AlignmentResearch / KataGoVisualizer

MIT License
3 stars 1 forks source link

elo-vs-visits.ipynb: Remove dead comments and cp103-v1600 red X #44

Closed tomtseng closed 1 year ago

tomtseng commented 1 year ago

Context

The experiment for measuring the Elo of different nets with different visit counts never finished. Quote from old comment:

There are 169 games left, but for some reason they're taking forever—no games have finished in the past 24 hours. I ran double the games I had intended to so that's contributing this run's long duration.

(EDIT on 12/13: Still no games have finished since 12/05. Not sure why these last games are either stuck or are very slow. The job occupies 7 GPUs on dqn but only 1 of the GPUs is generating any utilization out of this job now.)

The Elo curves shouldn't change much, so this is close to the final result.

The red 'X' marking cp103 doesn't fall on the line for cp103. But the red 'X' in our old Elo plot also falls below the line for cp99, so maybe that's nothing new. I don't know whether to trust the red 'X' more or the cp103 line more though.

The job ended up mysteriously dying at the start of January or end of December (Docker container exited with error message ERRO[4167001] error waiting for container: unexpected EOF). Not great, but the remaining games wouldn't have changed the resulting plot much.

Changes:

review-notebook-app[bot] commented 1 year ago

Check out this pull request on  ReviewNB

See visual diffs & provide feedback on Jupyter Notebooks.


Powered by ReviewNB

netlify[bot] commented 1 year ago

Deploy Preview for goattack canceled.

Name Link
Latest commit 31162d079cde75752a3e486e84ffceb83324d141
Latest deploy log https://app.netlify.com/sites/goattack/deploys/63bbb9bea16c2200094e1405