What's the common way to utilize distance functions for clustering?
Like does one set some thresholds for the distances and do grouping based on that?
Asked
Active
Viewed 21 times
1 Answers
0
I worked on K-means clustering,
you dont need any thresholds for clustering the data, your algorithm with club your data into appropriate buckets, for each input data , your bucket entries will change, only one thing you need to do : define $k$, i.e number of buckets , usually, $k = n^\frac{1}{2}, n=$ number of data set.
-
But how does one identify the clusters? Unless one limits the distances somehow? – mavavilj Jun 23 '18 at 11:51
-
p. 7 here: http://cs-people.bu.edu/evimaria/cs565/lect7.pdf. Shows that one creates the clusters by utilizing a distance function $f(X,d) = Γ$ that operates on partitions $Γ$ of the data set $X$. So in a sense, there's a "threshold". It's how the partitions are separated. – mavavilj Jun 23 '18 at 11:55
-
The way I coded : for each new entry i was changing the distance/threshold, and in the end depending upon the number of dataset, you will get optimal distance from each bucket/cluster. – ratnesh Jun 23 '18 at 12:07