Predictive learners (pred_learn package)
For evaluating agents that learn predictive models and then use those in RL setting.
Installation
Lots of requirements, might streamline this later. Tested on Python3.6.
Then clone this repo, install requirements, install this package:
git clone https://github.com/EqThinker/predictive-learners.git
pip install tensorflow # (consider GPU supporting version)
cd predictive-learners
pip install -r requirements.txt
pip install -e .
Requirements
- pytorch
- numpy
- https://github.com/EqThinker/pytorch-a2c-ppo-acktr (for RL)
- pandas
- opencv-python
- pyvirtualdisplay
- more see requirements.txt
Gyms (install only those that you're planning to run)
- gym
- gym_ple
- gym_tetris
- mujoco
- deepmind suite
Environments available
- gym classics: CartPole, MountainCar, etc.
- games: CarRacing, PixelCopter, Tetris, etc.
- Deepmind control suite:
Running experiments
Use files in scripts as starting point