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

The return value of predict_proba function in causality.pairwise #98

Open NawTun opened 3 years ago

NawTun commented 3 years ago

Dear Diviyan: I found that in doc image Does the return value of this function is between -1 and 1? but when I applied this function I often got some much larger values such as 13,14, do this larger values mean that the causal relation is much stronger? Thank you for your reply.

diviyank commented 3 years ago

Hello @NawTun,

It depends on the algorithm, but currently .predict_proba outputs the raw output of the algorithm. Which one are you using ?

Best, Diviyan

NawTun commented 3 years ago

Hi @Diviyan-Kalainathan,

Thanks for your reply, I used ANM algorithm,

Although this function returns a raw output, I can still compare the absolute value one output to another output in order to understand relative probability. For example, predict_proba(a,b) returns 4 when predict_proba(c,d) returns 2, so a->b is relatively more possible than c->d ,

am I right?

diviyank commented 3 years ago

Yes that is correct. The algorithm is more confident in a -> b than c ->d. Did you check with the original (matlab) implementation of ANM ? Those values does seem high for ANM...

NawTun commented 3 years ago

OK, Thanks for your affirmance, Diviyan, Maybe the original implementation can help to understand, I will check it later. In addition, CDT is really a great tool, :blush: Thanks again.

945716994 commented 2 years ago

If the value is between -1 and 1, doesn't that mean it doesn't know the direction of the edge.