MDFS: Build MultiDimensional Feature Selector from IGs

Description

Build MultiDimensional Feature Selector from IGs

Usage

MDFS(IGs, dimensions, divisions, response_divisions = 1, IG_bits = TRUE, IG_doubled = FALSE, ignore_lowest = length(IGs)%/%10, variable_number = length(IGs), calc_variable_number = TRUE, mode_1D = "exp", min_variable_number = variable_number%/%2, max_ignore_lowest = variable_number%/%3, max_iterations = 20, acceptable_error = 0.05)

Arguments

IGs

max conditional information gains

dimensions

number of dimensions

divisions

number of divisions

response_divisions

number of response divisions (i.e. categories-1)

IG_bits

input is in binary log (as opposed to natural log)

IG_doubled

input is doubled (to follow the chi-squared distribution)

ignore_lowest

number of variables with the lowest IG to ignore (ignored if computed)

variable_number

number of irrelevant variables (ignored if computed)

calc_variable_number

whether to compute the number of neglected and irrelevant variables

mode_1D

"exp" - exponential distribution, "lin" - linear function of chi-squared, "raw" - raw chi-squared

min_variable_number

minimum number of irrelevant variables

max_ignore_lowest

maximum number of ignored variables

max_iterations

maximum number of iterations in variable number calculation

acceptable_error

acceptable error level for distribution parameter

Value

MDFS (list-based S3 class object) with the following named elements: "IGs" is a vector of information gains (input copy) "order" is a vector of ordinal numbers (order of variables by decreasing score) "chi.squared" is a vector of chi-squared p-values "p.values" is a vector of eventual p-values "scores" is a list of two vectors FDR and FWER with FDR and FWER scores respectively "lo.sq.dev." is a vector of square deviations used to calculate the number of ignored variables "hi.sq.dev." is a vector of square deviations used to calculate the number of irrelevant variables "ign.lowest" is a number of ignored variables "var.number" is a number of irrelevant variables "dist.param." is an exponential distribution parameter or linear coefficient "err.param." is a square error of the parameter