The project is a mix of Reinforcement Learning concept and famous John Conway's Game of Life, a cellular automaton that generates thrilling structures that very looks like real living cells and complex molecules. We assume that such an environment could be an interesting source of new features as well as an inspiration itself. So.. The major goals are the experiments on SOTA in RL with the following transfer to real-world applications.
There are we have two envs:
There is a first loss we designed. It defines as: L1 distance between the amount of living cells and all the field cells: loss
There are three models:
- Random: random
- Linear: linear_observer_planter
- Actor-Critic model: a2c
There are a few hardcoded patterns that you can play with - gliders
For the moment - just clone the repo and run the latest notebooks
Latest examples of usage:
TBA
We are small at the moment and don't have any special requirements. We will be appreciated if you get one of the issues and help us. Just remember: new issue = new branch -> pull request.