ML_RMCL: Multi-Level Regularized Markov Clustering
In DavidGilgien/ML.RMCL: Multi-Level Regularized Markov Clustering

Description Usage Arguments Value Examples

This function implements the multi-level regularized markov clustering. It uses flow simulation of a graph to find clusters.

ML_RMCL(g, r = 2, reg = TRUE, iter.max = 200, n.threshold = 200,
  last.reg = TRUE, eps = 0.001, save.p1 = NULL, load.p1 = NULL,
  iter.curt = 4, coarse.weight = TRUE, init.weight.trans = FALSE,
  curt.weight.trans = FALSE, timer = FALSE)

`g`	graph object of class `ELgraph` on which to perform the clustering
`r`	inflation parameter. A higher value yields more refined clusters.
`reg`	boolean value. When `TRUE`, the MCL iterations in phase 2 are regularized.
`iter.max`	integer: maximum number of MCL iterations during phase 3 before the matrix is interpreted as a clustering. The function gives a warning when reached.
`n.threshold`	integer: number of nodes under which phase 1 stops the coarsening.
`last.reg`	boolean value. When `TRUE`, the MCL iterations in phase 3 are regularized.
`eps`	integer: epsilon threshold for the pruning operation. During pruning, values under `eps/m` are set to 0, where `m` is the number of non-zero entries in the column
`save.p1`	string: a name for the storing a the graph hierarchy after the coarsening. A file will be saved at "./phase1/`save.p1`.rds". The algorithm can then be used with this file as a strating point by giving the name as the `load.p1` parameter.
`load.p1`	string: a name for loading the graph hierarchy. When not `NULL`, `g` will not be used but the function will take the file "./phase1/`load.p1`.rds" as input. Can be used when phase 1 is time consuming and we want different tests with the same coarsening.
`iter.curt`	integer: number of iteration of MCL during phase 2.
`coarse.weight`	boolean value. When `TRUE`, the coarsening phase keeps the edges' weights as indications of the number of merged edges.
`init.weight.trans`	boolean value. When `TRUE`, the algorithm performs a weight transformation step at the very beginning, before the coarsening.
`curt.weight.trans`	boolean value. When `TRUE`, the algorithm performs a weight transformation step for each flow matrix initialisation during phase 2.
`timer`	boolean value. When `TRUE`, the algorithm returns a vector of phases duration in addition of the clustering.

A vector of size n describing the associated cluster of each node. When timer=TRUE, a list of two elements: res the clustering vector and time a vector of size 3 with the duration of each phase in seconds.