Thanks for the heads up. However, the proposed code by @ksemb doesn't seem to solve my problem. As far as I understood it, this issue is about the normalization step that doesn't seem to work in an underdetermined case. In @ksemb's proposed code, he still uses SVD to implement LDA (in contrast to the MATLAB implementation he used as a template). This means that regularization of the covariance matrix is only possible in the original MATLAB code, but not in @ksemb's Python version.
Following the discussion of https://github.com/scikit-learn/scikit-learn/issues/1649, I think that with ksemb's last comment he makes it clear that there are still problems even with his revised version. I suggest to use the classical straightforward formulation by e.g. Duda and Hart, pp. 117-121, where you simply compute the scatter matrices and get the weights via w = S_W^{-1} * (m_1 - m_2) This is exactly what happens in the MATLAB template that @ksemb is referring to. In this case, the calculation of the within-class scatter matrix S_W can be shrunk when computing S_1 and S_2 (and potentially of course for more than two classes). I've been using such an implementation (at least in MATLAB) for a long time and never experienced any problems. I will also post the gist of this message to the comments of issue 1649 to discuss the next steps. Clemens On 02/18/2014 06:20 PM, Alexandre Gramfort wrote: > https://github.com/scikit-learn/scikit-learn/issues/1649 > > have a look at the implemention from @ksemb > > https://gist.github.com/ksemb/4c3c86c6b62e7a99989b > > and the discussion on the issue > > HTH > A > > ------------------------------------------------------------------------------ > Managing the Performance of Cloud-Based Applications > Take advantage of what the Cloud has to offer - Avoid Common Pitfalls. > Read the Whitepaper. > http://pubads.g.doubleclick.net/gampad/clk?id=121054471&iu=/4140/ostg.clktrk > _______________________________________________ > Scikit-learn-general mailing list > Scikit-learn-general@lists.sourceforge.net > https://lists.sourceforge.net/lists/listinfo/scikit-learn-general > ------------------------------------------------------------------------------ Flow-based real-time traffic analytics software. Cisco certified tool. Monitor traffic, SLAs, QoS, Medianet, WAAS etc. with NetFlow Analyzer Customize your own dashboards, set traffic alerts and generate reports. Network behavioral analysis & security monitoring. All-in-one tool. http://pubads.g.doubleclick.net/gampad/clk?id=126839071&iu=/4140/ostg.clktrk _______________________________________________ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
