The title says it all: the last modification to the precision-recall implies that if the second argument to the precision-recall function is not bettwen 0 and 1, the code gives non-sensical results without a warning. It used to work.
I realise that the argument is called 'probas_pred', but let's face it, it can be anything that is an increasing function of the confidence to have a detection, in other words a test statistics. A comon use case is the plug in the output of a decision_function, as with SVMs, or with an F-score. I suggest to change it back to working with any non-bounded test statistic. Any reason not to? I am proposing to do the work. Cheers, Gaël PS: I've been overhauling a code that was written last February, using the scikit-learn to work with the latest version, and its broken in many subtle ways (like the one that I am mentionning in this email) due to subtle changes in behavior in the scikit. :( Will send pull requests for all. ------------------------------------------------------------------------------ The Windows 8 Center In partnership with Sourceforge Your idea - your app - 30 days. Get started! http://windows8center.sourceforge.net/ what-html-developers-need-to-know-about-coding-windows-8-metro-style-apps/ _______________________________________________ Scikit-learn-general mailing list Scikit-learn-general@lists.sourceforge.net https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
