Oliver Schulz
Oliver Schulz
> you can also provide custom likelihood in Turing.jl Oh, right, that's a nice feature! I guess another important difference is that BAT explicitly distinguishes between prior and likelihood, and...
> if you don't use the DSL macro you have more control Sure, you can of course use the samplers that power Turing directly - which BAT does, e.g. AdvancedHMC....
> DynamicPPL you can distinguish between the prior and the likelihood by working Sure, I didn't mean that DynamicPPL doesn't have a concept of prior and likelihood, but you don't...
> is my understanding correct that in BAT.jl you have to define likelihood manually? Yes. However, now that we finally have DensityInterface.jl in place, I would like to establish packages...
CC @cscherrer
Yes, we can definitely make that configurable.
Until we make `bat_report` more configurable, you can use this workaround: ```julia BAT.smallest_credible_intervals(smpls, nsigma_equivalent = 2) ``` with `nsigma_equivalent` set to `1`, `2`, or `3` to get smallest 68%, 95%...
> If I am correct that this was supported in the previous version, it might be worth mentioning the current status of this type of interface in BAT.jl. That why...
Indeed, JLD2 can serialize most Julia structures natively, but is not a long-term data preservation format. BAT itself is intended to work with user-provided likelihoods that may contain arbitrary code...
> It would be nice to write an example interface for pyhf as well I actually discussed just that with @lukasheinrich today. :-) Having pyhf import (and possibly re-export from...