add autotest code in new test framework #5912
Conversation
| @@ -1,12 +1,9 @@ | |||
| """ | |||
| Copyright 2020 The OneFlow Authors. All rights reserved. | |||
|
|
|||
There was a problem hiding this comment.
这个版权信息为什么会被改，在 build 里 make of_format 可以本地做代码格式化
There was a problem hiding this comment.
抄了这个PR https://github.com/Oneflow-Inc/oneflow/pull/5799/files 看见他几乎是重写了文件
| @autotest() | ||
| def test_sub_with_random_data(test_case): | ||
| device = random_device() | ||
| x = random_pytorch_tensor(2, 0, 3).to(device) |
There was a problem hiding this comment.
测试样例不够丰富，可以多测几个形状和情况（element-wise的、scalar的、broadcast的）
|
暂时不支持的参数，就不测了。不过最好要备忘一下，之后一起列个总结的单子。 |
…/oneflow into add_new_autotest_zgh
|
masked_fill是对齐了，测试样例已经写了，在这里： |
好的，已经看见了，然后还有一个 |
嗯，我看一下 |
|
| import oneflow as flow | ||
| import oneflow.unittest | ||
|
|
||
|
|
| import unittest | ||
|
|
||
| import numpy as np | ||
| from automated_test_util import * |
There was a problem hiding this comment.
标准库、第三方库、oneflow 分三类，这里没分好
| device = random_device() | ||
| x = random_pytorch_tensor(ndim=4, dim0=1).to(device) | ||
| k = random(10,100).to(int) | ||
| y = torch.mul(x, k) |
There was a problem hiding this comment.
原测试案例覆盖了各种情况： element-wise, scalar, broadcast 的，这里只测了 element-wise的。
请自己检查其它test case
There was a problem hiding this comment.
div 等应该也有这类覆盖不全的问题。
如果不知道要测那些情况，就应该看文档、自己试用、看源码。了解接口的功能。
…/oneflow into add_new_autotest_zgh
| @autotest(auto_backward=False) | ||
| def test_0shape_div(test_case): | ||
| @autotest() | ||
| def test_div_with_random_data_Number(test_case): |
There was a problem hiding this comment.
| def test_div_with_random_data_Number(test_case): | |
| def test_div_with_random_data(test_case): |
| @autotest() | ||
| def test_div_with_with_broadcast(test_case): | ||
| device = random_device() | ||
| x = random_pytorch_tensor(ndim=4, dim0=3).to(device) |
|
|
||
| @autotest() |
There was a problem hiding this comment.
我发现你把0-dim tensor的测试删了，这个不能删啊，这是专门构造的0-dim tensor测试样例
torch.min(input, dim, keepdim=False, *, out=None),oneflow缺少dim参数、keepdim参数、out参数Tensor.min(dim=None, keepdim=False),oneflow缺少dim参数、keepdim参数torch.max(input, dim, keepdim=False, *, out=None),oneflow缺少dim参数、keepdim参数、out参数Tensor.max(dim=None, keepdim=False),oneflow缺少dim参数、keepdim参数torch.mul(input, other, *, out=None)，oneflow缺少out参数torch.sub(input, other, *, alpha=1, out=None)，oneflow缺少alpha参数、out参数Tensor.sub(other, *, alpha=1),oneflow缺少alpha参数torch.div(input, other, *, rounding_mode=None, out=None)，oneflow缺少rounding_mode参数、out参数Tensor.div(value, *, rounding_mode=1),oneflow缺少rounding_mode参数Notice:
1、
oneflow.masked_fill和oneflow.Tensor.masked_fill接口似乎还有问题，原来的测试文档标注了
@unittest.skip("has bug now, need rewrite")2、
oneflow.suboneflow.Tensor.sub似乎没有对齐，
torch.sub(input, other, *, alpha=1, out=None)，onflow缺少alpha参数取消参数alpha：
3、
oneflow.min oneflow.max oneflow.Tensor.min oneflow.Tensor.maxoneflow没对齐？
torch.min(input, dim, keepdim=False, *, out=None)4、
oneflow.divoneflow.Tensor.divoneflow缺少参数rounding_mode
torch.div(input, other, *, rounding_mode=None, out=None)5、
oneflow.var存在问题