conda create --name stochastic-systems python=3.10
conda activate stochastic-systems
pip install swig
pip install -r requirements.txt
python es_grad_im.py --use_td3
python3 es_grad.py --use_td3
python3 es_grad_evo.py
-
Memory should be independent of agents - we can use same data to train many agents independently.
-
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
-
Add network architecture mutation. Test if it trains correctly.
-
Add CMAES.
-
Watch this spider walk