From: Chris T. <th...@sp...> - 2004-09-13 21:16:01
|
Ken Harris wrote: >That's a great speedup. Is there any noticable deterioration in output >quality? > > Ken Harris, I'm guessing that there isn't going to be any. The CEM is still run to completion, just as it was before. The only difference is in how the points are initially assigned to clusters (via k-means, versus randomly). I still have more testing to complete before much can be said about output quality, though the test dataset does complete with 100% correctness. The next step would be to consider implementing CEM on n-point spherical groupings of points instead of the full dataset, as you had suggested in an earlier posting. We could probably reduce the size of the dataset that CEM sees by an order of magnitude or more without much degradation to the output.. -Chris Thorp |