mlpack
Static Public Member Functions | Static Public Attributes | List of all members
SilhouetteScore Class Reference

The Silhouette Score is a metric of performance for clustering that represents the quality of clusters made as a result. More...

Static Public Member Functions

static double MeanDistanceFromCluster (const arma::colvec &distances, const arma::Row< size_t > &labels, const size_t &label, const bool &sameCluster=false)
 Find mean distance of element from a given cluster. More...

 
template
<
typename
DataType
,
typename
Metric
>
static double Overall (const DataType &X, const arma::Row< size_t > &labels, const Metric &metric)
 Find the overall silhouette score. More...

 
template
<
typename
DataType
>
static arma::rowvec SamplesScore (const DataType &distances, const arma::Row< size_t > &labels)
 Find the individual silhouette scores for precomputted dissimilarites. More...

 
template
<
typename
DataType
,
typename
Metric
>
static arma::rowvec SamplesScore (const DataType &X, const arma::Row< size_t > &labels, const Metric &metric)
 Find silhouette score of all individual elements. More...

 

Static Public Attributes

static const bool NeedsMinimization = false
 Information for hyper-parameter tuning code. More...

 

Detailed Description

The Silhouette Score is a metric of performance for clustering that represents the quality of clusters made as a result.

It provides an indication of goodness of fit and therefore a measure of how well unseen samples are likely to be predicted by the model, considering the inter-cluster and intra-cluster dissimilarities. Silhoutte Score is dependent on the metric used to calculate the dissimilarities. The best possible score is $ s(i) = 1.0 $. Smaller values of Silhouette Score indicate poor clustering. Negative values would occur when a wrong label was put on the element. Values near zero indicate overlapping clusters. For an element i $ a(i) $ is within cluster average dissimilarity and $ b(i) $ is minimum of average dissimilarity from other clusters. the Silhouette Score $ s(i) $ of a Sample is calculated by

\begin{eqnarray*} s(i) &=& \frac{b(i) - a(i)}{max\{b(i), a(i)\}} \end{eqnarray*}

The Overall Silhouette Score is the mean of individual silhoutte scores.

Definition at line 40 of file silhouette_score.hpp.

Member Function Documentation

◆ MeanDistanceFromCluster()

static double MeanDistanceFromCluster ( const arma::colvec &  distances,
const arma::Row< size_t > &  labels,
const size_t &  label,
const bool &  sameCluster = false 
)
static

Find mean distance of element from a given cluster.

Parameters
distancescolvec containing distances from other elements.
labelsLabels assigned to data by clustering.
labellabel of the target cluster.
sameClustertrue if calculating mean distance from same cluster.
Returns
(double) distance from the cluster.

◆ Overall()

static double Overall ( const DataType &  X,
const arma::Row< size_t > &  labels,
const Metric &  metric 
)
static

Find the overall silhouette score.

Parameters
XColumn-major data used for clustering.
labelsLabels assigned to data by clustering.
metricMetric to be used to calculate dissimilarity.
Returns
(double) silhouette score.

◆ SamplesScore() [1/2]

static arma::rowvec SamplesScore ( const DataType &  distances,
const arma::Row< size_t > &  labels 
)
static

Find the individual silhouette scores for precomputted dissimilarites.

Parameters
distancesSquare matrix containing distances between data points.
labelsLabels assigned to data by clustering.
Returns
(arma::rowvec) element-wise silhouette score.

◆ SamplesScore() [2/2]

static arma::rowvec SamplesScore ( const DataType &  X,
const arma::Row< size_t > &  labels,
const Metric &  metric 
)
static

Find silhouette score of all individual elements.

(Distance not precomputed).

Parameters
XColumn-major data used for clustering.
labelsLabels assigned to data by clustering.
metricMetric to be used to calculate dissimilarity.
Returns
(arma::rowvec) element-wise silhouette score.

Member Data Documentation

◆ NeedsMinimization

const bool NeedsMinimization = false
static

Information for hyper-parameter tuning code.

It indicates that we want to maximize the metric.

Definition at line 99 of file silhouette_score.hpp.

The documentation for this class was generated from the following file: