shaido987
commented
5 months ago Hello,
As of right now, we have implemented prior knowledge injection into the PC algorithm (https://github.com/huawei-noah/trustworthyAI/blob/master/gcastle/castle/algorithms/pc/pc.py). We plan to add this to more algorithms in the future but there is still no time plan for this, you can see issue #79.
Suppose I impose prior knowledge that X and Y are sink nodes, i.e. they have not outgoing edges. Then, I would like to forbid using them when building a skeleton using the separation set procedure.
The rationale is the following -- I know that X and Y cause nothing, they can only be children nodes in the causal graph. Thus, I cannot use them to discard causal links because, again, they are not causal drivers of anything.
I hope that makes sense. Let me know if you disagree or have questions.