save videos during training

[samuelarnesen]

Description

This addresses Issue #523 to automatically save videos during training time. This builds off of the following, earlier PR.

Known Limitations: (1) Will not necessarily save a video at the end of the training run - it just saves a video at the first episode after each checkpoint. (2) Saves videos during training episodes (and not in a separate, evaluation environment)

Testing

Added tests to test_scripts.

[AdamGleave]

A lot of the test failures on Mac seem to be down to ffmpeg not being installed. I think we removed that in https://github.com/HumanCompatibleAI/imitation/pull/539 as it was making Mac OS build times very high, although an alternative that might be worth revisiting is installing ffmpeg from a static executable (it was compiling it under brew that was slow). Alternatively we can just skip the video tests on Mac, as we do the others.

[codecov[bot]]

Codecov Report

Merging #597 (7714484) into master (d6f8ca2) will increase coverage by 0.04%. The diff coverage is 98.90%.

@@            Coverage Diff             @@
##           master     #597      +/-   ##
==========================================
+ Coverage   97.49%   97.54%   +0.04%     
==========================================
  Files          85       85              
  Lines        8099     8176      +77     
==========================================
+ Hits         7896     7975      +79     
+ Misses        203      201       -2     

Impacted Files	Coverage Δ
src/imitation/scripts/common/rl.py	97.43% <ø> (ø)
src/imitation/scripts/train_adversarial.py	94.87% <83.33%> (+0.20%)	:arrow_up:
src/imitation/scripts/common/common.py	97.82% <100.00%> (+0.29%)	:arrow_up:
src/imitation/scripts/config/train_adversarial.py	71.25% <100.00%> (+0.36%)	:arrow_up:
src/imitation/scripts/config/train_imitation.py	71.66% <100.00%> (+3.87%)	:arrow_up:
...ion/scripts/config/train_preference_comparisons.py	85.52% <100.00%> (+0.19%)	:arrow_up:
src/imitation/scripts/config/train_rl.py	79.22% <100.00%> (+0.27%)	:arrow_up:
src/imitation/scripts/eval_policy.py	100.00% <100.00%> (ø)
src/imitation/scripts/train_imitation.py	95.65% <100.00%> (+0.06%)	:arrow_up:
.../imitation/scripts/train_preference_comparisons.py	97.01% <100.00%> (+0.13%)	:arrow_up:
... and 6 more

:mega: We’re building smart automated test selection to slash your CI/CD build times. Learn more