search

LSSTDESC / derp

A first attempt at a simple LSST DRP catalog emulator
BSD 3-Clause "New" or "Revised" License
1 stars 1 forks source link

Implement an analytic model #10

Open drphilmarshall opened 5 years ago

drphilmarshall commented 5 years ago

Lots of people in DESC have one of these. Francois pointed out this one:

"The 3x2pt pipeline also has an "emulator" module, to make a fake metacalibration catalog from DC2. Code is here: https://github.com/LSSTDESC/TXPipe/blob/master/txpipe/input_cats.py Any developments/improvements on that piece of code would fit right in the 3x2pt pipeline."

It would be good if TXProtoDC2Mock could somehow inherit from or call derp - the metacal extras could either be handled by the TXPipe code or just be new properties to emulate (we'd need training data though).

Joe Z followed up with:

"Yes - I was imagining we would make several versions of that [TXPipe] module, depending on:

  • where we are starting from (extragalactic catalog, DRP, completely fake data)
  • exactly what is being simulated - just shapes, or photometry + shapes
  • which shape method we are mocking (metacal requires all the variant columns)"

So, this could be a good place to start in implementing a first "analytic" model in derp. We'd want to get the PC and PZ groups looking at it early on, for comparison with their own analytic emulators.

drphilmarshall commented 5 years ago

Oops - this is a duplicate of #8 :-)

drphilmarshall commented 5 years ago

Actually, scratch that: I think a simple analytic model (including optional modular noise addition) could be seen as a useful first step towards #8. I'm thinking it would already be useful to predict Object table quantities using a deterministic model that is not fitted to any training data, but just captures the wisdom of astronomers to make a set of sensible predictions. The regressor in #8 would presumably then go one step further by automatically tuning the analytic model's magic numbers.