random_search_resample

This function finds the optimal parameters of an algorithm using random search

random_search_resample(
  y,
  tune_iters = NULL,
  resampling_method = NULL,
  ALGORITHM = NULL,
  grid_params = NULL,
  DATA = NULL,
  Args = NULL,
  regression = FALSE,
  re_run_params = FALSE,
  UNLABELED_TEST_DATA = NULL,
  ...
)

Arguments

y	a numeric vector
tune_iters	a number
resampling_method	one of 'bootstrap', 'train_test_split', 'cross_validation'
ALGORITHM	a list of parameters
grid_params	a grid of parameters in form of a list
DATA	a list including the data
Args	a list with further arguments of the function
regression	a boolean (TRUE, FALSE)
re_run_params	a boolean (TRUE, FALSE)
UNLABELED_TEST_DATA	either NULL or a data.frame ( matrix ) with the same number of columns as the initial train data
...	ellipsis to allow additional parameters

Value

a list of lists

Details

This function takes a number of arguments (including a grid of parameters) of an algorithm and using random search it returns a list of predictions and parameters for the chosen resampling method.

Author

Lampros Mouselimis

Examples

if (FALSE) {


#..........................
# MULTICLASS CLASSIFICATION
#..........................

library(kknn)
data(glass)

str(glass)

X = glass[, -c(1, dim(glass)[2])]
y1 = glass[, dim(glass)[2]]

form <- as.formula(paste('Type ~', paste(names(X),collapse = '+')))

y1 = c(1:length(unique(y1)))[ match(y1, sort(unique(y1))) ]       # labels should begin from 1:Inf
ALL_DATA = glass
ALL_DATA$Type = as.factor(y1)


#........................
# randomForest classifier
#........................

wrap_grid_args3 = list(ntree = seq(30, 50, 5), mtry = c(2:3), nodesize = seq(5, 15, 5))

res_rf = random_search_resample(as.factor(y1), tune_iters = 15,

                               resampling_method = list(method = 'cross_validation', 
                                                        repeats = NULL, 
                                                        sample_rate = NULL,
                                                        folds = 5),

                               ALGORITHM = list(package = require(randomForest), 
                                                algorithm = randomForest),

                               grid_params = wrap_grid_args3,

                               DATA = list(x = X, y = as.factor(y1)),

                               Args = NULL,

                               regression = FALSE, re_run_params = FALSE)


#............
# Logit boost
#............

#...........................
# RWeka::WOW("LogitBoost") : gives info for the parameters of the RWeka control list
#...........................


lb_lst = list(control = RWeka::Weka_control(H = c(1.0, 0.5), 
                                            I = seq(10, 30, 5), 
                                            Q = c(TRUE, FALSE), 
                                            O = 4))


res_log_boost = random_search_resample(as.factor(y1), tune_iters = 15,

                                      resampling_method = list(method = 'cross_validation',
                                                               repeats = NULL, 
                                                               sample_rate = NULL, 
                                                               folds = 5),

                                      ALGORITHM = list(package = require(RWeka), 
                                                       algorithm = LogitBoost),

                                      grid_params = lb_lst,

                                      DATA = list(formula = form, data = ALL_DATA),

                                      Args = NULL,

                                      regression = FALSE, re_run_params = FALSE)
}