年度106
書名巨量資料之矩陣視覺化
部分章節The innovation of biomedical and industrial techniques with continued development of computer technology have caused dramatic changes of data generation and collection. Data scale tends to grow exponentially while data quality becomes unreliable. Statistical methods for validation and analysis of big data with its computation techniques became important research topics nowadays. Visualization and exploratory data analysis (EDA) are going to play essential roles in deep analytics on big data analysis. Yet there are some problems to be solved and techniques to be developed. Most current big data visualization methods focus on node-link diagram based dynamic network drawing. They mainly rely on the 2D and 3D scatterplots that do not consume much computing memory, power, and display space; however, the drawback is the limitation on dimensions of variable for visualization. This works first aims to resolve the potential difficulties for applying the techniques of matrix visualization for continuous type big data: (1) computation and permutation of proximity matrices; (2) display of big data. We shall integrate the strength of GAP (generalized association plots), SDA (symbolic data analysis), with Hadoop/Spark computing facility for taking care of these problems of computation and display and for creating environment for matrix visualization of continuous type big data. Here we apply the proposed MV for big data techniques on the 2000 Longitudinal Health Insurance Database (LHID2000) of National Health Insurance Research Database (NHIRD) published by National Health Research Institutes (NHRI) in Taiwan. We will then move on and expand the environment for matrix visualization of continuous type big data to binary, categorical, cartography, and other types of big data. We expect to face even more challenging difficulties while developing related techniques.
全部作者高君豪
出版日期2018-06-20
備註專書