search

mckinsey / causalnex

A Python library that helps data scientists to infer causation rather than observing correlation.
http://causalnex.readthedocs.io/
Other
2.21k stars 256 forks source link

Support of OR statements in the observations field for the InferenceEngine query() method #127

Open NShah19 opened 3 years ago

NShah19 commented 3 years ago

Description

The query() method for the InferenceEngine currently supports the computation of the marginal likelihood of states in the network given observations joined by conjunctions (ANDs). Support for disjunctive observations would allow for greater insights to be gained from the network.

Context

This extension would allow for more functionality within CausalNex for examining different scenarios.

oentaryorj commented 2 years ago

Hi @NShah19, thanks for your suggestion - this is certainly an interesting point. To our best knowledge, however, the standard notations for conditional/joint probabilities and Bayesian inference in the literature are defined based on conjunctive operation.

Would you be able to perhaps suggest a reference paper and/or use case example for us to investigate further? It might be possible to make use of the inclusion-exclusion principle to do this, but we would like to make sure that the inference mechanism is general enough if we were to implement it.