SGA-2024 algorithmic plans

[moustakas]

Starter ticket for the SGA-2024.

Working outline:

• Part 1:

  • Train & validate a neural net to recognize what large galaxies "look" like, using the SGA-2020 images, blob masks, etc.
  • Bonus: include labels pertaining to morphology, color, environmental density (e.g., mergers, galaxy clusters).
  • Of possible interest / relevance:
  • https://arxiv.org/pdf/2012.13083.pdf and
  • https://georgestein-galaxy-search-streamlit-app-o7accj.streamlit.app/

• Part 2:

  • Build early coadds from custom / arbitrary central "brick" positions.
  • Run legacypipe source detection & segmentation with a variety of kernels optimized to find large galaxies, including low surface-brightness systems.
  • Use the trained neural net to test all the candidate source positions.
  • Rinse and repeat.

• Part 3:

  • Use source injection to test the completeness and purity of the algorithm as a function of various "classes" (color, morphology) of large galaxies.

• Part 4:

  • Generate ellipse modeling / surface-brightness profiles using GALEX + grz + unWISE imaging.

[moustakas]

Here's a nice test case of a "known" SGA-2020 galaxy (PGC1054942) adjacent to an object which was missed: https://www.legacysurvey.org/viewer-dev/?ra=15.8299&dec=-4.6714&layer=ls-dr9&zoom=15&bricks&sga https://www.legacysurvey.org/viewer/jpeg-cutout?ra=15.8232&dec=-4.6630&width=600&height=600&layer=ls-dr9

salloc -N 1 -C cpu -A desi -t 01:00:00 --qos interactive --image=legacysurvey/legacypipe:DR10.1.2

SHIFTER=docker:legacysurvey/legacypipe:DR10.1.2
shifterimg pull $SHIFTER
shifter --image $SHIFTER bash

export LEGACY_SURVEY_DIR=/global/cfs/cdirs/cosmo/work/legacysurvey/dr9
export SKY_TEMPLATE_DIR=/global/cfs/cdirs/cosmo/data/legacysurvey/dr9/calib/sky_pattern
export GAIA_CAT_DIR=/global/cfs/cdirs/desi/target/gaia_edr3/healpix
export GAIA_CAT_PREFIX=healpix
export GAIA_CAT_SCHEME=nested
export GAIA_CAT_VER=E
export UNWISE_COADDS_DIR=/global/cfs/cdirs/cosmo/data/unwise/neo7/unwise-coadds/fulldepth:/global/cfs/cdirs/cosmo/data/unwise/allwise/unwise-coadds/fulldepth
export TYCHO2_KD_DIR=/global/cfs/cdirs/cosmo/staging/tycho2
export PS1CAT_DIR=/global/cfs/cdirs/cosmo/work/ps1/cats/chunks-qz-star-v3
export DUST_DIR=/global/cfs/cdirs/cosmo/data/dust/v0_1
export GALEX_DIR=/global/cfs/cdirs/cosmo/data/galex/images

export LEGACYPIPE_CODE_DIR=/src/legacypipe
# or, optionally, add a local check-out to your path
# export LEGACYPIPE_CODE_DIR=/path/to/my/legacypipe
# export PYTHONPATH=$LEGACYPIPE_CODE_DIR/py:$PYTHONPATH

# these will need to be determined algorithmically
export RUN=south 
export BRICK=015823m04663
export OUTDIR=$SCRATCH/SGA-2024/$BRICK

python $LEGACYPIPE_CODE_DIR/py/legacypipe/runbrick.py --radec 15.8232 -4.6630 \
  --width 600 --height 600 --pixscale 0.262 --bands g,r,z --run $RUN --threads 12 \
  --outdir $OUTDIR --pickle "$OUTDIR/runbrick-custom-$BRICK-%%(stage)s.pickle" \
  --no-unwise-coadds --stage srcs