BootstrapValidation<T>
Models & Types · AiDotNet.Clustering.Evaluation
Bootstrap validation for evaluating clustering quality and confidence.
For Beginners
Bootstrap asks “How confident are we in these results?”
By creating many random samples (with repeats allowed):
- We see how much our metrics vary
- We can give confidence intervals, not just single numbers
- We understand which results are reliable
If the bootstrap results vary a lot, be cautious about the clustering!
How It Works
Bootstrap validation uses resampling with replacement to estimate the uncertainty in clustering metrics. It provides confidence intervals for internal validation measures like silhouette score.
Algorithm:
- Create B bootstrap samples (sample with replacement)
- Cluster each bootstrap sample
- Compute metrics on each clustering
- Use the distribution to estimate confidence intervals
Constructors
| Constructor | Summary |
|---|---|
BootstrapValidation(int numBootstraps, int? randomState) |
Initializes a new BootstrapValidation instance. |
Methods
| Method | Summary |
|---|---|
BootstrapAnalysisResult EvaluateRange(Matrix<T> data, int minClusters, int maxClusters, double confidenceLevel) |
Evaluates bootstrap confidence intervals across a range of cluster counts. |
BootstrapResult Evaluate(Matrix<T> data, int numClusters, double confidenceLevel) |
Evaluates clustering with bootstrap confidence intervals. |
double[] ComputeAssignmentConfidence(Matrix<T> data, int numClusters) |
Evaluates cluster assignment confidence for each point. |