search

pdebench / PDEBench

PDEBench: An Extensive Benchmark for Scientific Machine Learning
https://arxiv.org/abs/2210.07182
Other
724 stars 86 forks source link

Configuration file for FNO on SWE equation #52

Closed HaoZhongkai closed 11 months ago

HaoZhongkai commented 11 months ago

Dear authors, Thanks for the nice benchmark as well as the data! I have a question about using FNO for shallow-water equation 2D (SWE). It seems there is only a configuration file for PINNs on this PDE. I want to know in which file is the configuration file or hyperparameters of FNO on SWE 2D?

best regards

kmario23 commented 11 months ago

Hi @HaoZhongkai, Thank you for your kind interest in our benchmark and bringing this issue to our notice. I have added the support for training an FNO with SWE-2D dataset .

The corresponding config file is now located at pdebench/models/config/config_rdb.yaml, but this is just for convenience (see original here: pdebench/models/config/args/config_rdb.yaml ).

Download the Radial Dam Break data to an appropriate path, update its absolute path in the config file pdebench/models/config/config_rdb.yaml#L22, then do:

$ python train_models_forward.py

You can adjust the desired parameters such as spatial and temporal subsampling, number of modes in FNO, batch size, lr, how many initial steps you wanna use as initial condition (i.e., initial_step) etc., in the config file.