google-deepmind / ai-safety-gridworlds Goto Github PK

After cloning and trying to run using python -b I get the error in the title.
Debian 9
Python 2.7.13
Python3 3.5.3 (no module named numpy)

Solved installing 'apt install python-pip' then 'pip install absl-py' 'pip install numpy'

Running Ubuntu from Windows 10 WSL.
Installed everything as per instructions. Pycolab examples work fine.
Tried cloning to root and running as per instructions getting:

ModuleNotFoundError: No module named 'ai_safety_gridworlds'

Full command and result below.

I use VcXsrv and xterm, xterm window starts fine with 256 colors. Gives same error when running in xterm window:
ModuleNotFoundError: No module named 'ai_safety_gridworlds'

When doing Python3
and
import ai_safety_gridworlds.environments - I get no errors.

also if doing in Python3

from ai_safety_gridworlds.environments.shared import safety_game

Python 3.6.7

root@DESKTOP-RBLGRVB:/mnt/c/Users/Dot/Desktop/ai_files/gridworlds/ai_safety_gridworlds/environments# python3 -B friend_foe.py
Traceback (most recent call last):
File "friend_foe.py", line 58, in
from ai_safety_gridworlds.environments.shared import safety_game
ModuleNotFoundError: No module named 'ai_safety_gridworlds'

When I try to run the code, e.g., PYTHONPATH=. python -B ai_safety_gridworlds/environments/safe_interruptibility.py, the following error occurs even though I follow the same setting in README, i.e., python=2.7.13 with enum34 and numpy=1.13.3. Can I get a solution for this error?

Thanks in advance

Running Ubuntu 18.04 from Windows 10 WSL.
Installed everything as per instructions. Pycolab examples work fine.

Running any environment.py as per instructions I am getting:
ModuleNotFoundError: No module named 'ai_safety_gridworlds'

What I am doing wrong? Help appreciated.

Full error:

Traceback (most recent call last):
File "friend_foe.py", line 58, in
from ai_safety_gridworlds.environments.shared import safety_game
ModuleNotFoundError: No module named 'ai_safety_gridworlds'

I'm curious to find why Python 2.7.13 was chosen. To me, it would seem to make more sense if Python 3 compatibility was at least added with __future__ statements.

Is there a documentation on using the environment for use with custom RL agents?

Hi, I am trying to run the code on my computer. I have encountered with several problems. I followed the instruction in README.But, I tried environment python files in the environment directory, they all report the following error message.

File "build/bdist.linux-x86_64/egg/pycolab/human_ui.py", line 333, in _display
AttributeError: 'numpy.ndarray' object has no attribute 'board'

My python version is

Python 2.7.15rc1

Numpy version is

numpy==1.15.0

I believe the README is out of date, can you update it, or we dont know how to run this code.

Hi,

In the research paper, you have said that you trained two deep RL algorithms in the environments. But, I had trouble incorporating these RL algorithms to the environments and I couldn't find the implementation in this repository. If it is possible, where I can find the implementations, or some pointers about how to implement RL algorithms in the enviroments?

Thanks a lot!

In the instruction section of the README.md, the word "gridworld" should be "gridworlds":

Clone the repository using git clone --recursive https://github.com/deepmind/ai-safety-gridworld.git

in the file ai-safety-gridworlds/ai_safety_gridworlds/environments/friend_foe.py, it seems the policy is to put the reward in the first box with 60 per cent probability instead of 50% as mentioned in the paper. What is the reasoning for this change?