Autoencoders with Pysindy

[buathukorala]

Is your feature request related to a problem? Please describe.

Hello, I was wondering whether you could include an example (may be from the paper data-driven discovery of coordinates and governing equations) that showcases the use of autoencoders with Pysindy package.

Thank you, Bhagya

Describe the solution you'd like

Describe alternatives you've considered

Additional context

[akaptano]

Hi Bhagya,

We don't currently support tensorflow as a dependency, but it would be great if you or someone else wants to use tensorflow and pysindy together in a script to show how it is done. The codes could be fairly separate... for instance, the auto encoder would take some data and build some low (or high) dimensional feature space, and then you could use PySINDy on this feature space. However, some algorithms call for minimizing the PySINDy loss and the auto encoder loss at the same time (see, for instance, Kathleen Champion's paper on this) and that would require integration of tensorflow in a complicated way into the PySINDy code. Goals for the future perhaps.

Best, Alan

[buathukorala]

Hi Alan, Thank you very much for the reply. I will check Kathleen Champion's paper and the corresponding GitHub repo. I don't think I'm an expert in python :), however, I will try to use TensorFlow and pysindy together to come up with an example.

Thank you again, Bhagya

[znicolaou]

Agreed, it would be really great if somebody made an example with the autoencoders. We'll need to do it eventually, but it isn't on my immediate agenda yet. There are a few papers (some published, some under review) in the Kutz/Brunton group that will have repositories to model after including those of Kathleen Champion and Joseph Bakarji. I think this will be a good summer project for somebody.

[buathukorala]

Thank you @znicolaou . I will keep eye on the upcoming papers as well.