Awesome

DBSCAN

This is an implementation of the DBSCAN clustering algorithm in .NET. The primary advantage of this library over other DBSCAN implementations is that this library allows the use of spatial indexes, and is agnostic to the index.

Most implementations of DBSCAN use an O(N) search over every data point to find nearby data points. Sincethis search is executed for every point, the overall clustering time is O(N²). This is fine for smaller data sets, but better tools are needed to review larger data sets (say 100,000 points).

If we can provide a better index for searching for nearby data points, then we can reduce the overall time for clustering. For example, using an R-Tree or a K-d tree can reduce time to find neighbors to O(log n), which cuts the overall clustering to O(n log n).

A default spatial index using the above is provided in the primary library as ListSpatialIndex<T>. Alternatively, an R-Tree library is available in the secondary library as RBushSpatialIndex<T>.

DBSCAN:

DBSCAN.RBush:

Install

Install with Nuget:

• Primary library: Install-Package DBSCAN
• Secondary library: Install-Package DBSCAN.RBush

Usage

Convenience functions have been provided in both libraries. Call the CalculateClusters() function in either DBSCAN or DBSCAN.RBush.

var clusters = Dbscan.CalculateClusters(
    data,
    epsilon: 1.0,
    minimumPointsPerCluster: 4);

or

var clusters = DbscanRBush.CalculateClusters(
    data,
    epsilon: 1.0,
    minimumPointsPerCluster: 4);

If you have implemented an alternative ISpatialIndex<T>, then you can provide the index directly to DBSCAN.CalculateClusters():

var clusters = DBSCAN.CalculateClusters(
    index,
    epsilon: 1.0,
    minimumPointsPerCluster: 4);

Development

Clone the repository and open DBSCAN.sln in Visual Studio.

Compatibility

DBSCAN should run on any .NET system that supports .NET Standard 1.2 (.NET Framework 4.5.1 or later; .NET Core 1.0 or later).