We have created an open-source auto reinforcement learning library called Kindo. You can see the code and example here https://github.com/NEU-AI-Skunkworks/kindo/tree/master

We, in particular, Kinesso fellow Fedor Grab https://www.linkedin.com/in/fedor-grab-14609b1a4/  have created an open-source auto reinforcement learning library called Kindo.  You can see the code and example here  https://github.com/NEU-AI-Skunkworks/kindo/tree/master

kindo

Kindo is a reinforcement learning high-level API enabling developers and analysts to use Stable Baselines 3 and TF-Agents algorithms.

Stable Baselines 3 is powered by PyTorch. TF-Agents is powered by Tensorflow 2.X
Kindo enables to train models using both Tensorflow 2.X and PyTorch deep learning frameworks.

Main features

  • Training algorithms from both tf_agents and stable_baselines3 packages
  • Monitoring and plotting training history, saving it in a consistent format
  • stable_baselines3 similar callbacks for both stable_baselines3 and tf_agents
  • Open AI gym environments compatibility
  • Open AI gym environment from .csv file creation
  • Bandit problems support

Installation

Prerequisites

Kindo requires python 3.8+

Install using pip

pip install git+https://github.com/NEU-AI-Skunkworks/kindo.git@master

Important Information

If you use conda environment, and your kernel was killed with an error
OMP: Error #15: Initializing libiomp5.dylib, but found libomp.dylib already initialized.
Check out this stackoverflow question

Examples

Kindo code example notebook is provided here