search

IvanNikolic21 / Lyman-alpha-bubbles

Codes used to derive the likelihood calculation for finding an ionized bubble given some Lyman-emitter data
0 stars 0 forks source link

Fixing the bug that lead to biased inference with the new speed-up #7

Closed IvanNikolic21 closed 6 months ago

IvanNikolic21 commented 6 months ago

There is a bug that leads to incorrect inference when using the new code which pre-computes forward-models. First attempt is to check whether there is a shift in the order of galaxies because of poor parallelization.

IvanNikolic21 commented 6 months ago

Note: The problem doesn't seem to be with the parallelization. Looking at the cache, it also seems alright. There is not an obvious problem in sight, so running a resolution check is the best way to proceed with some results expected on Friday.

IvanNikolic21 commented 6 months ago

The tau upgrade led to the correct likelihood on tau, but not on other likelihoods. Checking them out now.

IvanNikolic21 commented 6 months ago

One update for tomorrow's analysis. Getting the likelihood directly in jupyter doesn't reproduce the bug. This could indicate something wrong with the sampler or functions that are different.

IvanNikolic21 commented 6 months ago

The bug has been solved and my inference finds the correct result! Woohoo