters is only
......@@ -303,7 +295,7 @@ def subm_conv3d(x,
the definition and role of rulebook refers to
https://pdfs.semanticscholar.org/5125/a16039cabc6320c908a4764f32596e018ad3.pdf. The
default value is None.
name(str|None): For detailed information, please refer
name(str, optional): For detailed information, please refer
to :ref:`api_guide_Name`. Usually name is no need to set and
None by default.
......@@ -315,9 +307,7 @@ def subm_conv3d(x,
.. code-block:: python
import paddle
from paddle.fluid.framework import _test_eager_guard
with _test_eager_guard():
indices = [[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 1, 2], [1, 3, 2, 3]]
values = [[1], [2], [3], [4]]
indices = paddle.to_tensor(indices, dtype='int32')
......@@ -329,5 +319,16 @@ def subm_conv3d(x,
print(y.shape)
#(1, 1, 3, 4, 1)
"""
return _conv3d(x, weight, bias, stride, padding, dilation, groups, True,
key, data_format, name)
return _conv3d(
x,
weight,
bias,
stride,
padding,
dilation,
groups,
True,
key,
data_format,
name,
)
python/paddle/sparse/nn/functional/pooling.py
浏览文件 @ 61953b90
......@@ -19,13 +19,15 @@ from paddle.nn.functional.pooling import _update_padding_nd
__all__ = []
def max_pool3d(x,
def max_pool3d(
x,
kernel_size,
stride=None,
padding=0,
ceil_mode=False,
data_format="NDHWC",
name=None):
name=None,
):
"""
Implements sparse max pooling 3d operation.
See more details in :ref:`api_sparse_pooling_MaxPool3d` .
......@@ -37,18 +39,18 @@ def max_pool3d(x,
is a tuple or list, it must contain three integers,
(kernel_size_Depth, kernel_size_Height, kernel_size_Width).
Otherwise, the pool kernel size will be the cube of an int.
stride (int|list|tuple): The pool stride size. If pool stride size is a tuple or list,
stride (int|list|tuple, optional): The pool stride size. If pool stride size is a tuple or list,
it must contain three integers, [stride_Depth, stride_Height, stride_Width).
Otherwise, the pool stride size will be a cube of an int.
padding (string|int|list|tuple): The padding size. Padding could be in one of the following forms.
padding (string|int|list|tuple, optional): The padding size. Padding could be in one of the following forms.
1. A string in ['valid', 'same'].
2. An int, which means the feature map is zero padded by size of `padding` on every sides.
3. A list[int] or tuple(int) whose length is 3, [pad_depth, pad_height, pad_weight] whose value means the padding size of each dimension.
4. A list[int] or tuple(int) whose length is 6. [pad_depth_front, pad_depth_back, pad_height_top, pad_height_bottom, pad_width_left, pad_width_right] whose value means the padding size of each side.
5. A list or tuple of pairs of integers. It has the form [[pad_before, pad_after], [pad_before, pad_after], ...]. Note that, the batch dimension and channel dimension should be [0,0] or (0,0).
The default value is 0.
ceil_mode (bool): ${ceil_mode_comment}
data_format (string): The data format of the input and output data. An optional string from: `"NCDHW"`, `"NDHWC"`.
ceil_mode (bool, optional): ${ceil_mode_comment}
data_format (string, optional): The data format of the input and output data. An optional string from: `"NCDHW"`, `"NDHWC"`.
The default is `"NCDHW"`. When it is `"NCDHW"`, the data is stored in the order of:
`[batch_size, input_channels, input_depth, input_height, input_width]`. Currently only support `"NDHWC"` .
name(str, optional): For detailed information, please refer
......@@ -62,9 +64,7 @@ def max_pool3d(x,
.. code-block:: python
import paddle
from paddle.fluid.framework import _test_eager_guard
with _test_eager_guard():
dense_x = paddle.randn((1, 4, 4, 4, 3))
sparse_x = dense_x.to_sparse_coo(4)
kernel_sizes = [3, 3, 3]
......@@ -75,9 +75,12 @@ def max_pool3d(x,
"""
assert in_dynamic_mode(), "Currently, Sparse API only support dynamic mode"
assert x.is_sparse_coo(
assert (
x.is_sparse_coo()
), "Currently, sparse.relu only support the input of SparseCooTensor"
assert data_format == 'NDHWC', "Currently, sparse.max_pool3d only support data format of 'NDHWC'"
assert (
data_format == 'NDHWC'
), "Currently, sparse.max_pool3d only support data format of 'NDHWC'"
kernel_size = utils.convert_to_list(kernel_size, 3, 'pool_size')
if stride is None:
......@@ -87,12 +90,11 @@ def max_pool3d(x,
channel_last = True
padding, padding_algorithm = _update_padding_nd(padding,
3,
channel_last=channel_last,
ceil_mode=ceil_mode)
padding, padding_algorithm = _update_padding_nd(
padding, 3, channel_last=channel_last, ceil_mode=ceil_mode
)
#TODO(zkh2016): remove the dependency on dilation from the backend
# TODO(zkh2016): remove the dependency on dilation from the backend
dilation = [1, 1, 1]
return _C_ops.sparse_maxpool(x, kernel_size, padding, dilation, stride)
python/paddle/sparse/nn/layer/conv.py
浏览文件 @ 61953b90
......@@ -23,8 +23,8 @@ __all__ = []
class _Conv3D(Layer):
def __init__(self,
def __init__(
self,
in_channels,
out_channels,
kernel_size,
......@@ -37,9 +37,12 @@ class _Conv3D(Layer):
padding_mode='zeros',
weight_attr=None,
bias_attr=None,
data_format="NDHWC"):
data_format="NDHWC",
):
super(_Conv3D, self).__init__()
assert weight_attr is not False, "weight_attr should not be False in Conv."
assert (
weight_attr is not False
), "weight_attr should not be False in Conv."
self._param_attr = weight_attr
self._bias_attr = bias_attr
self._groups = groups
......@@ -49,47 +52,56 @@ class _Conv3D(Layer):
self._subm = subm
self._key = key
assert padding_mode == 'zeros', "Currently, only support padding_mode='zeros'"
assert (
padding_mode == 'zeros'
), "Currently, only support padding_mode='zeros'"
assert groups == 1, "Currently, only support groups=1"
valid_format = {'NDHWC'}
if data_format not in valid_format:
raise ValueError(
"data_format must be one of {}, but got data_format='{}'".
format(valid_format, data_format))
"data_format must be one of {}, but got data_format='{}'".format(
valid_format, data_format
)
)
channel_last = data_format == "NDHWC"
dims = 3
self._stride = utils.convert_to_list(stride, dims, 'stride')
self._dilation = utils.convert_to_list(dilation, dims, 'dilation')
self._kernel_size = utils.convert_to_list(kernel_size, dims,
'kernel_size')
self._kernel_size = utils.convert_to_list(
kernel_size, dims, 'kernel_size'
)
self._padding = padding
self._padding_mode = padding_mode
self._updated_padding, self._padding_algorithm = _update_padding_nd(
padding, channel_last, dims)
padding, channel_last, dims
)
# the sparse conv restricts the shape is [D, H, W, in_channels, out_channels]
filter_shape = self._kernel_size + [
self._in_channels, self._out_channels
self._in_channels,
self._out_channels,
]
def _get_default_param_initializer():
filter_elem_num = np.prod(self._kernel_size) * self._in_channels
std = (2.0 / filter_elem_num)**0.5
std = (2.0 / filter_elem_num) ** 0.5
return Normal(0.0, std)
self.weight = self.create_parameter(
shape=filter_shape,
attr=self._param_attr,
default_initializer=_get_default_param_initializer())
self.bias = self.create_parameter(attr=self._bias_attr,
shape=[self._out_channels],
is_bias=True)
default_initializer=_get_default_param_initializer(),
)
self.bias = self.create_parameter(
attr=self._bias_attr, shape=[self._out_channels], is_bias=True
)
def forward(self, x):
out = F.conv._conv3d(x,
out = F.conv._conv3d(
x,
self.weight,
bias=self.bias,
stride=self._stride,
......@@ -98,7 +110,8 @@ class _Conv3D(Layer):
groups=self._groups,
subm=self._subm,
key=self._key,
data_format=self._data_format)
data_format=self._data_format,
)
return out
def extra_repr(self):
......@@ -208,9 +221,7 @@ class Conv3D(_Conv3D):
.. code-block:: python
import paddle
from paddle.fluid.framework import _test_eager_guard
with _test_eager_guard():
indices = [[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 1, 2], [1, 3, 2, 3]]
values = [[1], [2], [3], [4]]
indices = paddle.to_tensor(indices, dtype='int32')
......@@ -223,7 +234,8 @@ class Conv3D(_Conv3D):
# (1, 1, 1, 2, 1)
"""
def __init__(self,
def __init__(
self,
in_channels,
out_channels,
kernel_size,
......@@ -234,8 +246,10 @@ class Conv3D(_Conv3D):
padding_mode='zeros',
weight_attr=None,
bias_attr=None,
data_format="NDHWC"):
super(Conv3D, self).__init__(in_channels,
data_format="NDHWC",
):
super(Conv3D, self).__init__(
in_channels,
out_channels,
kernel_size,
stride=stride,
......@@ -247,13 +261,14 @@ class Conv3D(_Conv3D):
padding_mode=padding_mode,
weight_attr=weight_attr,
bias_attr=bias_attr,
data_format=data_format)
data_format=data_format,
)
class SubmConv3D(_Conv3D):
r"""
**Sparse Submanifold Convlution3d Layer**
The Sparse submanifold convolution3d layer calculates the output based on the input, filter
**Submanifold Sparse Convlution3d Layer**
The submanifold sparse convolution3d layer calculates the output based on the input, filter
and strides, paddings, dilations, groups parameters. Input(Input) and
Output(Output) are multidimensional SparseCooTensors with a shape of
:math:`[N, D, H, W, C]` . Where N is batch size, C is the number of
......@@ -264,7 +279,7 @@ class SubmConv3D(_Conv3D):
.. math::
Out =(W \ast X + b
Out = W \ast X + b
In the above equation:
......@@ -345,9 +360,7 @@ class SubmConv3D(_Conv3D):
.. code-block:: python
import paddle
from paddle.fluid.framework import _test_eager_guard
with _test_eager_guard():
indices = [[0, 0, 0, 0], [0, 0, 0, 0], [0, 0, 1, 2], [1, 3, 2, 3]]
values = [[1], [2], [3], [4]]
dense_shape = [1, 1, 3, 4, 1]
......@@ -360,7 +373,8 @@ class SubmConv3D(_Conv3D):
# (1, 1, 3, 4, 1)
"""
def __init__(self,
def __init__(
self,
in_channels,
out_channels,
kernel_size,
......@@ -372,8 +386,10 @@ class SubmConv3D(_Conv3D):
key=None,
weight_attr=None,
bias_attr=None,
data_format="NDHWC"):
super(SubmConv3D, self).__init__(in_channels,
data_format="NDHWC",
):
super(SubmConv3D, self).__init__(
in_channels,
out_channels,
kernel_size,
stride=stride,
......@@ -385,4 +401,5 @@ class SubmConv3D(_Conv3D):
padding_mode=padding_mode,
weight_attr=weight_attr,
bias_attr=bias_attr,
data_format=data_format)
data_format=data_format,
)
python/paddle/sparse/nn/layer/norm.py
浏览文件 @ 61953b90
......@@ -83,9 +83,7 @@ class BatchNorm(paddle.nn.BatchNorm1D):
.. code-block:: python
import paddle
from paddle.fluid.framework import _test_eager_guard
with _test_eager_guard():
paddle.seed(123)
channels = 3
x_data = paddle.randn((1, 6, 6, 6, channels)).astype('float32')
......@@ -271,6 +269,8 @@ class SyncBatchNorm(paddle.nn.SyncBatchNorm):
will create ParamAttr as bias_attr. If the Initializer of the bias_attr
is not set, the bias is initialized zero. If it is set to False, this layer will not
have trainable bias parameter. Default: None.
data_format(str, optional): Specify the input data format, may be "NCHW". Default "NCHW".
name(str, optional): Name for the BatchNorm, default is None. For more information, please refer to :ref:`api_guide_Name`..
Shapes:
input: Tensor that the dimension from 2 to 5.
......@@ -283,10 +283,8 @@ class SyncBatchNorm(paddle.nn.SyncBatchNorm):
# required: gpu
import paddle
import paddle.sparse.nn as nn
import numpy as np
x = np.array([[[[0.3, 0.4], [0.3, 0.07]], [[0.83, 0.37], [0.18, 0.93]]]]).astype('float32')
x = paddle.to_tensor(x)
x = paddle.to_tensor([[[[0.3, 0.4], [0.3, 0.07]], [[0.83, 0.37], [0.18, 0.93]]]], dtype='float32')
x = x.to_sparse_coo(len(x.shape)-1)
if paddle.is_compiled_with_cuda():
......
python/paddle/sparse/nn/layer/pooling.py
浏览文件 @ 61953b90
......@@ -61,9 +61,7 @@ class MaxPool3D(Layer):
.. code-block:: python
import paddle
from paddle.fluid.framework import _test_eager_guard
with _test_eager_guard():
dense_x = paddle.randn((2, 3, 6, 6, 3))
sparse_x = dense_x.to_sparse_coo(4)
max_pool3d = paddle.sparse.nn.MaxPool3D(
......@@ -73,14 +71,16 @@ class MaxPool3D(Layer):
"""
def __init__(self,
def __init__(
self,
kernel_size,
stride=None,
padding=0,
return_mask=False,
ceil_mode=False,
data_format="NDHWC",
name=None):
name=None,
):
super(MaxPool3D, self).__init__()
self.ksize = kernel_size
self.stride = stride
......@@ -91,14 +91,17 @@ class MaxPool3D(Layer):
self.name = name
def forward(self, x):
return F.max_pool3d(x,
return F.max_pool3d(
x,
kernel_size=self.ksize,
stride=self.stride,
padding=self.padding,
ceil_mode=self.ceil_mode,
data_format=self.data_format,
name=self.name)
name=self.name,
)
def extra_repr(self):
return 'kernel_size={ksize}, stride={stride}, padding={padding}'.format(
**self.__dict__)
**self.__dict__
)
python/paddle/sparse/unary.py
浏览文件 @ 61953b90
......@@ -15,7 +15,11 @@
import numpy as np
from paddle import _C_ops, _legacy_C_ops
from paddle.fluid.framework import dygraph_only, core, convert_np_dtype_to_dtype_
from paddle.fluid.framework import (
dygraph_only,
core,
convert_np_dtype_to_dtype_,
)
__all__ = []
......@@ -517,12 +521,14 @@ def abs(x, name=None):
@dygraph_only
def coalesce(x):
def coalesce(x, name=None):
r"""
the coalesced operator include sorted and merge, after coalesced, the indices of x is sorted and unique.
Parameters:
x (Tensor): the input SparseCooTensor.
name (str, optional): Name for the operation (optional, default is None).
For more information, please refer to :ref:`api_guide_Name`.
Returns:
Tensor: return the SparseCooTensor after coalesced.
......
Markdown is supported
0% 或 .
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 或