Post by asgr on Jan 21, 2021 12:34:02 GMT 8
Hi All,
It’s been a little while since I posted anything related to ProFit (the main purpose of this exploder). There has been a lot of work on ProFound and ProSpect over the last year, and since I got back after the break I finally got round to some big-ish ProFit updates I’ve had at the back of my mind for a while.
The short version is that ProFit (and by proxy, I) was invited to join the current Euclid morphology challenge. This involved attempting decompositions for about 1.2 million mock targets on large mock Euclid frames (20k x 20k). To do this in a sensible manner I’m making use of Rfits’ capacity to read image subsets from FITS pointers (so only the part that is needed is in memory, and one on disk file can be shared between nodes and cores). On top of that I decided to add some very high level interfaces (most significantly profitDoFit) to more rapidly and almost entirely automatically fit galaxies with the most popular decomposition methods, namely:
Moffat (stars)
Single Sersic
1.5 component (PSF bulge, Sersic disk)
2 component (Sersic bulge, Sersic disk)
As well as fitting the target object, the new functions will also take care of all the busy work of co-fitting nearby sources that might produce enough extended flux to influence our solution. This part had always been a bit of pain for users (me included) and became critical for the Euclid work, so that motivated getting it working painlessly.
The profitAllStarDoFit function also make it very easy to create PSF models from a target image, automatically finding and simultaneously fitting stars on the frame.
The above both use my new-ish Highlander package (actually developed for ProSpect SED fitting) to rapidly achieve good fits with useful errors using a hybrid of a genetic algorithm and CHARM based MCMC to optimise. This is *much* faster than the previously suggested methods for getting reasonable fits- for the example galaxies you can get good solutions with errors in about a minute. If you want longer posterior chains this is still possible of course.
A vignette on how to use these new functions is already online at:
rpubs.com/asgr/715082
You can get the key packages from GitHub at:
ICRAR/ProFit
asgr/ProFound
asgr/Highlander (which also needs asgr/cmaeshpc)
Happy fitting!
Cheers,
Aaron
It’s been a little while since I posted anything related to ProFit (the main purpose of this exploder). There has been a lot of work on ProFound and ProSpect over the last year, and since I got back after the break I finally got round to some big-ish ProFit updates I’ve had at the back of my mind for a while.
The short version is that ProFit (and by proxy, I) was invited to join the current Euclid morphology challenge. This involved attempting decompositions for about 1.2 million mock targets on large mock Euclid frames (20k x 20k). To do this in a sensible manner I’m making use of Rfits’ capacity to read image subsets from FITS pointers (so only the part that is needed is in memory, and one on disk file can be shared between nodes and cores). On top of that I decided to add some very high level interfaces (most significantly profitDoFit) to more rapidly and almost entirely automatically fit galaxies with the most popular decomposition methods, namely:
Moffat (stars)
Single Sersic
1.5 component (PSF bulge, Sersic disk)
2 component (Sersic bulge, Sersic disk)
As well as fitting the target object, the new functions will also take care of all the busy work of co-fitting nearby sources that might produce enough extended flux to influence our solution. This part had always been a bit of pain for users (me included) and became critical for the Euclid work, so that motivated getting it working painlessly.
The profitAllStarDoFit function also make it very easy to create PSF models from a target image, automatically finding and simultaneously fitting stars on the frame.
The above both use my new-ish Highlander package (actually developed for ProSpect SED fitting) to rapidly achieve good fits with useful errors using a hybrid of a genetic algorithm and CHARM based MCMC to optimise. This is *much* faster than the previously suggested methods for getting reasonable fits- for the example galaxies you can get good solutions with errors in about a minute. If you want longer posterior chains this is still possible of course.
A vignette on how to use these new functions is already online at:
rpubs.com/asgr/715082
You can get the key packages from GitHub at:
ICRAR/ProFit
asgr/ProFound
asgr/Highlander (which also needs asgr/cmaeshpc)
Happy fitting!
Cheers,
Aaron