Folder Structure

├─config
│  ├─envs (config for environment)
│  ├─gs (config for grid search)
│  │  └─sddpg (algorithm name)
│  └─models (config for algorithm)
│      └─sddpg (algorithm)
├─data
│  ├─log
│  │  ├─1x1 (environment)
│  │  │  ├─2019-01-17_sddpg_sz (exp_name)
│  │  │  │  ├─2019-01-17_03-18-26-sddpg_sz_s1007_1790dfbd
│  ├─report (report generated by analysis.py or grid_search.py)
│  └─setting (environment files)
│      └─1x1 (environment)
├─tools 

Workflow

1. use tools/generate_setting.py to generate roadnet.json, signal.jsonl, flow.jsonl
2. use one of the sim_agent to generate observed.txt
3. create corresponding config/envs/env.json
4. create model config config/models/<alg>/<yourname>xxx.json
5. create grid search config config/gs/<alg>/<yourname>xxx.json
6. run grid_search.py with grid search config, use --report flag to generate report

Notice!
you can implement different replay buffer in replay_buffer.py
you can implement different network architecture in core.py
any modification to algorithm file should be noticed on Wechat.

All configs file are self-explained, if you have any questions, please contact me!

Happy Exploring!