Timeline for Alternative to Otsu for dividing data into two groups
Current License: CC BY-SA 3.0
6 events
| when toggle format | what | by | license | comment | |
|---|---|---|---|---|---|
| Aug 30, 2012 at 19:49 | vote | accept | John Robertson | ||
| Aug 17, 2012 at 12:07 | comment | added | Andy W | Thanks for updating to make clear your critique of k-means. I still don't grok how minimizing variance within groups doesn't maximize the variance between groups. The wikipedia article is close to worthless. As I said before, there are multiple Jenk's methods. My copy of Slocum et al. (2005) doesn't even mention minimizing means! It only gives an example of minimizing deviations from group medians when talking about "Jenk's optimal method". | |
| Aug 17, 2012 at 3:27 | history | edited | Has QUIT--Anony-Mousse | CC BY-SA 3.0 |
added 267 characters in body
|
| Aug 17, 2012 at 3:26 | comment | added | Has QUIT--Anony-Mousse | Jenk's according to Wikipedia not only minimizes in-cluster variance, but also tries to maximize cross-cluster variances. K-means only does the first. Secondly, on the algorithmic side, k-means does test non-contiguous regions, which is unneccessary on an ordered data set, i.e. it wastes computations, lots of them (which is why you should use 1D methods for 1D data) Last but not least, there are other methods such as searching for local minima in kernel density estimation, too. | |
| Aug 16, 2012 at 18:31 | comment | added | Andy W | I thought Jenk's method of optimization was K-means in one dimension (see this comment by whuber). I believe there are multiple "Jenk's" methods though, so it could be some confusions around this. Your second comment also seems to be confused, if you use K-means clusters in one dimension it doesn't assign groups in non-contiguous rank order. | |
| Aug 16, 2012 at 16:05 | history | answered | Has QUIT--Anony-Mousse | CC BY-SA 3.0 |