Weak SINDy#29
Conversation
|
One big change I may request is to make
|
|
|
||
| latent_dict = {'ae': Autoencoder} | ||
|
|
||
| ld_dict = {'sindy': SINDy} |
There was a problem hiding this comment.
I think workflow.py might change in future based on our discussion that the code will change toward a library of utility functions rather than a unified training framework.
So I won't leave much comments here, but wouldn't this require to include wSINDy?
| ld_type = config['latent_dynamics']['type'] | ||
| assert(ld_type in config['latent_dynamics']) | ||
| assert(ld_type in ld_dict) | ||
| latent_dynamics = ld_dict[ld_type](latent_space.n_z, physics.nt, config['latent_dynamics']) |
There was a problem hiding this comment.
I see that wSINDy takes different input parameters than this. So registering wSINDy in ld_dict will break this line anyway.
I may not ask for updating this function now, but we'll have to discuss about how to refactor workflow.py as soon as possible. We shouldn't leave it broken too long. maybe raise a RuntimeError for now if wsindy is chosen?
|
If you can write some description about this PR, that'd be great. |
3 participants
Add weak SINDy class and corresponding weak form derivatives