Run experiments - Githubissues

SebastianLarssonDTU / 02456-Reinforcement-Learning-Project

0 stars 0 forks source link

Open DoaFever opened 3 years ago

DoaFever commented 3 years ago

impala (value clipping), impala death for long (do they converge to the same solution?)
ppo, impala (value clipping), impala (death 5 / 20?) 200 levels
impala 20 death penalty 50 levels
ppo, impala, impala death with framestacking
data augmentation

DoaFever commented 3 years ago

I have started: Long run for ppo and impala, (needs to be continued again when done) Impala 20 death penalty 50 levels

DoaFever commented 3 years ago

Priority:

Run with 200 levels: IMPALA, IMPALA DEATH
Run Framestacking 50 levels: IMPALA, IMPALA DEATH. (framestacking 2, 32 env og 4, 16 env)
Run Impala death pen 50 levels for long (16M timesteps)!