Evidence: Different stopping times for training
Created on April 12|Last edited on April 12
Comment
Number of seeds: 5
Smac v1: 3m
Run set
40
We shall cut at various points, but most notably:
- QMix + cql degrades massively
- QMix + BCQ only enters the races later
Smac v1: 8m
Run set
40
Now maicq, dbc and idrqn+cql are the only algorithms in the races.
Notably, Maicq only enters the races after about 12k timesteps.
Add a comment