diagnose iSpeculator
changhoonhahn opened this issue · comments
There are currently discrepancies between the true galaxy properties and properties inferred from iSpeculator.
The above comparison is for the first 10 LGal galaxies.
This discrepancy can be caused by the following:
- bug in the code
- MCMC not converged
- model limitation (SFH bases, dust)
We need to diagnose the cause of this discrepancy.
The discrepancy in SFRs may also be due to priors on SFR (or SFH) set by the Dirichlet priors. See notebook for how the warped manifold transform used to sample the Dirichlet priors.
In this notebook I compared the observed photometry and spectroscopy to model(median of posterior) and find good agreement. This suggests that the discrepancy is likely not due to a bug or MCMC convergence issue (this still needs more testing).
I was able to reproduce the same results in the both notebooks on NERSC with my emulator data
@kgb0255 which results are you referring to?
@changhoonhahn Sorry for lack of clarity. I sshed to NERSC to access my data and was able to reproduce the same output in both notebooks you linked above - sfh_basis_transform, diagnose_ispeculator.ipynb.
Closing this issue since this falls under #52