Wangyx_2021-P-16224_MMD-LCS_datanet

Citation Author(s):
Wang
Yunxia
Submitted by:
Wang yunxia
Last updated:
Sun, 09/12/2021 - 06:37
DOI:
10.21227/yyqv-1293
License:
18 Views
Categories:
0
0 ratings - Please login to submit your rating.

Abstract 

It distinguishes direct causes from direct effects of a target variable from multiple manipulated datasets with unknown manipulated variables and nonidentical data distributions.

Instructions: 

We select five BNs with low to high dimensionality to conduct the simulation. First, we randomly choose 5-7 vertices as targets to evaluate their performance in each BN. Then, given a target variable, in an interventionalexperiment, we choose manipulated variables randomly to manipulate and make sure the multiple intervention experiments are conservative. We conduct two simulations to generate two types of multiple interventional datasets in each network. The fifirst one is that corresponding ten post-intervention DAGs and probability distributions, getting ten interventional datasets as a group. The second one is fifive interventional datasets as a group. For the experiments, we run these simulations for 10 times to generate 10 groups and use |data{i}| = 10 or 5 to represent the number Of datasets in each group. We compute the average metrics for each algorithm over the ten groups of datasets produced by the two types of simulations.