Astrobase

Fixes

- services: fixed infinite mirror-hop loop when submitting queries for
SIMBAD/GAIA, these should now bail out after `maxtries` number of submission
tries are reached.
- cpserver: fixed some issues with frontend JS breaking if GAIA xy positions are
not available for some reason

New stuff

- cpserver: added a `baseurl` kwarg to `checkplotserver` so one can launch
multiple instances of it and have them go to separate base URLs if
reverse-proxying to external users.
- `plotbase.plot_phased_mag_series`: allow output to an existing
`matplotlib.axes.Axes` object


Changes

- cpserver: checkplot filenames are now URI encoded in requests to
`checkplotserver`


Fixes

- cpserver: better handling of nans that may break JSON provided to the
frontend's javascript handlers from the python backend.
- `coordutils.xmatch_kdtree`: use correct order of `query_ball_tree` args
- `lcproc.collect_tfa_stats`: better handling of all-nan magcol arrays

New stuff

- checkplot: add `update_checkplot_objectinfo` function to retroactively update
checkplots with new information after they've been created
- services: added a SIMBAD TAP query client
- services.gaia: added `complete_query_later` kwarg to save still-running query
info if a timeout is reached. the next invocation of the same query will check
if it completed and will retrieve the results
- varclass.starfeatures: get all possible external object IDs from SIMBAD
- checkplot, lcproc, starfeatures: add `gaia_max_timeout`, `gaia_mirror`, and
`complete_query_later` kwargs to query and checkplot related functions


Changes

- services.gaia: now uses DR2
- services.gaia: now figures out which GAIA mirrors to use automatically, uses
CDS by default
- services.gaia: can now skip to another GAIA mirror if the current one doesn't
work. mirrors enabled are the GAIA main site, the Heidelberg mirror, and the
CDS mirror
- `lcproc.add_cpinfo_to_lclist`: add magcols to the output dict and run SIMBAD
query again if necessary
- `checkplot._pkl_finder_objectinfo`: allow direct use of dict as lclistpkl kwarg


Fixes

- checkplot: actually retry finder chart with forcefetch=True
- cpserver.js: handle errors correctly on checkplot load
- cpserver.js: more failure handling
- kbls: new chunk-size determination scheme, should fix Github 50
- services.gaia: automatically figure out table names for mirrors
- services.gaia: fix missing import, tests: use new expected GAIA ID for DR2
- services.gaia: handle timeouts and 503s a bit better
- services.gaia, skyview: add handling for corrupt cached results
- services.simbad: enforce random wait time between retries
- starfeatures: fix missing simbad_result var
- tests: more fixes for GAIA DR2
- `varfeatures.nonperiodic_lightcurve_features`: fix ndet inconsistency

New stuff

- checkplot, checkplotserver: handle cases of no PF results, also PF results w/
only 1 phased LC, allows making checkplots for objects with no period-finder
results just to see their mags, colors, finder chart, and GAIA info
- lcproc.runcp: allow empty pfpickle input for CPs w/o phased LCs
- lcproc: new `add_cpinfo_to_lclist` function to add back checkplot info to the
object catalog generated by `lcproc.make_lclist`
- lcproc, plotbase: add `overlay_zoomcontain` kwarg for `fits_finder_chart`
function to zoom finder charts automatically to the footprint of all annotated
objects in the overlay
- services.gaia: now randomly picks between GAIA archive mirrors and can skip
dead ones automatically

Changes

- checkplot: finder reticle now at radec -> xy of object instead of center of
chart
- services.gaia: return query params for info after an HTTPError

Fixes

- services.gaia: standardize ra, dec etc strings to make cache hits more
probable
- lcproc.runcp: workaround for list of pfmethods if not provided in the
period-finding result pickle
- cps UI: fix overview LC tile widths if no PF methods
- checkplot: handle 'objectid' not found in HAT LC objectinfo dicts

New stuff

- lcproc: added TFA (Kovacs et al. 2005;
http://adsabs.harvard.edu/abs/2005MNRAS.356..557K) implementation and parallel drivers
- lcproc: added EPD drivers for LC collections


Changes

- lcproc, checkplotlist: make nworkers depend on actual number of CPUs
- lcproc: add minobservations kwarg for `runcp`, `parallel_cp`, `parallel_cpdir`


Fixes

- lcproc: fix addressing of magcols so stuff like epd.mags, tfa.mags works
- lcproc.lcbin: write results to correct outdict keys based on magcols
- lcfit: fix missing imports
- `lcfit.spline_fit_magseries`: fix annoying 'x must be strictly increasing'
thing
- checkplot: fix axes -> ax typo

New stuff

- `checkplot.checkplot_pickle_to_png`: can now export to StringIO or BytesIO
instance in addition to a PNG filename.
- `checkplot`/`lcproc`: added a `maxnumneighbors` kwarg to control the number of
neighbors the function retrieves around the target object (added by
joshuawallace).
- cpserver UI: added object neighbor color and mag diffs to the UI if available
- cpserver UI: zoom effects for phased LC tiles to make reviewing easier
- `lcproc.make_lclist`/`filter_lclist`: can now make overlayed FITS finder
charts if a FITS file for the object catalog is provided
- `plotbase`: added a `fits_finder_chart` function to make an annotated finder
chart from a given FITS file with WCS information

Changes

- cpserver UI: finder chart reticle appearance changed slightly to obscure
nearby stars less
- cpserver UI: moved the 'Save to PNG' button to the overview tab so it's more
obvious
- `checkplot:checkplot_pickle`: can now fall back to GAIA proper motions if
'pmra' and 'pmdecl' are not provided in the input `objectinfo` dict kwarg
- cpserver: now gzips the response (helps if the checkplotserver is running
somewhere else other than localhost)
- cpserver: PNG export for a checkplot is now done in memory

Fixes

- `checkplot.checkplot_pickle_to_png`: fix py2 StringIO isinstance() issues
- cpserver: fix broken handling of missing nbr mag and color diffs
- cpserver UI: fix the CSV/JSON export to correctly include all previous
reviewed objects instead of just the ones from the current session
- cpserver: readonly mode now disables all input correctly
- cpserver UI: fix checkplot list traverse breaking at ends of list
- kbls.bls_snr: fixed transit duration calculation, pointed out in 45
- lcproc: fix readerfunc apply if sequences are returned