13:25:36	 From Rasmus  : transactions celaring
13:39:32	 From Troels Christian Petersen : I can mention, that in estimating housing values “others” (i.e. SKAT) uses this simple technique.
13:40:06	 From Troels Christian Petersen : SKAT does it for square meter price… of course different size houses have different value.
13:45:13	 From Camilla : Would KNN have difficulties dealing with two clusters where the distances between the points is different? That is the points are very close in one cluster but far from each  other in the other?
13:46:22	 From Camilla : Because you have to fix k beforehand right, so a k good for one cluster might be bad for the other?
13:46:29	 From Troels Christian Petersen : No, it should not have too many problems with this, since it simply takes the closest points, no matter if they are near or far…
13:47:10	 From Troels Christian Petersen : Yes, it might be that different k’s are optimal for different parts of the phase space, i.e. the resolution changes. But that is beyond the kNN to deal with!
13:47:33	 From Camilla : Ah yes, I guess that makes sense - thanks :-)
13:47:38	 From Troels Christian Petersen : But, typically (and maybe surprisingly) this doesn’t happen to a very large degree.
13:52:53	 From Troels Christian Petersen : I see a triangle, but OK...
13:55:00	 From Troels Christian Petersen  To  Vadim Rusakov(privately) : We probably need to move on soon!
13:56:14	 From Troels Christian Petersen : This is generally called “Ensemble Learning”
13:57:10	 From Jonathan Elsborg : does this help with both regression and classification?
13:57:49	 From Troels Christian Petersen : Yes...
14:00:04	 From Bia Fonseca : can you do an educated guess for the initial position of the centroids or is it always random?
14:00:43	 From Troels Christian Petersen : Hmmmm… I would think that you can give such. The other things to “know” is the number of clusters, which again likes to have an educated guess…
14:05:19	 From Beatriz Campos Estrada : could you open breakout rooms please?