Open gnarayan opened 2 years ago
@rknop and I had discussed building the ELAsTiCC metrics with proclam. Is this no longer the plan?
Talking to Gautham, the session on Friday is more going to be a workshop for people who want to access the database and pull down the information in order to construct their own metrics rather than building any built-in code for metrics on the TOM itself.
In addition to updating proclam to Python 3, we ought to make the metrics a little more flexible by separating the per-class averaging from the per-lightcurve metric calculations.
Also, ping me if you want access to the repo and don't already have it!
I'm sure many/all of you already know this, but just in case you were planning on doing it by hand: py2to3
is a tool for automatically updating python2 to python3. DM did this many years ago, and I can dig out the best-practice instructions for actually running it that we came up with, if that would be helpful.
Yeah, 2to3 is fine to start - there will be some numpy/matplotlib calls that have changed, but we should start with 2to3. Re. separating per-lightcurve from per-class - I agree we want this, but this is more than a day. I want to focus on just getting one classifier's results from one broker through fixed up proclam, and that's a good start for now.
Here are the slides I put together as an intro. It has links to URLs for various resources if you're interested in poking into the database of elasticc broker messages and alerts
There is now a DESC fork of proclam which will hopefully make permissions management easier.
Implementing ELAsTiCC Metrics
We will have ELAsTiCC Alerts ingested into FAST DB, as well as mock submissions from broker/classifier teams, and we will start to implement SQL/python to write evaluation metrics
Contacts: Time: All Day Main communication channel: #plasticc-public GitHub repo: https://github.com/lsstdesc/elasticc In-person/Virtual/Hybrid: Hybrid Pitch: https://docs.google.com/presentation/d/10jD12cBNtuuph_zO2EAuRWpQLybCl4ALCuudtioYk2U/edit#slide=id.g142601f6059_5_0 Zoom room (if applicable): Use the usual ELAsTiCC zoom - pinned in #plasticc-public
Goals and deliverable
Plots + metrics tables comparing a few mock classifiers on different criterion
Resources and skills needed
Python and ideally some SQL