(Added tips for generating MVN data) |
|||
| Line 14: | Line 14: | ||
--[[User:Ralazrai|Ralazrai]] 21:55, 17 February 2010 (UTC) | --[[User:Ralazrai|Ralazrai]] 21:55, 17 February 2010 (UTC) | ||
| + | '''Generating correlated multi-variate normal (MVN) data:''' | ||
| + | I don't know if anyone else ran into this issue, but FreeMat doesn't know how to generate MVN random samples. The solution is to generate independent standard normal data points and perform a linear transformation. Refer to the link below for details: | ||
| + | *http://www.stat.uiuc.edu/stat428/cndata.html | ||
| + | To make matters worse, FreeMat cannot perform Cholesky decomposition. Two ways to get the desired results: | ||
| + | *Instead of starting with the covariance matrix and taking the square root, start with the upper triangular matrix A and take A'A as the covariance. (Prof. Boutin's suggestion). | ||
| + | *Perform singular value decomposition using FreeMat's "svd" command on the covariance matrix to get [u s v]. Then <math> B = u \sqrt{s} v </math> would serve as the square root. | ||
| + | |||
| + | The transformed data, using either A or B, should have be the desired statistics (please verify!). | ||
| + | |||
| + | -Satyam | ||
---- | ---- | ||
[[ 2010 Spring ECE 662 mboutin|Back to 2010 Spring ECE 662 mboutin]] | [[ 2010 Spring ECE 662 mboutin|Back to 2010 Spring ECE 662 mboutin]] | ||
Revision as of 18:25, 18 February 2010
Put your content here . . .
Here is a link to a lab on Bayes Classifier that you might find helpful. Please use it as a reference.
Enjoy, Raj..
Here is a link for a theoretical and practical assignment on Bayes Classifier.
--Ralazrai 21:55, 17 February 2010 (UTC)
Generating correlated multi-variate normal (MVN) data: I don't know if anyone else ran into this issue, but FreeMat doesn't know how to generate MVN random samples. The solution is to generate independent standard normal data points and perform a linear transformation. Refer to the link below for details:
To make matters worse, FreeMat cannot perform Cholesky decomposition. Two ways to get the desired results:
- Instead of starting with the covariance matrix and taking the square root, start with the upper triangular matrix A and take A'A as the covariance. (Prof. Boutin's suggestion).
- Perform singular value decomposition using FreeMat's "svd" command on the covariance matrix to get [u s v]. Then $ B = u \sqrt{s} v $ would serve as the square root.
The transformed data, using either A or B, should have be the desired statistics (please verify!).
-Satyam
