New answers tagged

0

The Mahalanobis distance takes into account the distribution of the existing points. This distribution affects whether or not a point is considered "close" in a particular dimension. If your points extend along a particular direction, say the x-axis for simplicity, then a reduced x-distance is used to determine if new points belong to the cluster. As ...


1

Some academic paper is giving a precise answer to that problem, under some separation assumptions (stability/noise resilience) on the clusters of the flat partition. The coarse idea of the paper solution is to extract the flat partition by cutting at different levels in the dendrogram. Say you want to minimize intra-cluster variance (that is your ...


2

The statement assumes that if the data contains discrete clusters, the axis or principal component of greatest variance is likely to separate those clusters. In the plot you've shown, the two principal components exhibit variances of 17.37 and 8.35. When the original data is projected onto the first component, which accounts for 17.37/(17.37+8.35) = 68% of ...


2

This is usually formulated a a supervised learning problem, however typically not classification. Instead such similarity models are most often trained with Triplet Loss. https://en.m.wikipedia.org/wiki/Triplet_loss The triplet loss setup is very much like your propose. The triplets consist of one sample (called anchor), as well as another sample of the same ...


1

I know it's a little bit late but I just wanted to say that I'm currently studying density based clustering algorithms, and I found out the most suitable metric was in fact DBCV: 1) it deals with noise (which is intrinsic to the definition of the density-based clustering, and it's not taken into account in indexes such as Silhouette or Davies Bouldin) 2) ...


0

UPDATE (Looking back on the OP, I would recommend running FKM and GMM, then try to publish the analysis based on that, or give a talk based on use of FKM/GMM. There is nothing wrong with use of FKM/GMM for your OP. You may be "missing" a lot other unknown/unpopular methods which develop probabilities -- so construction of list that misses nothing would ...


1

Each cluster should contain all datasets with a similar trend. Note that, this is your assumption, but can we verify this assumption in data? Verifying this assumption on data is a very important step. This is because it is possible that data driven results will conflict with your knowledge driven results, i.e, data collected from different nodes will ...


1

If I understand correctly, you can construct a new "dataset", in which each node (or original dataset) is one observation (point). You need to devise a metric for calculating the "distance" between the nodes. Since you say that some nodes share similar properties and the associated underlying datasets follow a similar trend, you can start from these notions ...


0

A step by step example of DBSCAN for teaching can be found in these lecture slides: https://dbs.ifi.uni-heidelberg.de/files/Team/eschubert/lectures/KDDClusterAnalysis17-screen.pdf#page=215 The newer versions of these slides are currently not online; but I intend to make them accessible eventually.


1

Probably the most widely used method is the Adjusted Rand Index (ARI). The CCR (and some others that you list) would assume that you know how to match the found clusters to the true ones, which is not necessarily trivial, particularly not if the numbers of clusters are not identical. I prefer something like the ARI because it doesn't depend on matching. ...


0

From Cha, Sung-Hyuk. "Comprehensive survey on distance/similarity measures between probability density functions." City 1.2 (2007): 1. You can read the reference for a technical explanation, but, intuitively, some things to point out: You (probably) want to be using an actual distance metric (https://en.wikipedia.org/wiki/Metric_(mathematics)) You ...


Top 50 recent answers are included