jacaseyclyde / pop-classify Goto Github PK

we'll have to work on fake data (or just any star data) until we build the SQL query to select stars with classes, but we can start implementing this now

we'll have to work on fake data (or just any star data) until we build the SQL query to select stars with classes, but we can start implementing this now

ROC curves can be used to evaluate the performance of each classifier by examining the true positive rate as a function of false positive rate for each classifier. the closer the curve is to the upper left corner, the better it is. see Ivezic 395/396 for more info/examples

Completeness-efficiency curves are another useful tool in characterizing ML methods. they can tell us about things like the completeness (e.g. % of the overall true population of class A that is getting correctly classified as class A) vs the efficiency (the % being classified as class A that are actually class A). More info in Ivezic 395/396

we'll have to work on fake data (or just any star data) until we build the SQL query to select stars with classes, but we can start implementing this now

Like the title says, we need to pull down stellar data from SDSS. Not sure how much data we need to pull down just yet, but we need to grab data from stars with spectra and a valid subclass (selected from the subclasses on this page http://www.sdss.org/dr14/spectro/catalogs/). Still not sure what data we should pull down, beyond color/subclass. Might depend on exactly what's available to us.

we'll have to work on fake data (or just any star data) until we build the SQL query to select stars with classes, but we can start implementing this now

We should cross check which data is available to us in both photometric-only and spectroscopic point sources, so that we don't train using data not available in photometric.