DBSCANOptions<T>
Options & Configuration · AiDotNet.Clustering.Options
Configuration options for DBSCAN clustering.
Inheritance: ModelOptions → ClusteringOptions<T> → DBSCANOptions<T>
For Beginners
DBSCAN works by finding dense regions in your data.
Key concepts:
- Epsilon (eps): The maximum distance between two points to be neighbors
- MinPoints: Minimum neighbors needed to form a dense region
- Core point: Has at least MinPoints neighbors within epsilon
- Border point: Near a core point but not enough neighbors to be core
- Noise: Points that don’t belong to any cluster
Advantages over K-Means:
- No need to specify number of clusters
- Can find arbitrarily shaped clusters
- Robust to outliers (marks them as noise)
- Doesn’t assume spherical clusters
Choosing parameters:
- eps: Use a k-distance graph (elbow method)
- MinPoints: 2 × dimensions is a good starting point
How It Works
DBSCAN (Density-Based Spatial Clustering of Applications with Noise) finds clusters based on density. Unlike K-Means, it doesn’t require specifying the number of clusters and can discover clusters of arbitrary shape.
Constructors
| Constructor | Summary |
|---|---|
DBSCANOptions() |
Properties
| Property | Summary |
|---|---|
NeighborAlgorithm Algorithm { get; set; } |
Gets or sets the algorithm for computing core sample neighborhoods. |
bool AutoEpsilon { get; set; } |
When true, automatically estimates epsilon from the data using the k-distance method. |
double Epsilon { get; set; } |
Gets or sets the epsilon radius for neighborhood queries. |
double P { get; set; } |
Gets or sets the power parameter for Minkowski distance. |
int LeafSize { get; set; } |
Gets or sets the leaf size for BallTree or KDTree. |
int MinPoints { get; set; } |
Gets or sets the minimum number of points to form a dense region. |
int NumJobs { get; set; } |
Gets or sets the number of parallel jobs (-1 for all cores). |