ref data for hNd pNd - Githubissues

i207M / PINNacle

Codebase for PINNacle: A Comprehensive Benchmark of Physics-Informed Neural Networks for Solving PDEs.

https://arxiv.org/abs/2306.08827

MIT License

236 stars 38 forks source link

ref data for hNd pNd #4

Closed VhaoYan closed 11 months ago

VhaoYan commented 11 months ago

Hi, could you please also provide your ref data for high dim?

i207M commented 11 months ago

Sure. Actually we've designed the two high-dimensional problems to have analytical solutions. Please refer to these two sections of code: https://github.com/i207M/PINNacle/blob/d36c53f1c373fe1ad632a2a3b3e6ee978cbcfb79/src/pde/heat.py#L280C13-L280C20 https://github.com/i207M/PINNacle/blob/d36c53f1c373fe1ad632a2a3b3e6ee978cbcfb79/src/pde/poisson.py#L302C19-L302C19 Feel free to comment if you have additional questions.

VhaoYan commented 11 months ago

I was wondering: 1: did you use the lbfgs alone, or adam_lbfgs for optimization? In your paper I found only states lbfgs. A bit confused. 2: is below the hyperparameters for all the equations? "[adam_param={'lr': 1e-3, 'betas': (0.9, 0.999)}, lbfgs_param={'lr': 1, 'max_iter': 20]" or you've also optimized the hypers?