docs! #29
Assignees
Labels
organization
documentation and clean code
Comments
joshspeagle
added
organization
joshspeagle
added a commit
that referenced
this issue
Aug 4, 2019
This commit adds in a number of planned features before the first big `brutus` production run in the upcoming week over ~80M objects (!!!). A summary of changes is below. Note that this commit also includes a number of updates to the datafiles needed to run the code in the Dropbox folder (linked through the README), which should hopefully finish uploading sometime during the week. Clusters -------- - The baseline isochrones used to generate these had an error that caused weird gaps in coverage. I think I've resolved this, and the isochrone file has been updated accordingly. - The `clusters` module now contains a pretty hefty log-likelihood function that can be used to fit isochrones to co-spatial, co-eval populations of stars. I don't have any documentation of this quite yet, but I've confirmed in testing that this does actually work! I want to build this out in the future, but for now I just wanted to push this since it resolves #26. MIST Models ----------- - Using a series of ~solar and sub-solar benchmark clusters, I have derived improved corrections to the underlying SEDs on the lower-MS and associated zero-points. These are now implemented natively in the various SED generation tools by default, and so do not necessitate any more additional correction files. This resolves #27. Binaries -------- - I have improved flexibility for modeling binaries when generating individual SEDs and isochrones. These now allow users to specify where binaries are allowed and expand the default range. Priors ------ - Added in the new 3-D dust map from Bayestar19, which is substantially higher resolution than Bayestar17 and looks very nice. - Modified the default behavior of the dust prior based on the results from Green et al. (2019) so that we are now by default only adding scatter, not multiplying the existing scatter. - Added support for an age prior that whose behavior tracks our metallicity priors in the disk and the halo. Per-Object Fitting ------------------ - Cleaned up internals so things are a bit better organized (at least for me). - Modified defaults to be more aggressive on adaptive thresholding and save fewer samples to improve runtime. - Changed how parallaxes are used so they behave more like an additional "band" wherever possible, rather than being seen as a "special" additional piece of information. This is propagated through to various parts of the code and to the saved quantities (including the effective `Nbands`). - Modified outputs so that names now are easier to understand. - Added log-posteriors to outputs (both per-object and for individual draws). **THIS TOTALLY BREAKS BACKWARDS COMPATIBILITY.** I don't expect to have to substantially change the output format again, but I guess it's stil below v1 for a reason! Finally, here's a list of upcoming updates in the following week: - A new model grid and zero-point file will be uploaded after updated Gaia passbands are added in and I can run additional systematics tests on high-latitude, low-reddening fields. - Hopefully I can start trying to put together documentation (see #29) since there is now enough features in the code that people who aren't me will probably struggle even with all the excessive docstrings I add in.
|
This is now slated to be part of my Gaia EDR3 hack week project, along with getting the paper out the door. |
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
It's reaching the point where I need to start putting in documentation. If I can get anything half as good as what I have with
dynesty, I'll consider that a win.The text was updated successfully, but these errors were encountered: