pweglik / stochastic-systems

Stochastic Systems

Setup

conda create --name stochastic-systems python=3.10

conda activate stochastic-systems

pip install swig
pip install -r requirements.txt

How to run

python es_grad_im.py --use_td3
python3 es_grad.py --use_td3
python3 es_grad_evo.py

TODO

1. Memory should be independent of agents - we can use same data to train many agents independently.

2. Add basic population mechanisms (population, selection, mutation):

   • Train 100 networks (RL)
   • Choose 10 best. (selection)
   • Random Gaussian step of weights (mutation) into 100 networks again.
   • Rinse and repeat

3. Add network architecture mutation. Test if it trains correctly.

4. Add CMAES.

5. Watch this spider walk