2013/6/17 Andreas Mueller <[email protected]>: > On 06/17/2013 09:59 AM, Olivier Grisel wrote: >> 2013/6/17 Andreas Mueller <[email protected]>: >>> On 06/14/2013 02:32 AM, Lars Buitinck wrote: >>>> 2013/6/14 Mathieu Blondel <[email protected]>: >>>>> We could return a tuple X, y, metadata. >>>> That would save a lot of newb confusion, I think. The question "how do >>>> I create a Bunch" comes up regularly -- people seem to think it's our >>>> preferred dataset format, even though it's just meant for the >>>> examples. >>>> >>> With a long enough deprecation cycle, I'm +1. >> I am not sure I am +1 on this one as I am not sure that all datasets >> will be always represented as a X, y pair. >> >> For instance for learning to rank problems it's more a triple like >> (data, target, query_id). >> > So what is the problem with just returning this?
It's not homogeneous, but fair enough. > (also, what is query_id?) A integer index for each sample to group the samples into query result sets when doing learning to rank. This can then be used to generate pairs of samples on the fly when doing the pairwise learning to rank reduction. -- Olivier http://twitter.com/ogrisel - http://github.com/ogrisel ------------------------------------------------------------------------------ This SF.net email is sponsored by Windows: Build for Windows Store. http://p.sf.net/sfu/windows-dev2dev _______________________________________________ Scikit-learn-general mailing list [email protected] https://lists.sourceforge.net/lists/listinfo/scikit-learn-general
