※ 引述《newlittle (小新)》之铭言： : 之前正相关的作法是将 covariance-variance matrix : 利用 cholesky decomposition 分成上下相同的半矩阵 : 之後藉由标准常态值 即可产生具有相关性的多维常态 : 可是现在要产生具有负相关性的值 : 例如其 covariance-variance matrix 为 : 1 -0.5 -0.5 : -0.5 1 -0.5 : -0.5 -0.5 1 : 因为他不是正定矩阵 所以无法使用cholesky decomposition : 这样一来该如何产生呢? : 谢谢 This is an interesting question. We can generate arbitrary positive equicorrelated exchangeable random variables, but not negative ones. In fact, if I remember correctly, the smallest negative correlation coefficient allowed is -1/(k-1), where k is the number of random variables. One way to generate the most negative equicorrelated exchangeable r.v.'s is to generate k i.i.d. r.v.'s, say X1,..., Xk, then (X1 -m, ..., Xk-m), where m is the average of these k r.v.'s, will yield the smallest correlation coefficient. --

※ 发信站: 批踢踢实业坊(ptt.cc) ◆ From: 166.137.8.145