XuhuiM / PPINN

Demo code for PPINN paper: https://www.sciencedirect.com/science/article/pii/S0045782520304357
11 stars 2 forks source link

问题 #1

Closed hyw-freeway closed 1 year ago

hyw-freeway commented 1 year ago

你好。 我是一名来自中国的研究生。最近,我阅读了您的论文PPINN: Parareal physics-informed neural network for time-dependent pde。我有几个疑问,如果您能抽出宝贵的时间来解释,我将不胜感激。 对于pinn模型,用训练数据训练后,是否需要将数据分为测试数据和验证数据来验证模型的有效性?在您的论文中,训练的模型数据似乎直接与其他解析方程的结果进行了比较。 再次感谢大家的聆听。如果你能回答我的问题,我将非常感激。祝你生活幸福!

XuhuiM commented 1 year ago

PPINN is developed for solving forward PDE problems in parallel. For forward problems, we only have training data on boundary/initial conditions. We compare the PINN predictions with the results from numerical solvers after training. We do not use validation data in the forward problems.

hyw-freeway commented 1 year ago

Thank you very much for your answer, which is very useful to me! Thank you, have a nice life!

------------------ 原始邮件 ------------------ 发件人: "XuhuiM/PPINN" @.>; 发送时间: 2023年2月17日(星期五) 上午8:04 @.>; @.**@.>; 主题: Re: [XuhuiM/PPINN] 问题 (Issue #1)

PPINN is developed for solving forward PDE problems in parallel. For forward problems, we only have training data on boundary/initial conditions. We compare the PINN predictions with the results from numerical solvers after training. We do not use validation data in the forward problems.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

hyw-freeway commented 1 year ago

I'm sorry to bother you again. What you mean by "we only have training data on boundary/initial conditions" is that the input dataset is the boundary conditions and the initial data at the time of t=0, not include the y value of the whole time T, right? The y value of each time step trained by the neural network is compared with the y value of the standard solution of the corresponding time step to evaluate the performance of the model.

------------------ 原始邮件 ------------------ 发件人: "胡玉炜" @.>; 发送时间: 2023年2月17日(星期五) 上午9:24 @.>;

主题: 回复: [XuhuiM/PPINN] 问题 (Issue #1)

Thank you very much for your answer, which is very useful to me! Thank you, have a nice life!

------------------ 原始邮件 ------------------ 发件人: "XuhuiM/PPINN" @.>; 发送时间: 2023年2月17日(星期五) 上午8:04 @.>; @.**@.>; 主题: Re: [XuhuiM/PPINN] 问题 (Issue #1)

PPINN is developed for solving forward PDE problems in parallel. For forward problems, we only have training data on boundary/initial conditions. We compare the PINN predictions with the results from numerical solvers after training. We do not use validation data in the forward problems.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

XuhuiM commented 1 year ago

Yes.

On Sat, Feb 18, 2023 at 12:56 PM huyuwei @.***> wrote:

I'm sorry to bother you again. What you mean by "we only have training data on boundary/initial conditions" is that the input dataset is the boundary conditions and the initial data at the time of t=0, not include the y value of the whole time T, right? The y value of each time step trained by the neural network is compared with the y value of the standard solution of the corresponding time step to evaluate the performance of the model.

------------------ 原始邮件 ------------------ 发件人: "胡玉炜" @.>; 发送时间: 2023年2月17日(星期五) 上午9:24 @.>;

主题: 回复: [XuhuiM/PPINN] 问题 (Issue #1)

Thank you very much for your answer, which is very useful to me! Thank you, have a nice life!

------------------ 原始邮件 ------------------ 发件人: "XuhuiM/PPINN" @.>; 发送时间: 2023年2月17日(星期五) 上午8:04 @.>; @.**@.>; 主题: Re: [XuhuiM/PPINN] 问题 (Issue #1)

PPINN is developed for solving forward PDE problems in parallel. For forward problems, we only have training data on boundary/initial conditions. We compare the PINN predictions with the results from numerical solvers after training. We do not use validation data in the forward problems.

— Reply to this email directly, view it on GitHub, or unsubscribe. You are receiving this because you authored the thread.Message ID: @.***>

— Reply to this email directly, view it on GitHub https://github.com/XuhuiM/PPINN/issues/1#issuecomment-1435484487, or unsubscribe https://github.com/notifications/unsubscribe-auth/ALMCURRDCMA4ILVHPZBOZVTWYBJAPANCNFSM6AAAAAAU57A5PY . You are receiving this because you commented.Message ID: @.***>