Weak lensing: updated baseline shear estimator + manuscript figures

[timwhite0]

• Regenerated catalog after DC2 files were re-synced. Added ellipticity_1_true_dc2 and ellipticity_2_true_dc2 to the catalog.
• Estimated galaxy ellipticities using LSST-measured second moments Ixx_pixel, Ixy_pixel, Iyy_pixel. Switched to using these LSST estimates in our baseline shear estimator instead of the true ellipticities.
• Updated tuning procedure for the baseline shear estimator. We now use only the training set to tune (I was previously using all DC2 images by mistake), and we use R-squared to select the optimal weighted-average kernel size and standard deviation.
• Created some figures for the manuscript.

See evaluate_encoder.ipynb and manuscript/figure*.ipynb for recent results. Here's a summary from a fairly recent training run:

BLISS:
test loss: -10.34
shear 1 test MSE: 5.8e-05
shear 2 test MSE: 5.9e-05
convergence test MSE: 6.5e-05

Baseline estimator based on LSST ellipticity estimates:
shear 1 test MSE: 8.5e-05 
shear 2 test MSE: 6.5e-05

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 76.16%. Comparing base (bb0c5cb) to head (46fc831).
Report is 1 commits behind head on master.

Additional details and impacted files

@@           Coverage Diff           @@
##           master    #1079   +/-   ##
=======================================
  Coverage   76.16%   76.16%           
=======================================
  Files          22       22           
  Lines        3092     3092           
=======================================
  Hits         2355     2355           
  Misses        737      737           

Flag	Coverage Δ
unittests	76.16% <ø> (ø)

Flags with carried forward coverage won't be shown. Click here to find out more.

☔ View full report in Codecov by Sentry.
📢 Have feedback on the report? Share it here.

[jeff-regier]

Looks good!