search

NeuroDiffGym / neurodiffeq

A library for solving differential equations using neural networks based on PyTorch, used by multiple research groups around the world, including at Harvard IACS.
http://pypi.org/project/neurodiffeq/
MIT License
680 stars 89 forks source link

Use a single network for ODE/PDE systems #154

Closed Edward-CY closed 2 years ago

Edward-CY commented 2 years ago

Thanks for your great job! I am learning to solve systems of odes. After I read the docs, I find that the systems of odes is solved by many 1NNN1 networks parallelly. I wonder if this package can solve odes by one network whose number of output units equal to the number of functions. And which method, you think, is better. Thank you!

shuheng-liu commented 2 years ago

There is a single_net feature in neurodiffeq but it was deprecated and we have stopped working on it a long time ago. For toy problems, we didn't see much of an boost by doing it this way. I can take some time to update it to work with the latest version of neurodiffeq, but there's no ETA on it.

If you believe a single network will outperform multiple networks each with a single output unit, please let us know and we can discuss further.

Edward-CY commented 2 years ago

Thanks for your reply!