Replies: 2 comments 2 replies
-
I think you should post the loss_history, and try more |
Beta Was this translation helpful? Give feedback.
2 replies
-
Hi @rayannott, I have worked with some SIR models before but mostly for inverse problems. I usually use 3 layers with 32-64 neurons and relu as activation. Also you could try to add some input transform, Dr. Lu has a good example in this repo https://github.com/lu-group/sbinn I hope it helps! |
Beta Was this translation helpful? Give feedback.
0 replies
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
-
I picked the original paper for review during my masters seminar "Deep Learning in Physics" at the Technical University of Munich. I wanted to try out some other equations and systems, but all of them failed to match the "exact" numerical solution (RK45).
This problem is similar to #1583
For example, here is my code for the SIR-model simulation.
This sets up the reference solution using
scipy.integrate.solve_ivp
.This is the model creation and training:
The loss looks great, but the solution doesn't match the reference:
Beta Was this translation helpful? Give feedback.
All reactions