get_optimal_k: get_optimal_k

In charles-bernard/mage: Mine Associated Gene Expressions (from single-cell Rna Seq data)

Description

a wrapper to fpc::prediction.strength function. This function is based on consensus clustering using clara algorithm for different k (2 to 10 precisely).

The prediction strength is defined according to Tibshirani and Walther (2005) who recommand to choose as optimal number of clusters the largest number of clusters that leads to a prediction strength above a cutoff of 0.8 or 0.9.

Usage

get_optimal_k(scores_tab, ix = NULL, cutoff = 0.8, nrep = 20)

Arguments

scores_tab	data.table; the table of scores
ix	boolean vector of indexes to which scores_tab must be reduced (sampling output)
cutoff	cutoff for the prediction strength (recommanded to range from 0.8 up to 0.9)
nrep	nb of times the clustering is performed for each value of k (bootstrapping)

Value

the optimal number of clusters

charles-bernard/mage documentation built on May 14, 2019, 2 a.m.