tag:blogger.com,1999:blog-4446292666398344382.post6369610443316079696..comments2020-04-04T15:03:09.798-07:00Comments on Machined Learnings: An Encoding TrickPaul Mineirohttp://www.blogger.com/profile/05439062526157173163noreply@blogger.comBlogger2125tag:blogger.com,1999:blog-4446292666398344382.post-3286106753469242072013-01-10T15:07:28.964-08:002013-01-10T15:07:28.964-08:00Yes, that is a typo, thanks for noting (I'll p...Yes, that is a typo, thanks for noting (I'll preserve it so your comment continues to make sense ... exercise for the reader and all that ...)<br /><br />Re: comparison, I haven't done anything systematic. So much of data science is heuristic because domain variability. <br /><br />Note this technique corresponds to linear (quadratic with -q) spline interpolation. What you suggest sounds like piecewise linear (quadratic) but without enforcing continuity at the endpoints. If you suspect there should be jumps at your control points then maybe it would work better.Paul Mineirohttps://www.blogger.com/profile/05439062526157173163noreply@blogger.comtag:blogger.com,1999:blog-4446292666398344382.post-34156390324176690232013-01-10T11:28:01.725-08:002013-01-10T11:28:01.725-08:00Interesting technique! I think there's a typo...Interesting technique! I think there's a typo: should the weight on LOG2_DISTANCE be .17 instead of .19?<br /><br />Have you done much comparison with other methods of doing this type of thing? For example, what about just taking the closest point, and making the feature value the difference or percent change between the discretized value and the actual value?David Rosenberghttps://www.blogger.com/profile/11804108370682039789noreply@blogger.com