论文部分内容阅读
Matrix computations form a core for many of the traditional multivariate statistical methods. Especially the matrix decompositions, such as singular value decomposition, have turned out to be very useful also from the application point of view. Principal components analysis is one example. Quite recently it has been demonstrated how exploratory factor analysis can be considered as specific data matrix decomposition with fixed unknown matrix parameters. In this so called direct factor analysis approach all model unknowns including common and unique factor scores are estimated simultaneously by minimizing a specific object function with an alternating least squares (ALS) algorithm utilizing singular value decomposition (SVD) of data matrices. Such technique also allows performing factor analysis in cases with more variables than observations. Other useful matrix decompositions for the pragmatic data-analyses are the CUR decompositions, in which a low-rank matrix approximation is based on a small number of actual columns of the data matrix. The utility of this approach is that the interpretations of the results of an application are straightforward. Practical application of such techniques requires appropriate tools for the analyses. We demonstrate the methods and their implementation using the Survo R system that allows to freely mix natural language and computation schemes in so called editorial environment.