matlab的相关系数计算函数corr中'rows’选项的疑问

qinkong

题目说的不清楚。先把help文档贴上来：
corr Linear or rank correlation.
    RHO = corr(X) returns a P-by-P matrix containing the pairwise linear
    correlation coefficient between each pair of columns in the N-by-P
    matrix X.

    RHO = corr(X,Y,...) returns a P1-by-P2 matrix containing the pairwise
    correlation coefficient between each pair of columns in the N-by-P1 and
    N-by-P2 matrices X and Y.

    [RHO,PVAL] = corr(...) also returns PVAL, a matrix of p-values for
    testing the hypothesis of no correlation against the alternative that
    there is a non-zero correlation.  Each element of PVAL is the p-value
    for the corresponding element of RHO.  If PVAL(i,j) is small, say less
    than 0.05, then the correlation RHO(i,j) is significantly different
    from zero.

    [...] = corr(...,'PARAM1',VAL1,'PARAM2',VAL2,...) specifies additional
    parameters and their values.  Valid parameters are the following:

         Parameter  Value
          'type'    'Pearson' (the default) to compute Pearson's linear
                    correlation coefficient, 'Kendall' to compute Kendall's
                    tau, or 'Spearman' to compute Spearman's rho.
          'rows'    'all' (default) to use all rows regardless of missing
                    values (NaNs), 'complete' to use only rows with no
                    missing values, or 'pairwise' to compute RHO(i,j) using
                    rows with no missing values in column i or j.
          'tail'    The alternative hypothesis against which to compute
                    p-values for testing the hypothesis of no correlation.
                    Choices are:
                       TAIL         Alternative Hypothesis
                    ---------------------------------------------------
                      'both'     correlation is not zero (the default)
                      'right'    correlation is greater than zero
                      'left'     correlation is less than zero

    The 'pairwise' option for the 'rows' parameter can produce RHO that is not
    positive definite.  The 'complete' option always produces a positive
    definite RHO, but when data are missing, the estimates may be based on
    fewer observations.

看这个文档中‘rows’选项的介绍，分不清pairwise和complete的区别。最后好像提到他们的区别是最终那个相关系数矩阵是不是正定的。我不懂这个到底代表着什么含义，“正定”或者“不正定”有什么差别，pairwise和complete两个选项具体执行过程是怎样的，有什么差别。
谢谢大家了！

s0mrjkl

matlab的相关系数计算函数corr中'rows’选项的疑问