Unable to reproduce CollectMineralShards result

[vwxyzjn]

Hi,

I was trying to reproduce the CollectMineralShards result you posted here:

by running

python run_agent.py --map_name CollectMineralShards --model_name my_mineral_shards_model

However, after 60k steps, the rewards still remain around 25

I did not use the flag --n_envs 32. Would that be the cause of this irreproducibility?

Thank you.

[pekaalto]

Hi, actually the x-axis in this chart is number or episodes. In your screenshot can see that you have only 2100 episodes so far. So it's long way to go.

Other notes:
-The training is probably more robust with multiple environments even if run same number of episodes.
-There is lot of randomness involved in the learning process. You will get different training graph each run.

[vwxyzjn]

I see. Thank you!