Divergence of gridworld rewards and reward heatmap improvements

[AdamGleave]

New features:

• Adds support in tabular.py for affine transformations, and an EM-like optimization approach, to compute divergence in gridworlds.
• Write CLI script to plot divergence heatmaps for gridworld rewards.
• Add new reward, center_goal
• Make hatch color depend on facecolor

TODO:

• [X] Consider adding unit tests for some of the tabular.py methods? Useful to do but not urgent: had limited coverage before anyway.
• [X] May be able to consolidate some of the tabular.py and gridworld_heatmap.py code. Not easily without significant rewriting -- gridworld_heatmap makes some different design decisions due to visualization focus.

[codecov[bot]]

Codecov Report

Merging #11 into master will increase coverage by 2.75%. The diff coverage is 93.62%.

@@            Coverage Diff             @@
##           master      #11      +/-   ##
==========================================
+ Coverage   81.42%   84.17%   +2.75%     
==========================================
  Files          43       45       +2     
  Lines        2745     2799      +54     
==========================================
+ Hits         2235     2356     +121     
+ Misses        510      443      -67

Impacted Files	Coverage Δ
src/evaluating_rewards/analysis/stylesheets.py	71.42% <ø> (ø)	:arrow_up:
tests/test_scripts.py	100% <100%> (ø)	:arrow_up:
src/evaluating_rewards/analysis/visualize.py	83.68% <100%> (+0.44%)	:arrow_up:
...c/evaluating_rewards/analysis/gridworld_rewards.py	100% <100%> (ø)
...uating_rewards/analysis/plot_divergence_heatmap.py	69.89% <100%> (-0.95%)	:arrow_down:
src/evaluating_rewards/tabular.py	49.48% <84.84%> (+49.48%)	:arrow_up:
...c/evaluating_rewards/analysis/gridworld_heatmap.py	96.12% <90.9%> (-1.93%)	:arrow_down:
...ting_rewards/analysis/plot_gridworld_divergence.py	94.52% <94.52%> (ø)
...aluating_rewards/analysis/plot_gridworld_reward.py	97.72% <97.72%> (ø)
... and 3 more

---

Continue to review full report at Codecov.

Legend - Click here to learn more Δ = absolute <relative> (impact), ø = not affected, ? = missing data Powered by Codecov. Last update 6bdeec5...88afaee. Read the comment docs.