DaMiR.ModelSelect: Select the best classification model

In BioinfoMonzino/DaMiRseq: Data Mining for RNA-seq data: normalization, feature selection and classification

Description

This function selects the bestmodels learned by the DaMiR.EnsL_Train and a tested by DaMiR.EnsL_Test.

Usage

DaMiR.ModelSelect(
  df,
  type.sel = c("mode", "median", "greater"),
  th.sel = 0.85,
  npred.sel = c("min", "rnd"),
  metric.idx = 1,
  npred.idx = 2
)

Arguments

df	A data frame of performance metrics. At least two columns representing a specific classification metrics (e.g., Accuracy) and the number of predictors must be provided. Additionally, other classification metrics (e.g., MCC, Sensitivity, Specificity,PPV, NPV, AUC, ...) can be appended (from the third column onwards) and used for the evaluation, by correctly setting the 'metric.idx' parameter.
type.sel	The method to select the best models. Only "mode","median" and "greater" values are allowed. For a specific classification metrics, "mode" selects all models whose score is the mode of all scores; "median" selects all models whose score is the median of all scores; and, "greater" selects all models whose score is greater than the value specified in "th.sel". Default: "mode".
th.sel	Threshold for the evaluation of the performance when "type.sel" is equal to "greater". Default: 0.85
npred.sel	The method to select the best model. Only "min" and "rnd" values are allowed. Taking into account the subset of models found by 'type.sel', this parameter selects one single model with the minimum number of predictors ("min") or randomly ("rnd"). Default: "min".
metric.idx	The index of the 'df' column (i.e., classification metrics) to be considered for the models evaluation. Default: 1.
npred.idx	The index of the 'df' column representing the number of predictors. Default: 2.

Details

This function finds the best model, taking into account specific classification metrics.

Value

The index of df (row), representing the model selected and a bubble chart

Author(s)

Mattia Chiesa, Luca Piacentini

Examples

# use example data:
set.seed(1)

BioinfoMonzino/DaMiRseq documentation built on Aug. 22, 2021, 3:11 p.m.