News for hackers. Stuff that matters.
Re:Data mining without prejudice by mbq on Monday December 19, 2011 @12:41PM (#3370322)
This boils down to a following old algo: 1. plot all possible scatterplots 2. order them due to an area of white space left 3. claim the top-whatever novel exciting relationships. This way it is anti-robust and produces too much false-postives to be seriously useful.
Re:Data mining without prejudice by baltcode on Monday December 19, 2011 @12:59PM (#3370398)
I think you can correct for the false-positives to some extent. I am more skeptical of the computational load of basically testing all sorts of relationships. They have an approximate algorithm for doing that.
Re:Data mining without prejudice by mbq on Monday December 19, 2011 @01:38PM (#3370549)
Yeah... do it exact and no hit will survive; do it with Bonferroni and nothing will really change. And in fact it is only checking two-way relationships, which is already a minor fraction of "all sorts".