miicTeam / miic_R_package

Learning causal or non-causal graphical models using information theory
GNU General Public License v3.0
26 stars 3 forks source link

Comparison scheme to choose a top Z in case of ties #89

Closed vcabeli closed 3 years ago

vcabeli commented 3 years ago

Tied scores can happen with some data (e.g. discretizing semi-continuous variables) and cause the final choice of contributing nodes to be order-dependent. This order can have slight variations depending on the number of threads with a dynamic schedule.

In the case of a tie, we need an order-independent test to choose a new top Z. This commit implements a comparison between the Zs' entropies by computing I(X;Z)+I(Y;Z), and another test that depends on the seeded RNG if those sums are also equal.