Timeline for Obtaining covariance matrix from correlation matrix
Current License: CC BY-SA 4.0
9 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Feb 26 at 9:53 | comment | added | Avraham |
@M.Beausoleil, it's a fast way of doing the "sweep". Consider that you are multiplying each entry in the correlation matrix by the product of the component standard deviations. For example, the covariance matrix entry at [1, 2] will be the SD[1] * SD[2] * cor[1, 2]. By turning the SD vector into a matrix and multiplying it by itself (tcrossprod(S) would be even faster), you are "pregenerating" the needed 4x4 matrix of multiplications of SD entries each by each other. All that is left to do is to scale them, elementwise, by the correlations, thus the * R. Which could be first or second.
|
|
| Jun 8, 2022 at 16:28 | history | made wiki | Post Made Community Wiki by kjetil b halvorsen♦ | ||
| Jun 8, 2022 at 16:06 | comment | added | M. Beausoleil |
What is the idea behind R * smat %*% t(smat)?
|
|
| S Jun 8, 2022 at 16:06 | history | suggested | M. Beausoleil | CC BY-SA 4.0 |
added microbenchmark and reproducible answer and new way to compute the covariance matrix which is even more efficient!
|
| Jun 8, 2022 at 16:00 | review | Suggested edits | |||
| S Jun 8, 2022 at 16:06 | |||||
| May 12, 2019 at 15:48 | comment | added | whuber♦ |
On my system (Microsoft's version of R), the outer solution is far faster for large dimensions. It is about eight times faster than the sweep implementation. (If you were to replace the inner sweep by R*S it would be almost twice as fast, but still four times slower than outer.) Replacing outer(S,S) by d <- length(S); (matrix(S,d,1) %*% matrix(S,1,d)) is a little faster. cc @BenBolker
|
|
| May 12, 2019 at 15:00 | comment | added | kjetil b halvorsen♦ | @Ben Bolker: It would be a good exercise to compare speedwise the efficiencies ... | |
| May 12, 2019 at 14:13 | comment | added | Ben Bolker |
also outer(S,S) * R since * performs the Hadamard (elementwise) product ...
|
|
| May 12, 2019 at 13:13 | history | answered | kjetil b halvorsen♦ | CC BY-SA 4.0 |