NeuralOperators.jl icon indicating copy to clipboard operation
NeuralOperators.jl copied to clipboard
verified publisher icon SciML

Mutli-dimensional input-grid for DeepONet

Open dynamic-queries opened this issue 3 years ago • 4 comments

In https://github.com/SciML/NeuralOperators.jl/blob/master/src/DeepONet.jl#L122-L124, multi-dimensional input grid is handled using an Argument error. Is there a reason this was done this way?

Also, if it is a matter of contracting AbstractArray(s), we could look into OMEinsum.jl for its implemention.

Please advise.

dynamic-queries avatar Jun 24 '22 11:06 dynamic-queries

multi-dimensional input grid is handled using an Argument error. Is there a reason this was done this way?

@Abhishek-1Bhatt Any ideas?

For contracting AbstractArray(s), this package uses Tullio.jl for it. Should work.

yuehhua avatar Jun 26 '22 03:06 yuehhua

If this is still unresolved. I have a quick fix with OMEinsum.jl Should I open a PR?

dynamic-queries avatar Jun 30 '22 13:06 dynamic-queries

Could you try setting it up with Tullio? That should make it easier for making it support GPUs.

ChrisRackauckas avatar Jun 30 '22 13:06 ChrisRackauckas

It is welcome to support multi-dimensional input for DeepONet. However, it will be nice to use Tullio to make deps simple.

yuehhua avatar Jun 30 '22 14:06 yuehhua