The goal of grideR is to provides general infrastructure based on data.table for grid search in model hyperparameters space. Key feature: all preprocessing steps are performed within resamples. This package aims to be as easy to extend as possible.
You can install dev version of grideR from GitHub with:
devtools::install_github("statist-bhfz/grideR")library(resampleR)
#> Loading required package: data.table
library(grideR)
library(parallel)
# Input data
dt <- as.data.table(mtcars)
# data.table with resamples
splits <- resampleR::cv_base(dt, "hp")
# data.table with tunable model hyperparameters
xgb_grid <- CJ(
max_depth = c(6, 8),
eta = 0.025,
colsample_bytree = 0.9,
subsample = 0.8,
gamma = 0,
min_child_weight = c(3, 5),
alpha = 0,
lambda = 1
)
# Non-tunable parameters for xgboost
xgb_args <- list(
nrounds = 500,
early_stopping_rounds = 10,
booster = "gbtree",
eval_metric = "rmse",
objective = "reg:linear",
verbose = 0
)
# Dumb preprocessing function
# Real function will contain imputation, feature engineering etc.
# with all statistics computed on train folds and applied to validation fold
preproc_fun_example <- function(data) return(data[])
cl <- makePSOCKcluster(2L)
clusterExport(cl, list("dt", "splits",
"xgb_grid", "xgb_args",
"preproc_fun_example"))
clusterEvalQ(cl, {
suppressMessages(library(resampleR))
suppressMessages(library(grideR))
suppressMessages(library(checkmate))
suppressMessages(library(data.table))
suppressMessages(library(xgboost))
})
#> [[1]]
#> [1] "xgboost" "checkmate" "grideR" "resampleR" "data.table"
#> [6] "stats" "graphics" "grDevices" "utils" "datasets"
#> [11] "methods" "base"
#>
#> [[2]]
#> [1] "xgboost" "checkmate" "grideR" "resampleR" "data.table"
#> [6] "stats" "graphics" "grDevices" "utils" "datasets"
#> [11] "methods" "base"
res <- clusterApply(
cl,
splits,
function(split) across_grid(data = dt,
target = "hp",
split = split,
fit_fun = xgb_fit,
preproc_fun = preproc_fun_example,
grid = xgb_grid,
args = xgb_args,
metrics = c("rmse", "mae")))
res <- rbindlist(res, idcol = "split")
res
#> split max_depth eta colsample_bytree subsample gamma
#> 1: 1 6 0.025 0.9 0.8 0
#> 2: 1 6 0.025 0.9 0.8 0
#> 3: 1 8 0.025 0.9 0.8 0
#> 4: 1 8 0.025 0.9 0.8 0
#> 5: 2 6 0.025 0.9 0.8 0
#> 6: 2 6 0.025 0.9 0.8 0
#> 7: 2 8 0.025 0.9 0.8 0
#> 8: 2 8 0.025 0.9 0.8 0
#> 9: 3 6 0.025 0.9 0.8 0
#> 10: 3 6 0.025 0.9 0.8 0
#> 11: 3 8 0.025 0.9 0.8 0
#> 12: 3 8 0.025 0.9 0.8 0
#> 13: 4 6 0.025 0.9 0.8 0
#> 14: 4 6 0.025 0.9 0.8 0
#> 15: 4 8 0.025 0.9 0.8 0
#> 16: 4 8 0.025 0.9 0.8 0
#> 17: 5 6 0.025 0.9 0.8 0
#> 18: 5 6 0.025 0.9 0.8 0
#> 19: 5 8 0.025 0.9 0.8 0
#> 20: 5 8 0.025 0.9 0.8 0
#> min_child_weight alpha lambda nrounds_best rmse mae
#> 1: 3 0 1 254 25.82795 23.442172
#> 2: 5 0 1 176 28.40845 24.490010
#> 3: 3 0 1 234 29.13382 26.868994
#> 4: 5 0 1 159 30.21812 26.535628
#> 5: 3 0 1 205 25.17265 20.216885
#> 6: 5 0 1 155 29.64170 25.587408
#> 7: 3 0 1 171 27.14264 20.503756
#> 8: 5 0 1 156 31.21857 27.003484
#> 9: 3 0 1 157 10.68297 9.809113
#> 10: 5 0 1 194 10.21284 8.511509
#> 11: 3 0 1 156 11.42183 10.083486
#> 12: 5 0 1 186 11.24998 9.353838
#> 13: 3 0 1 171 36.60020 27.660245
#> 14: 5 0 1 247 36.68939 25.642899
#> 15: 3 0 1 222 31.90733 27.354782
#> 16: 5 0 1 199 39.93131 27.828404
#> 17: 3 0 1 273 50.17284 31.402931
#> 18: 5 0 1 326 54.55717 32.347914
#> 19: 3 0 1 290 52.87341 30.706714
#> 20: 5 0 1 232 57.24948 33.237526
stopCluster(cl)