search

omniscientoctopus / Physics-Informed-Neural-Networks

Investigating PINNs
MIT License
519 stars 132 forks source link

Helmholtz pytorch predicted problem #4

Open FemLearner opened 1 year ago

FemLearner commented 1 year ago

I try the tensorflow version of the Helmholtz problem and it was working correctly: image

But when I try the version of pytorch , and whatever i did(changing the layers or iter) ,the result was bad , if there is some parameters make this problem happen?I checked out the code and I think there is nothing wrong?
image

aho38 commented 5 months ago

I fixed it if you wanna check out mine (https://github.com/aho38/Pytorch-PINN-Marine-Lake/tree/main/Data_Discovery). The issue was in autograd of the parameters. It's computing the wrong input.