成分数据典型相关分析的增量算法

The approach of connecting linear correlations between several sets of multidimensional compositional variables known as canonical correlation analysis (CCA) for compositional data streams is widely applicable to the study of economics, administration, geology, and chemistry. In the context of massive data, it is of great significance to study how to perform CCA for compositional data streams. Propose an incremental modeling method for the CCA on compositional data streams, which provides accurate results based on the decomposition of the covariance matrix. Furthermore, two incremental modeling methods for compositional data streams are also derived. The first is the sequential block algorithm, which conducts CCA in the order of data stream blocks. The second is the parallel block algorithm, which can improve the calculating efficiency. The proposed methods do indeed outperform non-incremental ones in terms of running time while maintaining the accuracy of canonical correlation computing, according to extensive simulation studies on compositional data with various sample sizes and probability distributions.