Hi there!

Currently, the Ledoit Wolf covariance estimation does not standardize the data 
before computing the shrinkage parameter. However, I think this is not correct, 
because the data should have zero mean and unit L2 norm for the algorithm to 
work correctly.

I only found one reference which explicitly states this (page 241, right after 
Eq. 2):
http://ba.stat.cmu.edu/journal/2010/vol05/issue02/gramacy.pdf

I implemented a real-world test using shrinkage LDA. On my data set, I get 100% 
accuracy when using the empirical covariance. I also get 100% when I use the 
correctly shrunk cov (standardizing the data). However, I only get around 50% 
when using the current sklearn implementation.

I have also posted an issue on GitHub:
https://github.com/scikit-learn/scikit-learn/issues/3508

Should I create a PR to fix this, or am I completely missing something?

Clemens

------------------------------------------------------------------------------
Infragistics Professional
Build stunning WinForms apps today!
Reboot your WinForms applications with our WinForms controls. 
Build a bridge from your legacy apps to the future.
http://pubads.g.doubleclick.net/gampad/clk?id=153845071&iu=/4140/ostg.clktrk
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to