【Hackathon No.2】为 Paddle 新增 cdist API

[GreatV]

PR types

New features

PR changes

APIs

Description

Add cdist op

• RFC: paddle.cdist 设计文档

[paddle-bot]

你的PR提交成功，感谢你对开源项目的贡献!
请关注后续CI自动化测试结果，详情请参考Paddle-CI手册。
Your PR has been submitted. Thanks for your contribution!
Please wait for the result of CI firstly. See Paddle CI Manual for details.

[paddle-bot]

❌ The PR is not created using PR's template. You can refer to this Demo.
Please use PR's template, it helps save our maintainers' time so that more developers get helped.

[GGBond8488]

Hello, 对于非必要新增op的API ，目前我们不倾向新增OP，最好以组合的形式实现该API

[GreatV]

Hello, 对于非必要新增op的API ，目前我们不倾向新增OP，最好以组合的形式实现该API

只用python端实现类似这样的效果，就可以了吗

import paddle

def _cdist(x : paddle.Tensor, y : paddle.Tensor, p=2, name=None):
    r1 = x.shape[-2]
    r2 = y.shape[-2]
    if r1 == 0 or r2 == 0:
        return paddle.empty((r1, r2), dtype=x.dtype, name=name)
    
    return paddle.norm(x[..., None, :] - y[..., None, :, :], p=p, axis=-1, name=name)

[GGBond8488]

Hello, 对于非必要新增op的API ，目前我们不倾向新增OP，最好以组合的形式实现该API

只用python端实现类似这样的效果，就可以了吗

import paddle

def _cdist(x : paddle.Tensor, y : paddle.Tensor, p=2, name=None):
    r1 = x.shape[-2]
    r2 = y.shape[-2]
    if r1 == 0 or r2 == 0:
        return paddle.empty((r1, r2), dtype=x.dtype, name=name)
    
    return paddle.norm(x[..., None, :] - y[..., None, :, :], p=p, axis=-1, name=name)

对的

[GGBond8488]

LGTM

[sunzhongkai588]

辛苦修改～同时请提供中文文档

[GreatV]

辛苦修改～同时请提供中文文档

好的，下个PR，提供中文文档。

[jeff41404]

Some descriptions in rfc need to be modified:
chapter1，“功能目标” 中 “为 Paddle 新增 cdist API。dist API 用于计算两个输入 Tensor 的 p 范数（p-norm），计算结果为形状为 [1] 的 Tensor，而 cdist API 则用于计算两个输入 Tensor 的所有行向量对的 p 范数（p-norm），输出结果的形状和两个 Tensor 乘积的形状一致。” we have corrected the semantics of the 0-dimensional tensor in the develop version, and the shape of output of dist is [], not [1].
chapter5，“命名与参数设计“ 中”API设计为 paddle.cdist(x1, x2, p=2.0, compute_mode='use_mm_for_euclid_dist_if_necessary')” is different from the parameters in the code: paddle.cdist(x, y, p=2.0, compute_mode='use_mm_for_euclid_dist_if_necessary')”

[GreatV]

Some descriptions in rfc need to be modified: chapter1，“功能目标” 中 “为 Paddle 新增 cdist API。dist API 用于计算两个输入 Tensor 的 p 范数（p-norm），计算结果为形状为 [1] 的 Tensor，而 cdist API 则用于计算两个输入 Tensor 的所有行向量对的 p 范数（p-norm），输出结果的形状和两个 Tensor 乘积的形状一致。” we have corrected the semantics of the 0-dimensional tensor in the develop version, and the shape of output of dist is [], not [1]. chapter5，“命名与参数设计“ 中”API设计为 paddle.cdist(x1, x2, p=2.0, compute_mode='use_mm_for_euclid_dist_if_necessary')” is different from the parameters in the code: paddle.cdist(x, y, p=2.0, compute_mode='use_mm_for_euclid_dist_if_necessary')”

Got it.

[sunzhongkai588]

LGTM for docs，中文文档链接之后请附上

[jeff41404]

LGTM

[luotao1]

请 @XieYunshen 看下单测时间，超时的理由在 https://github.com//pull/53836#issuecomment-1566777967

[XieYunshen]

LGTM
单测超时时间设置