(this issue is just for my personal self-learning ) not flexible enough: paddle.scatter()

[OccupyMars2025]

问题描述 Please describe your issue

"torch code, can be replaced with torch.nn.functional.one_hot"
def one_hot_torch(x, num_classes, on_value=1., off_value=0., device='cuda'):
x = x.long().view(-1, 1)
return torch.full((x.size()[0], num_classes), off_value, device=device).scatter_(1, x, on_value)

“paddle version”
def one_hot(x, num_classes, on_value=1., off_value=0., device='cuda'):
x = x.astype("int64").reshape([-1])
out = paddle.nn.functional.one_hot(x, num_classes=num_classes)
out = paddle.where(out > 0.5, on_value, off_value)
out = paddle.to_tensor(out, place=paddle.CUDAPlace(0) if paddle.is_compiled_with_cuda() else paddle.CPUPlace())
return out

TODO: When I study the above code, I find out that compared to torch.scatter(), paddle.scatter() is not flexible enough. The key problems are how to "index" and how to handle the case when there are repeated numbers in "index" especially in GPU mode ( simultaneous running threads that work on the same element in a tensor may have conflicts ???)

The following two PRs may have the same problems:
paddle.index_add
paddle.index_fill

[paddle-bot-old]

您好，我们已经收到了您的问题，会安排技术人员尽快解答您的问题，请耐心等待。请您再次检查是否提供了清晰的问题描述、复现代码、环境&版本、报错信息等。同时，您也可以通过查看官网API文档、常见问题、历史Issue、AI社区来寻求解答。祝您生活愉快～

Hi! We've received your issue and please be patient to get responded. We will arrange technicians to answer your questions as soon as possible. Please make sure that you have posted enough message to demo your request. You may also check out the API，FAQ，Github Issue and AI community to get the answer.Have a nice day!

[SunNy820828449]

Thanks for your issue. We will feedback this issue to our developer.

[Ligoml]

In fact, paddle.put_along_axis and torch.scatter are corresponding, we suggest you try paddle.put_along_axis to solve this issue. We will add clarification later in the documentation.

[OccupyMars2025]

Thank you for your answer.