sequence_expand_as¶

paddle.static.nn. sequence_expand_as ( x, y, name=None ) [source]

Sequence Expand As Layer. This OP will expand the input variable x according to the zeroth level lod of y. Current implementation requires the level number of y’s lod must be 1, and the first dimension of x should be equal to the size of y’s zeroth level lod, thus the expanded Tensor has the same lod info as y. The expanded result has nothing to do with x’s lod, so the lod of Input(X) is not considered.

Note

Please note that the input x should be Tensor or Tensor, and input y must be Tensor.

Following examples will explain how sequence_expand_as works:

Case 1:

Consider 4 sequences [a], [b], [c], [d], now we want to expand them to [a][a][a], [b][b][b], [c] and [d].
It's obvious that the lod info of expanded sequences is [0, 3, 6, 7, 8].
Given a 1-level Tensor ``x``:
    x.data = [[a], [b], [c], [d]]
    x.dims = [4, 1]
and input ``y``
    y.lod = [[3, 3, 1, 1]] #lod based on length may be easier to understand

then we get 1-level Tensor out:
    Out.lod =  [[0,            3,              6,  7,  8]] #based on offset
    Out.data = [[a], [a], [a], [b], [b], [b], [c], [d]]
    Out.dims = [8, 1]


Case 2:

Given a common Tensor ``x``:
    x.data = [[a, b], [c, d], [e, f]]
    x.dims = [3, 2]
and input ``y``:
    y.lod = [[0, 2, 3, 6]]

then we get a 1-level Tensor:
    out.lod =  [[0,             2,     3,                    6]]
    out.data = [[a, b], [a, b] [c, d], [e, f], [e, f], [e, f]]
    out.dims = [6, 2]

Parameters

• x (Tensor) – The input variable which is a Tensor or Tensor, with the dims [M, K]. The data type should be float32, float64, int32 or int64.

• y (Tensor) – The input variable which is a Tensor with 1-level lod.

• name (str, optional) – For detailed information, please refer to Name. Usually name is no need to set and None by default.

Returns

Tensor, The expanded variable which is a Tensor with the dims [N, K]. N depends on the lod of y, and the lod level must be 1. The data type is same as input.

Examples

>>> import paddle
>>> import paddle.base as base
>>> paddle.enable_static()
>>> import numpy as np

>>> x = paddle.static.data(name='x', shape=[4, 1], dtype='float32')
>>> y = paddle.static.data(name='y', shape=[8, 1], dtype='float32', lod_level=1)
>>> out = paddle.static.nn.sequence_expand_as(x=x, y=y)

>>> exe = base.Executor(base.CPUPlace())
>>> place = base.CPUPlace()

>>> np_data = np.array([[1], [2], [3], [4]]).astype('float32')
>>> x_lod_tensor = base.create_lod_tensor(np_data, [[2, 2]], place)
>>> print(x_lod_tensor)
- lod: {{0, 2, 4}}
- place: Place(cpu)
- shape: [4, 1]
- layout: NCHW
- dtype: float32
- data: [1 2 3 4]

>>> np_data = np.array([[1], [2], [3], [4], [5], [6], [7], [8]]).astype('float32')
>>> y_lod_tensor = base.create_lod_tensor(np_data, [[3,3,1,1]], place)
>>> print(y_lod_tensor)
- lod: {{0, 3, 6, 7, 8}}
- place: Place(cpu)
- shape: [8, 1]
- layout: NCHW
- dtype: float32
- data: [1 2 3 4 5 6 7 8]

>>> out_main = exe.run(base.default_main_program(),
...                 feed={'x': x_lod_tensor, 'y': y_lod_tensor},
...                 fetch_list=[out], return_numpy=False)
>>> print(out_main[0])
- lod: {{0, 3, 6, 7, 8}}
- place: Place(cpu)
- shape: [8, 1]
- layout: NCHW
- dtype: float32
- data: [1 1 1 2 2 2 3 4]