Submissions Q&A

It seems evaluation server dead.
Mine stuck for 7 hours

View post 1

Evalutation error : Unity environment took too long to respond

In local docker, it always tested well.
I am getting same error for every submission.

The Unity environment took too long to respond.

View post 0

Problem on agent

It runs correctly in my local machines.
Also, set realtime_mode=True doesn’t help.

View post 0

Testing agent in local with docker

I followed the instruction in README.md.
I successfully built docker image.

I run docker image with tow terminals as described in Run Docker image section.
The agent looks good and it waits for the environment.
When I run the environment, anything happens.
Below are my console messages for both.

Agent

root
INFO:mlagents_envs:Start training by pressing the Play button in the Unity Editor.
Traceback (most recent call last):
  File "run.py", line 27, in <module>
    env = ObstacleTowerEnv(args.environment_filename, docker_training=args.docker_training)
  File "/srv/conda/lib/python3.6/site-packages/obstacle_tower_env.py", line 45, in __init__
    timeout_wait=timeout_wait)
  File "/srv/conda/lib/python3.6/site-packages/mlagents_envs/environment.py", line 69, in __init__
    aca_params = self.send_academy_parameters(rl_init_parameters_in)
  File "/srv/conda/lib/python3.6/site-packages/mlagents_envs/environment.py", line 491, in send_academy_parameters
    return self.communicator.initialize(inputs).rl_initialization_output
  File "/srv/conda/lib/python3.6/site-packages/mlagents_envs/rpc_communicator.py", line 80, in initialize
    "The Unity environment took too long to respond. Make sure that :\n"
mlagents_envs.exception.UnityTimeOutException: The Unity environment took too long to respond. Make sure that :
	 The environment does not need user interaction to launch
	 The Academy and the External Brain(s) are attached to objects in the Scene
	 The environment and the Python interface have compatible versions.

Environment

+ ENV_PORT=
+ ENV_FILENAME=
+ '[' -z '' ']'
+ ENV_PORT=5005
+ '[' -z '' ']'
+ ENV_FILENAME=/home/otc/ObstacleTower/obstacletower.x86_64
+ touch otc_out.json
+ APP_PID=7
+ xvfb-run --auto-servernum '--server-args=-screen 0 640x480x24' /home/otc/ObstacleTower/obstacletower.x86_64 --port 5005 2
+ TAIL_PID=8
+ wait 7
+ tail -f otc_out.json

View post 0

Problem on agent

I tested the Obstacle tower environment with local machines.
I confirmed that the action space is consist of 4 numbers in list, like [0, 0, 0, 1]
I submitted a starter kit agent for a test, and it evaluated successfully.

Then, I tested my agent for submission which slightly modified from starter kit.
The modification was to force jump action 0 from env.action_space.sample()
Actual source code is below. It is part of run.py in run_episode(env) function.

while not done:
    action = env.action_space.sample()
    action[2] = 0
    obs, reward, done, info = env.step(action)

From evaluation log, it stuck at step 0.

It is my first try to participate in this kind of challenges, therefore I am not familiar with the environment.
What is the problem with my code?

View post 1