fabiopardo / tonic_data

Data from the Tonic RL library

Tonic

Welcome to the Tonic deep reinforcement learning library.

The source code can be found here.

This repository contains the full logs from the benchmark, as well as the configuration and final checkpoint of the best seed for each agent on each environment.

Instructions

The repository can be downloaded with:

git clone https://github.com/fabiopardo/tonic_data.git

More instructions can be found on the Tonic repository.

Credit

Citing Tonic

If you use Tonic in your research, please cite:

@misc{tonic2020,
  author={Pardo, Fabio},
  title={Tonic RL Library},
  year={2020},
  publisher={GitHub},
  journal={GitHub repository},
  howpublished={\url{https://github.com/fabiopardo/tonic}},
}

tonic_data's Issues

Do the results vary significantly between Tonic + TensorFlow vs. Pytorch? And are the default hyperparams in Tonic the same as the ones used to generate this data?

Would it be apples to oranges to compare this data to data generated from the Pytorch version?

How can I reproduce the hyperparams used here? Particularly in Pytorch?

How to unpickle the log file?

Hi, I am having a lot of trouble getting to the data in the log.pkl files without knowing the structure of the pkl file itself. Could you share a code snippet showing how I can read this data?

SAC seems to be underperforming

From looking at the benchmarks, SAC seems to be underperforming quite a bit. Have you seen this library's implementation? https://autonomous-learning-library.readthedocs.io/en/stable/index.html

Are the saved benchmarks the best of 10 seeds or the average of 10 seeds? I noticed they have standard deviation shading around them, so is it average? In the readme it says best of.