【Hackathon 9th No.13】Fix accuracy for paddle.unique

[ooooo-create]

paddle.unique 在 axis=None，return_inverse = True时，返回的 inverse 是被 flatten 之后的，但是 numpy 和 torch 会保持原来输入的形状

import paddle
import torch
import numpy as np

# 构造一个二维张量
x_np = np.array([[1, 2, 1], [3, 2, 1]])
x_pd = paddle.to_tensor(x_np)
x_th = torch.tensor(x_np)

# paddle.unique
out_pd, inverse_pd = paddle.unique(x_pd, return_inverse=True)
print("Paddle unique output:", out_pd.numpy())
print("Paddle inverse:", inverse_pd.numpy(), "shape:", inverse_pd.shape)

# torch.unique
out_th, inverse_th = torch.unique(x_th, return_inverse=True)
print("Torch unique output:", out_th.numpy())
print("Torch inverse:", inverse_th.numpy(), "shape:", inverse_th.shape)

"""
Paddle unique output: [1 2 3]
Paddle inverse: [0 1 0 2 1 0] shape: [6]
Torch unique output: [1 2 3]
Torch inverse: [[0 1 0]
 [2 1 0]] shape: torch.Size([2, 3])
"""

[paddle-bot]

Thanks for your contribution!

[cangtianhuang]

LGTM