search

HERA-Team / hera_cal

Library for HERA data reduction, including redundant calibration, absolute calibration, and LST-binning.
MIT License
11 stars 8 forks source link

Compare new omnical (in redcal) to old omnical #364

Closed jsdillon closed 6 years ago

jsdillon commented 6 years ago

We should make sure that they are giving the same results.

jsdillon commented 6 years ago

After some issues reproducing the results (it seems that some of the visibilities in the omnical model visibility files have are complex conjugated when they shouldn't be), I managed to get basically matching solutions with identical chisq, computed via the same code (the one in hera_cal.utils).

screen shot 2018-09-13 at 2 22 53 pm

However, a detailed look reveals that their chisq differ at the ~0.2% level:

screen shot 2018-09-13 at 2 24 12 pm

I imagine that this has to do with different weightings (my chisq uses a uniform weighting) or different convergence criteria.

For now, I think this is good enough to go forward, especially given the other improvements seen in http://reionization.org/wp-content/uploads/2013/03/HERA_memo050_Omnical-Convergence.pdf, but it may be worth further investigation.