diceR: consensus_evaluate – R documentation

Become an expert in R — Interactive courses, Cheat Sheets, certificates and more!

consensus_evaluate

Evaluate, trim, and reweigh algorithms

Description

Evaluates algorithms on internal/external validation indices. Poor performing algorithms can be trimmed from the ensemble. The remaining algorithms can be given weights before use in consensus functions.

Usage

consensus_evaluate(
  data,
  ...,
  cons.cl = NULL,
  ref.cl = NULL,
  k.method = NULL,
  plot = FALSE,
  trim = FALSE,
  reweigh = FALSE,
  n = 5
)

Arguments

`data`	data matrix with rows as samples and columns as variables
`...`	any number of objects outputted from `consensus_cluster()`
`cons.cl`	matrix of cluster assignments from consensus functions such as `kmodes` and `majority_voting`
`ref.cl`	reference class
`k.method`	determines the method to choose k when no reference class is given. When `ref.cl` is not `NULL`, k is the number of distinct classes of `ref.cl`. Otherwise the input from `k.method` chooses k. The default is to use the PAC to choose the best k(s). Specifying an integer as a user-desired k will override the best k chosen by PAC. Finally, specifying "all" will produce consensus results for all k. The "all" method is implicitly performed when there is only one k used.
`plot`	logical; if `TRUE`, `graph_all` is called
`trim`	logical; if `TRUE`, algorithms that score low on internal indices will be trimmed out
`reweigh`	logical; if `TRUE`, after trimming out poor performing algorithms, each algorithm is reweighed depending on its internal indices.
`n`	an integer specifying the top `n` algorithms to keep after trimming off the poor performing ones using Rank Aggregation. If the total number of algorithms is less than `n` no trimming is done.

Details

This function always returns internal indices. If ref.cl is not NULL, external indices are additionally shown. Relevant graphical displays are also outputted. Algorithms are ranked across internal indices using Rank Aggregation. Only the top n algorithms are kept, the rest are trimmed.

Value

consensus_evaluate returns a list with the following elements

k: if ref.cl is not NULL, this is the number of distinct classes in the reference; otherwise the chosen k is determined by the one giving the largest mean PAC across algorithms
pac: a data frame showing the PAC for each combination of algorithm and cluster size
ii: a list of data frames for all k showing internal evaluation indices
ei: a data frame showing external evaluation indices for k
trim.obj: A list with 4 elements
- alg.keep: algorithms kept
- alg.remove: algorithms removed
- rank.matrix: a matrix of ranked algorithms for every internal evaluation index
- top.list: final order of ranked algorithms
- E.new: A new version of a consensus_cluster data object

Examples

# Consensus clustering for multiple algorithms
set.seed(911)
x <- matrix(rnorm(500), ncol = 10)
CC <- consensus_cluster(x, nk = 3:4, reps = 10, algorithms = c("ap", "km"),
progress = FALSE)

# Evaluate algorithms on internal/external indices and trim algorithms:
# remove those ranking low on internal indices
set.seed(1)
ref.cl <- sample(1:4, 50, replace = TRUE)
z <- consensus_evaluate(x, CC, ref.cl = ref.cl, n = 1, trim = TRUE)
str(z, max.level = 2)

diceR

Diverse Cluster Ensemble in R

v1.0.3

MIT + file LICENSE

Authors

Derek Chiu [aut, cre], Aline Talhouk [aut], Johnson Liu [ctb, com]

Initial release