GoingMyWay
GoingMyWay
> @GoingMyWay Cool. Thanks! > > This problem is caused by not giving the docker container enough memory to run the task. Could you add `--shm-size 8G` (or higher) when...
I ran the command ``` python -m megaverse_rl.train --train_for_seconds=360000000 --train_for_env_steps=2000000000 --algo=APPO --gamma=0.997 --use_rnn=True --rnn_num_layers=2 --num_workers=12 --num_envs_per_worker=2 --ppo_epochs=1 --rollout=32 --recurrence=32 --batch_size=2048 --actor_worker_gpus 0 --num_policies=1 --with_pbt=False --max_grad_norm=0.0 --exploration_loss=symmetric_kl --exploration_loss_coeff=0.001 --megaverse_num_simulation_threads=1 --megaverse_use_vulkan=False --policy_workers_per_policy=2...
> @GoingMyWay > Re: Resource usage > It could be normal, but some sanity checks are: > > 1. Give the container full access to the resources > 2. Let...
Hi, when runing the tasks, there is an exception ``` [2021-07-07 10:40:36,827][08441] Visible devices: 1 [2021-07-07 10:40:36,828][08441] Unknown exception in rollout worker Traceback (most recent call last): File "/workspace/sample-factory/sample_factory/algorithms/appo/actor_worker.py", line...
> I will be also interested in an answer. As a matter of fact I have difficulties understanding how this work : from my understanding when doing the actor update...
@scotthuang1989 Hi, how can I fix it with the lasted gym?
Hi, I got the error ``` In [1]: from gym_compete import new_envs ERROR: Could not open disk Press Enter to exit ... ``` It seems that it fails to load...
@yadrimz @tomchen1000 In `main.py`, I found ``` tot_reward = tf.Variable(0.) tf.summary.scalar("DOCA/Total Reward", tot_reward) cum_reward = tf.Variable(0.) tf.summary.scalar("DOCA/Cummulative Reward", tot_reward) ``` I think `tf.summary.scalar("DOCA/Cummulative Reward", tot_reward)` should be changed to `tf.summary.scalar("DOCA/Cummulative...
> Hey @GoingMyWay! Thank you for noticing this and bringing this up. I never noticed this in my own experiments. I checked one of my old experiments and the tendency...
> If I figure out a way to fix this, then yes. But I'm not sure yet how to do it without potentially breaking things (like custom observers). > >...