search

bancaditalia / black-it

Black-box abm calibration kit by the Bank of Italy
https://bancaditalia.github.io/black-it/
GNU Affero General Public License v3.0
43 stars 2 forks source link

Code for the new ABM paper #57

Open wangy8989 opened 1 year ago

wangy8989 commented 1 year ago

I am wondering where is the code for this paper: Combining search strategies to improve performance in the calibration of economic ABMs Thanks!

AldoGl commented 1 year ago

Hi @wangy8989!

I imagine you are interested specifically in the use of reinforcement learning to combine search methods. The idea is implemented in Black-it as a RL scheduler, following the general abstraction defined here. In the 'examples' folder you can also find a Jupyter notebook to learn how to use it.

The current RL scheduler implements the exact $\epsilon$-greedy scheme of the paper, but thanks to the work of @marcofavoritobi it is very easy to extend it and use any standard RL library, since the calibration process is wrapped within an OpenAI gym environment.

Btw, the new version of the paper has a direct link to the Jupyter notebook and some more numerical experiments. Let me know if you need more help