Mercurial > hg > camir-aes2014
annotate toolboxes/FullBNT-1.0.7/KPMstats/cwr_readme.txt @ 0:e9a9cd732c1e tip
first hg version after svn
| author | wolffd |
|---|---|
| date | Tue, 10 Feb 2015 15:05:51 +0000 |
| parents | |
| children |
| rev | line source |
|---|---|
| wolffd@0 | 1 This directory implements Cluster Weighted Regression, as described in |
| wolffd@0 | 2 Neil Gershenfeld, "The nature of mathematical modelling", p182. |
| wolffd@0 | 3 (See also http://www.media.mit.edu/physics/publications/books/nmm/files/index.html) |
| wolffd@0 | 4 |
| wolffd@0 | 5 Written by K. Murphy, 2 May 2003 |
| wolffd@0 | 6 |
| wolffd@0 | 7 The model is as follows: |
| wolffd@0 | 8 |
| wolffd@0 | 9 X<--| |
| wolffd@0 | 10 | Q |
| wolffd@0 | 11 v | |
| wolffd@0 | 12 Y<-- |
| wolffd@0 | 13 |
| wolffd@0 | 14 where Q is a discrete latent mixture variable. |
| wolffd@0 | 15 |
| wolffd@0 | 16 A mixture of experts has an X->Q arc instead of a Q->X arc; |
| wolffd@0 | 17 the X->Q arc is modelled by a softmax, which is slightly harder to fit than a |
| wolffd@0 | 18 mixture of Gaussians. |
| wolffd@0 | 19 |
| wolffd@0 | 20 |