Hackathon 5th No.63 PhyCRNet: Physics-informed Convolutional-Recurrent Network for Solving Spatiotemporal PDEs #557
Conversation
|
Thanks for your contribution! |
|
黑客松的赛题有要求需要基于PaddleScience套件进行开发， |
|
|
luotao1
assigned luotao1, zhiminzhang0830 and wangguan1995 and unassigned zhiminzhang0830
|
更新使用 PaddleScience套件，增加ppsci.arch.PhyCRNet |
|
没有使用ppsci.utils.load_mat_file，ppsci.utils.load_mat_file中会对数据reshape，没有明白怎么要对数据reshape |
|
|
||
| ## 5. 结果展示 | ||
|
|
||
| PhyCRNet 案例针对 epoch=200 和 learning\_rate=1e-4 的参数配置进行了实验，结果返回Loss为 17.86。 |
There was a problem hiding this comment.
compute the error
error = frobenius_norm(np.array(ten_pred)-np.array(ten_true)) / frobenius_norm(
np.array(ten_true))
print('The predicted error is: ', error)
u_pred = output[:-1, 0, :, :].detach().cpu().numpy()
u_pred = np.swapaxes(u_pred, 1, 2) # [h,w] = [y,x]
u_true = truth[:, 0, :, :]
t_true = np.linspace(0, 2, 1001)
t_pred = np.linspace(0, 2, time_steps)
plt.plot(t_pred, u_pred[:, 32, 32], label='x=32, y=32, CRL')
plt.plot(t_true, u_true[:, 32, 32], '--', label='x=32, y=32, Ref.')
plt.xlabel('t')
plt.ylabel('u')
plt.xlim(0, 2)
plt.legend()
plt.savefig(fig_save_path + "x=32,y=32.png")
plt.close("all")
# plt.show()
# plot train loss
plt.figure()
plt.plot(train_loss, label = 'train loss')
plt.yscale('log')
plt.legend()
plt.savefig(fig_save_path + 'train loss.png', dpi = 300)
科学计算类论文复现，更关注 物理error和最后的物理场效果，参考原文代码复现error和对比图
（可以从checkpoint加载模型直接eval，不需要重新训练）
There was a problem hiding this comment.
增加变量graph，设置为True生成error，对比图
| # initialize logger | ||
| logger.init_logger("ppsci", f"{OUTPUT_DIR}/train.log", "info") | ||
| # set training hyper-parameters | ||
| EPOCHS = 200 if not args.epochs else args.epochs |
| return result_dict["loss"] | ||
|
|
||
|
|
||
| def output_graph(model, input_dataset, fig_save_path): |
| @@ -0,0 +1,273 @@ | |||
| # Copyright (c) 2023 PaddlePaddle Authors. All Rights Reserved. | |||
There was a problem hiding this comment.
FN_2d_solver_[HighOrder].py
FitzHugh- Nagumo RD 方程数据生成代码需要补齐
| np.array(ten_pred) - np.array(ten_true) | ||
| ) / functions.frobenius_norm(np.array(ten_true)) | ||
|
|
||
| print("The predicted error is: ", error) |
There was a problem hiding this comment.
error要求达到：源码，论文（1e-3）取较大值
| --8<-- | ||
| ``` | ||
|
|
||
| ### 3.7 模型训练与评估 |
|
|
||
| self.input_channels = input_channels | ||
| self.hidden_channels = hidden_channels | ||
| self.hidden_kernel_size = 3 |
| self.input_kernel_size = input_kernel_size | ||
| self.input_stride = input_stride | ||
| self.input_padding = input_padding | ||
| self.num_features = 4 |
| class loss_generator(nn.Layer): | ||
| """Loss generator for physics loss""" | ||
|
|
||
| def __init__(self, dt=(10.0 / 200), dx=(20.0 / 128)): |
| if __name__ == "__main__": | ||
| # grid size | ||
| M, N = 128, 128 | ||
| n_simu_steps = 30000 |
| layer_state_dict = paddle.load("output/phycrnet.pdparams") | ||
| model.set_state_dict(layer_state_dict) | ||
| model.register_output_transform(None) | ||
| output_graph(model, input_dict_val, fig_save_path) |
There was a problem hiding this comment.
input_dict_val进行修改，尝试完成时间变量t，从0s到4s的u_predict和u_true的对比，以及a-RMSE计算
|
暂时关闭 |
PR types
Others
PR changes
Others
Describe
PaddlePaddle/Paddle#57262
Loss精度
运行 200 次截图
torch
paddle