/offline-RL-congestion-control

Primary LanguagePythonApache License 2.0Apache-2.0

NOTES

This repo is for the Ten-TMS team to participate in the RL challenge.

emulate_data_dir is supposed to located at "./emulated_dataset" testbed_data_dir is supposed to located at "./testbed_dataset" We also suppose a directory called data_dir for further data split which is "./" by default

Before training, the program will split the dataset into train, validation, test data, which will be moved to folders './emulated_dataset', './train_eval_dataset', './eval_dataset', respectively.

We pre-store the behavior of baseline model for after-train evaluation. The file is baseline_eval.pkl, which is located at './emulated_dataset'.

Please feel free to contact us if necessary. zhangwei.666@bytedance.com

Quick Evaluation

create ./figs directory, then run run_ten_tms_model.py

1. Dependencies

pip install -r requirements.txt

2. Generate Test Set and Validation Set

Set the path_config.json, then run test_set_generator.py

3. Generate Pretrain Model

Set the pretrain_config.json, then run pretrain/train.py

4. Finetune Model

Set the finetune_config.json, the run fintune/train.py