Rory Conlin
Rory Conlin
> Should put a note in the docstring of _Objective of what `"auto"` for `jac_chunk_size` does now (uses a conservative heuristic estimate based on problem size) "auto" is still only...
+1 for adding this feature, it would be super useful in several projects for me.
> Ah, this is an easy one; we unconditionally use throw=True in the backward pass. (We have nowhere to pipe the error to.) On this note, I think it would...
Yeah I'm 99.9% sure using `jnp.inf` would be fine, I don't think it needs any `where` etc. The only possible issue I could see is if using `jax.debug_infs` you might...
This is what I had way back when: https://github.com/PlasmaControl/DESC/blob/v0.2.0/desc/gfile_helpers.py It's probably not super useful, I don't think it did any conversion between DESC and EFIT, it was just for plotting...
Just added a barebones function for doing this in #732 Preferably someone else can double check the units etc and make a testcase for it.
Yeah I don't see how this actually fixes the underlying issue in #1686. From the plots above it looks like this doesn't get rid of those spikes which are likely...
Also the plots only show a reduction of ~500 MB, shouldn't it be a lot more?
Could do something like this: https://github.com/PlasmaControl/DESC/blob/master/desc/equilibrium/coords.py#L106-L107
Also don't forget to add the new objective to the docs