베이지안 네트워크 기반 가뭄 예측을 위한 Pgmpy 라이브러리의 개선

초록

Pgmpy Library is the Python-based library that can construct and calculate probabilistic graphical models such as Bayesian network or Markov chain. In this paper, we improve of Pgmpy library by adding Likelihood Weighting Sampling inference function for Bayesian network composed with probability variables depending on linear Gaussian conditional probability distribution. We verified result of improved library by comparing with result of Bayesian network written with Matlab. We compared hydrological drought index forecasting results obtained with data of MODIS meteorological satellite and MME climate prediction data of APEC Climate Center. Based on the comparison results, we can think that the calculation by Pgmpy library has no difference within allowable error ranges with the result from Matlab. Therefore, we can conclude the improvement proceeded well and expect improved library will be able to perform more various Bayesian network calculation than before.

키워드