search

FenTechSolutions / CausalDiscoveryToolbox

Package for causal inference in graphs and in the pairwise settings. Tools for graph structure recovery and dependencies are included.
https://fentechsolutions.github.io/CausalDiscoveryToolbox/html/index.html
MIT License
1.08k stars 198 forks source link

Bayesian network parameters #133

Open XMAHA opened 2 years ago

XMAHA commented 2 years ago

Nice package, help me a lot~ I'm confused when testing the GS method and other BN algorithms. When only input data, edges in the generated graph are almost bi-directional. But when inputting data and a skeleton, the edges are unidirectional and the graph is sparser. So if only input data, is the causal graph accurate？ And only methods of BN structure learning are implemented, would BIC score based method and the probability parameter learning be added？

diviyank commented 2 years ago

Hello, thanks ! It comes from the theory: bayesian networks -based causal methods actually can only optimize up to the markov equivalence class, and moreover the tests are quite conservative. But ! It you add some expert knowledge (ex: skeleton) then it rules out many possibilities, and can help out a lot in the optimization and improve the results (at the condition that the skeleton is correct - otherwise it could amplify the errors)

You have quite a lot of methods ! and for example GES (added) is a BIC-score based method. Others can be added as well, please check the doc to have the full list.

Best regards,