This work was designed to develop an algorithm, epsilon-machine (e-machine), for network inference from non-time series data. The key idea of our method is reweighting the observed configurations to make their frequencies equally likely. We show that this inference outforms previous methods, especilly in the regime of strong interaction and/or small sample size of observations.
Use Binder to run our code online. You are welcome to change the parameters and edit the jupyter notebooks as you want.
- Code Documentation
- Code Source
- https://github.com/danhtaihoang/e-machine