data-raw/05c0_rf_grid.md

In bennysalo/predict-recidivism: Analyses for prediction of recidivism in Finnish data using supervised learning

title: "Create rf_grid" author: "Benny Salo" date: "2019-02-14" output: github_document

library(dplyr)
devtools::load_all(".")

Get the part of model_grid that are random forest models.

rf_grid <- model_grid %>% filter(model_type == "Random forest")

In the first run we test eleven possible values for the tuning parameter mtry, including the often recommended square root of the number of predictors. We test five smaller and five bigger values in relation to this. The sequence of tested values are the number of predictors raised to the power of 1/12, 1/6, 1/4, 1/3, 5/12, 1/2, (i.e. the square root), 7/12, 2/3, 3/4, 5/6, and 11/12.

We create a new column for this argument

write_mtry_seq <- function(predictor_vector) {
  n_preds  <- length(predictor_vector)
  powers   <- (1:11)/12
  mtry_seq <- as.integer(round(n_preds^powers))
  # We could remvove possible dupicate mtrys
  # mtry_seq <- unique(mtry_seq)
}

rf_grid$mtry_seq <- 
  purrr::map(
    .x = rf_grid$rhs,
    .f = ~ write_mtry_seq(.x)
  )

Assertions

library(assertthat)
# All entries in rf_grid$mtry_seq should be of class integer
assert_that(all(purrr::map_chr(rf_grid$mtry_seq, class) == "integer"))

## [1] TRUE

# All entries in rf_grid$mtry_seq should have length 7.
# assert_that(all(purrr::map_chr(rf_grid$mtry_seq, length) == 11))

# The sixth element should be the sqrt of the number of predictors
sixth  <- purrr::map_int(rf_grid$mtry_seq, 6)
n_preds <- purrr::map_int(rf_grid$rhs, length)

assert_that(all(sixth == round(sqrt(n_preds))))

## [1] TRUE

Save and make available in /data

devtools::use_data(rf_grid, overwrite = TRUE)

## Warning: 'devtools::use_data' is deprecated.
## Use 'usethis::use_data()' instead.
## See help("Deprecated") and help("devtools-deprecated").

## <U+2714> Saving 'rf_grid' to 'data/rf_grid.rda'

Print sessionInfo

sessionInfo()

## R version 3.5.2 (2018-12-20)
## Platform: x86_64-w64-mingw32/x64 (64-bit)
## Running under: Windows >= 8 x64 (build 9200)
## 
## Matrix products: default
## 
## locale:
## [1] LC_COLLATE=Swedish_Finland.1252  LC_CTYPE=Swedish_Finland.1252   
## [3] LC_MONETARY=Swedish_Finland.1252 LC_NUMERIC=C                    
## [5] LC_TIME=Swedish_Finland.1252    
## 
## attached base packages:
## [1] stats     graphics  grDevices utils     datasets  methods   base     
## 
## other attached packages:
##  [1] recidivismsl_0.0.0.9000 assertthat_0.2.0       
##  [3] caret_6.0-81            lattice_0.20-38        
##  [5] bindrcpp_0.2.2          ggplot2_3.1.0          
##  [7] dplyr_0.7.8             testthat_2.0.1         
##  [9] purrr_0.2.5             magrittr_1.5           
## 
## loaded via a namespace (and not attached):
##  [1] nlme_3.1-137            fs_1.2.6               
##  [3] xopen_1.0.0             usethis_1.4.0          
##  [5] lubridate_1.7.4         devtools_2.0.1         
##  [7] rprojroot_1.3-2         tools_3.5.2            
##  [9] backports_1.1.3         utf8_1.1.4             
## [11] R6_2.3.0                rpart_4.1-13           
## [13] lazyeval_0.2.1          colorspace_1.4-0       
## [15] nnet_7.3-12             withr_2.1.2            
## [17] ResourceSelection_0.3-4 tidyselect_0.2.5       
## [19] prettyunits_1.0.2       processx_3.2.1         
## [21] compiler_3.5.2          glmnet_2.0-16          
## [23] cli_1.0.1               xml2_1.2.0             
## [25] desc_1.2.0              scales_1.0.0           
## [27] randomForest_4.6-14     readr_1.3.1            
## [29] callr_3.1.1             commonmark_1.7         
## [31] stringr_1.3.1           digest_0.6.18          
## [33] pkgconfig_2.0.2         sessioninfo_1.1.1      
## [35] highr_0.7               rlang_0.3.1            
## [37] ggthemes_4.0.1          rstudioapi_0.9.0       
## [39] bindr_0.1.1             generics_0.0.2         
## [41] ModelMetrics_1.2.2      Matrix_1.2-15          
## [43] Rcpp_1.0.0              munsell_0.5.0          
## [45] fansi_0.4.0             furniture_1.8.7        
## [47] stringi_1.2.4           pROC_1.13.0            
## [49] yaml_2.2.0              MASS_7.3-51.1          
## [51] pkgbuild_1.0.2          plyr_1.8.4             
## [53] recipes_0.1.4           grid_3.5.2             
## [55] forcats_0.3.0           crayon_1.3.4           
## [57] splines_3.5.2           hms_0.4.2              
## [59] knitr_1.21              ps_1.3.0               
## [61] pillar_1.3.1            reshape2_1.4.3         
## [63] codetools_0.2-15        clisymbols_1.2.0       
## [65] stats4_3.5.2            pkgload_1.0.2          
## [67] glue_1.3.0              evaluate_0.12          
## [69] data.table_1.12.0       remotes_2.0.2          
## [71] foreach_1.4.4           gtable_0.2.0           
## [73] rcmdcheck_1.3.2         tidyr_0.8.2            
## [75] xfun_0.4                gower_0.1.2            
## [77] prodlim_2018.04.18      roxygen2_6.1.1         
## [79] class_7.3-14            survival_2.43-3        
## [81] timeDate_3043.102       tibble_2.0.1           
## [83] iterators_1.0.10        memoise_1.1.0          
## [85] lava_1.6.4              ipred_0.9-8

bennysalo/predict-recidivism documentation built on May 29, 2019, 10:34 a.m.