Weak lensing updates

[timwhite0]

Most recent results on DC2:

BLISS:
shear 1 test MSE = 3.2e-05
shear 2 test MSE = 4.7e-05
convergence test MSE = 3.5e-05

Predicting zero in every tile:
shear 1 test MSE = 6.8e-05
shear 2 test MSE = 6.9e-05
convergence test MSE = 1.4e-04

See notebooks/dc2/evaluate_encoder.ipynb in our case study for corresponding shear/convergence maps and scatterplots. In our next PR, we'll provide results for an estimator based on averaging ellipticities of neighboring galaxies (already implemented in utils/weighted_avg_ellip.py, just needs to be tuned).

A couple more updates for both DC2 and our simulator:

• Slightly more complicated prior than our last PR. We still generate images with constant shear and convergence, but now the values of shear/convergence are allowed to differ from image to image.
• We now plot a few shear and convergence maps during training, instead of just one. We also plot a scatterplot of our estimates vs. the ground truth for shear1, shear2, and convergence.
• Some housekeeping things: separate configs for DC2 and synthetic images, new weak lensing README.

[codecov]

Codecov Report

All modified and coverable lines are covered by tests ✅

Project coverage is 76.15%. Comparing base (1868ae0) to head (ab273b8).
Report is 1 commits behind head on master.

Additional details and impacted files

@@            Coverage Diff             @@
##           master    #1074      +/-   ##
==========================================
+ Coverage   76.14%   76.15%   +0.01%     
==========================================
  Files          22       22              
  Lines        3089     3091       +2     
==========================================
+ Hits         2352     2354       +2     
  Misses        737      737              

Flag	Coverage Δ
unittests	76.15% <100.00%> (+0.01%)	⬆️

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]

Great!