On Sun, Jan 06, 2013 at 09:53:58PM +0100, Andreas Mueller wrote: > >> I'm a bit confused by our usage of random numbers in the tests. > >> Usually all tests have fixed random seeds. But then there is a global > >> random state that is set > >> explicitly and printed for reproducability. > >> What is that state for? > > It's a way to reproduce failures caused by missing random_state. > This is the only reason?
Yes. We had a long discussion with the Debian packager maintainers that were asking for this feature to reproduce failures on their box. I somewhat thought that it wasn't necessary, because we should be setting specific seeds in the tests themselves. However, it made their life easier, as we kept checking in tests that didn't follow this policy (which we shouldn't). > Ok, then. But why not just set it to one fixed value there? > >> One of the open issues was caused by the test using this global > >> state - actually the test even said it was using the global random > >> state! > > I would rather avoid using using the shared random number generator to > > avoid having tests that depends on the running order. +1. G ------------------------------------------------------------------------------ Master Visual Studio, SharePoint, SQL, ASP.NET, C# 2012, HTML5, CSS, MVC, Windows 8 Apps, JavaScript and much more. Keep your skills current with LearnDevNow - 3,200 step-by-step video tutorials by Microsoft MVPs and experts. SALE $99.99 this month only -- learn more at: http://p.sf.net/sfu/learnmore_122412 _______________________________________________ Scikit-learn-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
