A Fast HITON_PC Algorithm

Abstract

The HITON_PC algorithm which is a state-of-the-art local causal discovery algorithm can deal with a dataset with a very small sample-to-variable ratio efficiently. But it cannot perform inefficiently on a dataset with a very large sample. To address this problem, a fast HITON_PC algorithm is presented which uses a new yet simple search strategy from high order to low order to improve the efficiency of HITON-PC. Experimental results show our fast HITON_PC outperforms the HITON_PC algorithm. Moreover, we also apply the new search strategy to MMPC algorithm. Our method also is superior to MMPC.