[PaddlePaddle hackathon] paddle.nn.PixelShuffle单测 #226
Conversation
| self.types = [np.float32] | ||
|
|
||
|
|
||
| obj = TestClipGradByNorm(paddle.nn.ClipGradByNorm) |
…test_clip_grad_by_norm.py, modify test_pixel_shuffle.py
justld
changed the title
|
PR types：New features PR types：New features PR types：New features |
justld
changed the title
|
PR types：New features PR types：New features PR types：New features PR types：New features |
|
hi，几个任务尽量别放在一个PR里，一个Task对应一个PR，避免Task间的相互影响合入，以后的PR需要注意下 谢谢~ |
| def test_clip_grad_by_global_norm1(): | ||
| """ | ||
| input shape | ||
| """ |
There was a problem hiding this comment.
doc部分详细描述下该测试用例的验证点，测试输入，预期结果等，通过doc部分可以清除的知晓该用例的测试点。，其他的case也类似
|
您好，已经添加了测试部分的注释 |
|
|
||
| # compare grad value computed by numpy and paddle | ||
| for res, p_res in zip(np_res, paddle_res): | ||
| compare(res[1], p_res[1]) |
There was a problem hiding this comment.
需要完善测试case
1、尝试不同输入类型下，目前我看都是float32，需要验证其他数据类型下是否符合预期。
2、输入shape为一维、
3、该类的功能中说明(将一个 Tensor列表 t_list 中所有Tensor的L2范数之和，限定在 clip_norm 范围内),case中需要显示验证这点。
| up_factor = 3 | ||
| data_format = "NCHW" | ||
| res = pixel_shuffle_np(x, up_factor, data_format=data_format) | ||
| obj.run(res=res, data=x, upscale_factor=up_factor, data_format=data_format) |
There was a problem hiding this comment.
需要补充up_factor的值和输入x中的不能整除时的情况case。当前的case中up_factor(3)都是刚好能正常分解输入x。
| tensor_x = paddle.to_tensor(x) | ||
| paddle_res = paddle.nn.UpsamplingBilinear2D(size=size, scale_factor=scale_factor, data_format=data_format)(tensor_x) | ||
| paddle_res = paddle_res.numpy() | ||
| compare(res, paddle_res, delta, rtol) |
There was a problem hiding this comment.
size (list|tuple|Tensor|None) - 输出Tensor，输入为4D张量，形状为为(out_h, out_w)的2-D Tensor。如果 size 是列表，每一个元素可以是整数或者形状为[1]的变量。如果 size 是变量，则其维度大小为1。默认值为None。
scale_factor (float|Tensor|list|tuple|None)-输入的高度或宽度的乘数因子。 size 和 scale_factor 至少要设置一个。 size 的优先级高于 scale_factor 。默认值为None。如果 scale_factor 是一个list或tuple，它必须与输入的shape匹配。
针对类的不同参数添加更丰富的case，不同的输入类型，shape等、异常情况等
justld
changed the title
| data_format = "NCHW" | ||
| with pytest.raises(ValueError): | ||
| pixel_shuffle = paddle.nn.PixelShuffle(upscale_factor=up_factor, data_format=data_format) | ||
| pixel_shuffle(x) |
There was a problem hiding this comment.
可以使用apibase里面的异常类。需要判断paddle抛出异常时的报错信息
There was a problem hiding this comment.
您好，这里需要在实例化类的时候传递参数，然后运行该实例时才会引发异常，请问该使用apibase的哪个类呢。
There was a problem hiding this comment.
您好，我已经改好了，请帮忙再review一下。
PR types
New features
PR changes
添加/framwork/api/nn/test_pixel_shuffle.py
Describe
Task: #35904
添加paddle.nn.PixelShuffle单测