Owen L
Owen L
Are any of the new solvers compatible with weakly controlled terms? I've been using them a lot for certain problems where my noise is a constant so I don't want...
Or is this related to Patricks PR which does "I realised that it's not sufficient to have term_compatible_contr_kwargs have a single thing on the solver -- we need something with...
Can confirm, it runs off Patricks branch
I have a branch started for this, but I want to know the scope of the change you are looking for here. Having spent (a little) time thinking about it...
For context, the simplest sort of approach to 1. is shown in https://github.com/patrick-kidger/diffrax/pull/434 (needs more tests/docs/whatever but the core idea of just making minimal change to allow lineax in control...
I've actually encountered something very similar before, in which I needed the first MVP (the second wouldn't work). My solution was to make a sort of `filter_scan` and this worked...
Revived PR for this: https://github.com/patrick-kidger/equinox/pull/674
This should now be complete with the latest version https://github.com/patrick-kidger/equinox/releases/tag/v0.11.4
Hard to say exactly what the issue is (if you have the MVC that would be helpful) from this description. I usually do the neural network defined for a single...
That is what I meant. It was mostly a design pattern note, I wouldn’t expect any numerical differences