forward_selection_byrank {structToolbox}R Documentation

forward selection by rank

Description

Forward selection by rank is a stepwise procedures that includes features incrementally based on their rank. Any measure for ranking the features may be used e.g. PLS VIP score, ttest p-value etc.

Usage

forward_selection_byrank(
  min_no_vars = 1,
  max_no_vars = 100,
  step_size = 1,
  factor_name,
  variable_rank,
  ...
)

Arguments

min_no_vars

minimum number of features to test

max_no_vars

maximum numbe ro features to test

step_size

the size of the incremenent between min and max no of vars

factor_name

the sample-meta colum to use

variable_rank

a vector of values that can be used to rank the features, where the smallest value is the first rank.

...

additional slots and values passed to struct_class

Value

A struct object

Examples

# some data
D = MTBLS79_DatasetExperiment(filtered=TRUE)

# normalise, impute and scale then remove QCs
P = pqn_norm(qc_label='QC',factor_name='class') +
    knn_impute(neighbours=5) +
    glog_transform(qc_label='QC',factor_name='class') +
    filter_smeta(mode='exclude',levels='QC',factor_name='class')
P = model_apply(P,D)
D = predicted(P)

# forward selection using a PLSDA model
M = forward_selection_byrank(factor_name='class',
                             min_no_vars=2,
                             max_no_vars=11,
                             variable_rank=1:2063) *
    (mean_centre() + PLSDA(number_components=1,
                           factor_name='class'))
M = run(M,D,balanced_accuracy())


[Package structToolbox version 1.0.1 Index]