Hi Dave.
It could be related to this bug:
https://github.com/scikit-learn/scikit-learn/issues/1649
But I'm just guessing.

If you could provide some toy data and the resulting models to reproduce, we could try to hunt
down the problem.
Any help there would be much appreciated.

Cheers,
Andy

On 02/14/2013 03:09 PM, David Reed wrote:
I was hoping I might be able to get some assistance here. I'm working on a classification problem and have started to see some pretty large discrepancies between results from Matlab and Sklearn.

The problem arises when doing univariate LDA. My set is fairly large with an N of 5000, but he class distribution is very poor with the two classes lying almost on top of one another, and so I expect the balanced accuracy to be poor.

Doing the classification problem using sklearn I end with a balanced accuracy of 0.56, using Matlab, I get 0.61. Initially I was getting a large discrepancy using xval, and so I attributed the bias to this, but these results are from resubstituion, or testing on the training set.

I would expect for such a problem with LDA, that the solution would be unique in this case, and it appears that it is not.

Does anyone have any experience with this? Or an ideas as to why this is happening?

-Dave




------------------------------------------------------------------------------
Free Next-Gen Firewall Hardware Offer
Buy your Sophos next-gen firewall before the end March 2013
and get the hardware for free! Learn more.
http://p.sf.net/sfu/sophos-d2d-feb


_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

------------------------------------------------------------------------------
Free Next-Gen Firewall Hardware Offer
Buy your Sophos next-gen firewall before the end March 2013 
and get the hardware for free! Learn more.
http://p.sf.net/sfu/sophos-d2d-feb
_______________________________________________
Scikit-learn-general mailing list
[email protected]
https://lists.sourceforge.net/lists/listinfo/scikit-learn-general

Reply via email to