【Hackathon 6th No.39】XPINN 迁移至 PaddleScience #849
Merged
Conversation
This file contains bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
|
Thanks for your contribution! |
HydrogenSulfate
requested changes
Apr 15, 2024
examples/xpinn/xpinn.py
Outdated
|
|
||
| import ppsci | ||
|
|
||
| # For the use of the second derivative: paddle.cos, paddle.exp |
There was a problem hiding this comment.
可以删除注释里的paddle.exp，默认模式已经支持无限阶微分
examples/xpinn/xpinn.py
Outdated
Comment on lines
34
to
126
| def xpinn_2d( | ||
| x1: paddle.Tensor, | ||
| y1: paddle.Tensor, | ||
| u1: paddle.Tensor, | ||
| x2: paddle.Tensor, | ||
| y2: paddle.Tensor, | ||
| u2: paddle.Tensor, | ||
| x3: paddle.Tensor, | ||
| y3: paddle.Tensor, | ||
| u3: paddle.Tensor, | ||
| xi1: paddle.Tensor, | ||
| yi1: paddle.Tensor, | ||
| u1i1: paddle.Tensor, | ||
| u2i1: paddle.Tensor, | ||
| xi2: paddle.Tensor, | ||
| yi2: paddle.Tensor, | ||
| u1i2: paddle.Tensor, | ||
| u3i2: paddle.Tensor, | ||
| ub: paddle.Tensor, | ||
| ub_pred: paddle.Tensor, | ||
| ): | ||
| u1_x = get_grad(u1, x1) | ||
| u1_y = get_grad(u1, y1) | ||
| u1_xx = get_grad(u1_x, x1) | ||
| u1_yy = get_grad(u1_y, y1) | ||
|
|
||
| u2_x = get_grad(u2, x2) | ||
| u2_y = get_grad(u2, y2) | ||
| u2_xx = get_grad(u2_x, x2) | ||
| u2_yy = get_grad(u2_y, y2) | ||
|
|
||
| u3_x = get_grad(u3, x3) | ||
| u3_y = get_grad(u3, y3) | ||
| u3_xx = get_grad(u3_x, x3) | ||
| u3_yy = get_grad(u3_y, y3) | ||
|
|
||
| u1i1_x = get_grad(u1i1, xi1) | ||
| u1i1_y = get_grad(u1i1, yi1) | ||
| u1i1_xx = get_grad(u1i1_x, xi1) | ||
| u1i1_yy = get_grad(u1i1_y, yi1) | ||
|
|
||
| u2i1_x = get_grad(u2i1, xi1) | ||
| u2i1_y = get_grad(u2i1, yi1) | ||
| u2i1_xx = get_grad(u2i1_x, xi1) | ||
| u2i1_yy = get_grad(u2i1_y, yi1) | ||
|
|
||
| u1i2_x = get_grad(u1i2, xi2) | ||
| u1i2_y = get_grad(u1i2, yi2) | ||
| u1i2_xx = get_grad(u1i2_x, xi2) | ||
| u1i2_yy = get_grad(u1i2_y, yi2) | ||
|
|
||
| u3i2_x = get_grad(u3i2, xi2) | ||
| u3i2_y = get_grad(u3i2, yi2) | ||
| u3i2_xx = get_grad(u3i2_x, xi2) | ||
| u3i2_yy = get_grad(u3i2_y, yi2) | ||
|
|
||
| uavgi1 = (u1i1 + u2i1) / 2 | ||
| uavgi2 = (u1i2 + u3i2) / 2 | ||
|
|
||
| # Residuals | ||
| f1 = u1_xx + u1_yy - (paddle.exp(x1) + paddle.exp(y1)) | ||
| f2 = u2_xx + u2_yy - (paddle.exp(x2) + paddle.exp(y2)) | ||
| f3 = u3_xx + u3_yy - (paddle.exp(x3) + paddle.exp(y3)) | ||
|
|
||
| # Residual continuity conditions on the interfaces | ||
| fi1 = (u1i1_xx + u1i1_yy - (paddle.exp(xi1) + paddle.exp(yi1))) - ( | ||
| u2i1_xx + u2i1_yy - (paddle.exp(xi1) + paddle.exp(yi1)) | ||
| ) | ||
| fi2 = (u1i2_xx + u1i2_yy - (paddle.exp(xi2) + paddle.exp(yi2))) - ( | ||
| u3i2_xx + u3i2_yy - (paddle.exp(xi2) + paddle.exp(yi2)) | ||
| ) | ||
|
|
||
| loss1 = ( | ||
| 20 * paddle.mean(paddle.square(ub - ub_pred)) | ||
| + paddle.mean(paddle.square(f1)) | ||
| + 1 * paddle.mean(paddle.square(fi1)) | ||
| + 1 * paddle.mean(paddle.square(fi2)) | ||
| + 20 * paddle.mean(paddle.square(u1i1 - uavgi1)) | ||
| + 20 * paddle.mean(paddle.square(u1i2 - uavgi2)) | ||
| ) | ||
|
|
||
| loss2 = ( | ||
| paddle.mean(paddle.square(f2)) | ||
| + 1 * paddle.mean(paddle.square(fi1)) | ||
| + 20 * paddle.mean(paddle.square(u2i1 - uavgi1)) | ||
| ) | ||
|
|
||
| loss3 = ( | ||
| paddle.mean(paddle.square(f3)) | ||
| + 1 * paddle.mean(paddle.square(fi2)) | ||
| + 20 * paddle.mean(paddle.square(u3i2 - uavgi2)) | ||
| ) | ||
| return loss1, loss2, loss3 |
There was a problem hiding this comment.
- 函数内重复代码过多，建议化简下
- 这样的XPINN写法感觉不太具备可扩展性，换一个案例这个函数就不能用了
There was a problem hiding this comment.
重新实现了XPINN的loss函数
examples/xpinn/xpinn.py
Outdated
Comment on lines
129
to
156
| def loss_fun( | ||
| output_dict: Dict[str, paddle.Tensor], | ||
| label_dict: Dict[str, paddle.Tensor], | ||
| *args, | ||
| ): | ||
| loss1, loss2, loss3 = xpinn_2d( | ||
| output_dict["x_f1"], | ||
| output_dict["y_f1"], | ||
| output_dict["u1"], | ||
| output_dict["x_f2"], | ||
| output_dict["y_f2"], | ||
| output_dict["u2"], | ||
| output_dict["x_f3"], | ||
| output_dict["y_f3"], | ||
| output_dict["u3"], | ||
| output_dict["xi1"], | ||
| output_dict["yi1"], | ||
| output_dict["u1i1"], | ||
| output_dict["u2i1"], | ||
| output_dict["xi2"], | ||
| output_dict["yi2"], | ||
| output_dict["u1i2"], | ||
| output_dict["u3i2"], | ||
| label_dict["ub"], | ||
| output_dict["ub_pred"], | ||
| ) | ||
|
|
||
| return loss1 + loss2 + loss3 |
examples/xpinn/xpinn.py
Outdated
| ) | ||
|
|
||
| error_u_total = paddle.linalg.norm( | ||
| paddle.squeeze(u_exact) - u_pred.flatten(), 2 |
There was a problem hiding this comment.
squeeze改为flatten，两边语义一致
examples/xpinn/xpinn.py
Outdated
Comment on lines
225
to
238
| "input_keys": ( | ||
| "x_f1", | ||
| "y_f1", | ||
| "x_f2", | ||
| "y_f2", | ||
| "x_f3", | ||
| "y_f3", | ||
| "xi1", | ||
| "yi1", | ||
| "xi2", | ||
| "yi2", | ||
| "xb", | ||
| "yb", | ||
| ), |
There was a problem hiding this comment.
分解后的input_keys可以由基本的input_keys+XPINN生成，不建议使用hard code
There was a problem hiding this comment.
input_keys 放在了配置文件中
examples/xpinn/xpinn.py
Outdated
| "xb", | ||
| "yb", | ||
| ), | ||
| "label_keys": ("ub", "u_exact", "u_exact2", "u_exact3"), |
There was a problem hiding this comment.
u_exact==>u_exact1?
There was a problem hiding this comment.
数据集中的名字是 u_exact
examples/xpinn/xpinn.py
Outdated
Comment on lines
352
to
363
| x1=input_["x_f1"], | ||
| y1=input_["y_f1"], | ||
| x2=input_["x_f2"], | ||
| y2=input_["y_f2"], | ||
| x3=input_["x_f3"], | ||
| y3=input_["y_f3"], | ||
| xi1=input_["xi1"], | ||
| yi1=input_["yi1"], | ||
| xi2=input_["xi2"], | ||
| yi2=input_["yi2"], | ||
| xb=input_["xb"], | ||
| yb=input_["yb"], |
There was a problem hiding this comment.
不建议使用x_f1、xb、yi1这类命名方式，看不太出来每个字段的含义
There was a problem hiding this comment.
好的，修改了命名
examples/xpinn/xpinn.py
Outdated
Comment on lines
283
to
296
| "input_keys": ( | ||
| "x_f1", | ||
| "y_f1", | ||
| "x_f2", | ||
| "y_f2", | ||
| "x_f3", | ||
| "y_f3", | ||
| "xi1", | ||
| "yi1", | ||
| "xi2", | ||
| "yi2", | ||
| "xb", | ||
| "yb", | ||
| ), |
There was a problem hiding this comment.
不建议使用x_f1、xb、yi1这类命名方式，看不太出来每个字段的含义
There was a problem hiding this comment.
好的，修改了命名
examples/xpinn/xpinn.py
Outdated
| ): | ||
| for key in in_: | ||
| in_[key] = paddle.cast(in_[key], paddle.float64) | ||
| return in_, _label, _weight |
HydrogenSulfate
requested changes
May 11, 2024
ppsci/utils/reader.py
Outdated
| if not np.issubdtype(data_dict[key].dtype, np.integer): | ||
| if ( | ||
| not np.issubdtype(data_dict[key].dtype, np.integer) | ||
| and data_dict[key].dtype != np.float64 |
There was a problem hiding this comment.
这里这么改会导致float64类型的数据不能被转换为float32，模型训练时数据与权重参数类型不匹配而报错。
There was a problem hiding this comment.
之前误以为有些算子只支持float64，已经修改。文档还在编写中。
HydrogenSulfate
requested changes
May 14, 2024
examples/xpinn/xpinn.py
Outdated
Comment on lines
33
to
36
| def get_grad(outputs: paddle.Tensor, inputs: paddle.Tensor) -> paddle.Tensor: | ||
| grad = paddle.grad(outputs, inputs, retain_graph=True, create_graph=True) | ||
| return grad[0] | ||
|
|
There was a problem hiding this comment.
这个函数看起来只被单独调用，建议放到get_second_derivatives上方，并且函数名前加上单个下划线
examples/xpinn/xpinn.py
Outdated
| residual_func (Callable, optional): residual calculation function. Defaults to lambda x,y : x - y. | ||
| """ | ||
|
|
||
| def get_second_derivatives( |
examples/xpinn/xpinn.py
Outdated
| return grad[0] | ||
|
|
||
|
|
||
| def xpinn_loss( |
| ] | ||
| ) | ||
|
|
||
| # the shape of label_dict["residual_u_exact"] is [22387, 1], and be cut into [18211, 1] `_eval_by_dataset`(ppsci/solver/eval.py). |
There was a problem hiding this comment.
这个注释是什么意思呢？是输入的样本点数不等于输出的样本点数，导致在_eval_by_dataset被丢掉？
There was a problem hiding this comment.
是的，输入的样本点数小于标签点数，数据集提供的label_dict["residual_u_exact"] 本来包含了label_dict["residual2_u_exact"] 和 label_dict["residual3_u_exact"]。
examples/xpinn/xpinn.py
Outdated
| error_u_total = paddle.linalg.norm( | ||
| u_exact.flatten() - u_pred.flatten(), 2 | ||
| ) / paddle.linalg.norm(u_exact.flatten(), 2) | ||
| return {"total": error_u_total} |
This comment was marked as outdated.
This comment was marked as outdated.
Sorry, something went wrong.
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.
This suggestion is invalid because no changes were made to the code.
Suggestions cannot be applied while the pull request is closed.
Suggestions cannot be applied while viewing a subset of changes.
Only one suggestion per line can be applied in a batch.
Add this suggestion to a batch that can be applied as a single commit.
Applying suggestions on deleted lines is not supported.
You must change the existing code in this line in order to create a valid suggestion.
Outdated suggestions cannot be applied.
This suggestion has been applied or marked resolved.
Suggestions cannot be applied from pending reviews.
Suggestions cannot be applied on multi-line comments.
Suggestions cannot be applied while the pull request is queued to merge.
Suggestion cannot be applied right now. Please check back later.
PR types
Others
PR changes
Others
Describe
原案例精度：0.04685
复现精度：0.04177