Talk:Vector quantization: Difference between revisions

Revision as of 08:48, 2 August 2013

Revision as of 19:48, 26 August 2013

Line 66:

E.g. clustering a 1-d normally distributioned data (10k samples) with k-means (6 clusters) results in groups with very different numbers of points assigned to each group (700 to 2400).

I would not call this difference "approximately the same". Or am i missing something?

== No mention of LBG or other methods ==

Article's "alternate training" method seems biased towards simulated annealing. No mention is made at all of the Linde–Buzo–Gray algorithm which is a fundamental starting point for most VQ implementations and is the most widely-cited paper in VQ work. No mention is made of PNN (Pair Nearest Neighbor) or other codebook generation methods either.