Unity 机器学习代理工具包 (ML-Agents) 是一个开源项目,它使游戏和模拟能够作为训练智能代理的环境。
您最多选择25个主题 主题必须以中文或者字母或数字开头,可以包含连字符 (-),并且长度不得超过35个字符
 
 
 
 
 
GitHub aee5d336 Fix discrete state (#33) 7 年前
..
ppo Fix discrete state (#33) 7 年前
unityagents Merge pull request #9 from Unity-Technologies/socket-timeout-fix 7 年前
Basics.ipynb formating and added documentation 7 年前
PPO.ipynb Fix discrete state (#33) 7 年前
README.md Update README.md 7 年前
ppo.py Fix discrete state (#33) 7 年前
requirements.txt Remove scipy from requirements 7 年前
setup.py PPO additions and warnings 7 年前
test_unityagents.py Initial commit 7 年前

README.md

alt text

Unity ML - Agents (Python API)

Python Setup

Requirements

  • Jupyter
  • docopt
  • Matplotlib
  • numpy
  • Pillow
  • Python (2 or 3)
  • Tensorflow (1.0+)

Installing Dependencies

To install dependencies, run:

pip install .

or

pip3 install .

If your Python environment doesn't include pip, see these instructions on installing it.

Provided Jupyter Notebooks

  • Basic - Demonstrates usage of UnityEnvironment class for launching and interfacing with Unity Environments.
  • PPO - Used for training agents. Contains an implementation of Proximal Policy Optimization Reinforcement Learning algorithm.

Running each notebook

To launch jupyter, run:

jupyter notebook

Then navigate to localhost:8888 to access each training notebook.

To monitor training progress, run the following from the root directory of this repo:

tensorboard --logdir=summaries

Then navigate to localhost:6006 to monitor progress with Tensorboard.

Training PPO directly

To train using PPO without the notebook, run:

python3 ppo.py <env_name> --train

For a list of additional hyperparameters, run:

python3 ppo.py --help

Using Python API

See this documentation for a detailed description of the functions and uses of the Python API.

Training on AWS

See this related blog post for a description of how to run Unity Environments on AWS EC2 instances with the GPU.