Peter Melchior

In other words, you don't need Lipschitz constants for the prior, simply computing the grad is sufficient

Thanks, this looks pretty good. Do you have the trained pixelCNN for us to play with? Am not (yet?) a fan of putting tf code into scarlet's main classes. I'd...

You're doing it correctly now. ExtendedSource has an implementation that predates Observation, so we haven't closed the loop. One concern is that weight is an optional argument, but bg_rms is...

Beside the weight vs bg_rms discussion we'll address in #129, and improve the packaging and dependencies later on, there are important changes we need. * The prior needs to be...

@fred3m is there a way to allow the data download from AWS or would it be better to make a static download for Silke?

The solution we came up is to open-source the test data in the form of a git-lfs repo, so everyone (incl us) who needs it can simply clone the repo.

I was so expecting you to come up with such an idea! Give it a try!

This method is now class `ThresholdConstraint`, but remains largely untested.

So, to clarify, we run the regression test through Jarvis and only report the results to SQuaRE? I'm worried about Jarvis timeouts.

Do you want to trigger this for every branch or only after when merging to master? I can see waiting for 1h until Jarvis comes back as quite annoying