Skip to content

Latest commit

 

History

History
91 lines (62 loc) · 2.39 KB

README.md

File metadata and controls

91 lines (62 loc) · 2.39 KB

Dependencies

Extras to install to be able to run the exact code in this repository

pip install pandas opencv-python tqdm lz4 requests

tensorflow

Don't install tf 2.0

pip install tensorflow==1.14

ray

pip install ray ray[debug] ray[rllib]

cityflow

On MacOS, brew install boost is equivalent to apt install libboost-all-dev

apt update && apt-get install -y build-essential libboost-all-dev cmake

git clone --recursive https://github.com/cityflow-project/CityFlow.git

go to cityflow directory

pip install .

gym

git clone https://github.com/openai/gym
cd gym
pip install .

gym_cityflow

Environment Usage

gym_cityflow文档结构遵循 gym 环境的规则，详见此处

使用前，cd到gym_cityflow目录，运行 pip install -e .进行环境包的安装。否则，无法在 ray_dqn_agent.py 中直接 import gym_cityflow

ray中可以自定义环境以及神经网络模型，详见此处

Configuration

Config包含 agent_config 和 env_config

  1. agent_config：遵照 ray 中 agent/[algorithm].py的 DEFAULT_CONFIG 以及 agents/trainer.py的 COMMON_CONFIG，两者是继承关系，DEFAULT_CONFIG继承 COMMON_CONFIG
  2. env_config：为 agent_config字典下的一个属性，即其为大字典中的小字典，用于配置 Cityflow 环境参数

Future Research

Just change this part

    trainer = DQNTrainer(
        env=CityflowGymEnv,
        config=config_agent)

and

import ray.rllib.agents.dqn as dqn
from ray.rllib.agents.dqn import DQNTrainer

into your algorithm, then it would work.

Remember: Follow the DEFAULT_CONFIG in each algorithm and change your own configuration.

TensorBoard

For Tensorboard, just

tensorboard --logdir = ~/ray_results

注: ray_results目录位置需根据本机位置修改, 可直接写绝对路径