HierarchicalClusteringOptions Properties

An upper limit on the number of threads the clustering algorithm will use.

By default the algorithm will use the same number of threads that there are cores in the computer. With this parameter it is possible to lower the number of threads used.

This is a hint to the clustering algorithm how much physical memory it can use. For large problems different internal algorithms will be selected depending on this value. The fastest algorithm is in most cases based on storing the entire distance matrix.

The distance matrix for hierarchical clustering is growing qudratically in the number of elements in the input. Thus for large input data all of the distance matrix will not fit in memory. Given 500 MBytes the algorithm can store the distance matrix up to around 15000 input elements. For larger input, HierarchicalClustering is using caching and recalculation algorithms based on the distance measure and clustering method. These can be slower, but are much faster than a full distance matrix algorithm that is swapping.

Raising the value too high compared to the physical memory in the computer will lead to out of memory exit and/or swapping on large problems. 500 MBytes is a good value on a 2 GByte computer.

If true a subprocess will be used for the clusterng computation.