paithiov909 / baritsu

Wrappers for the 'mlpack' R package; at times and in some situations, baritsu may be more powerful than a swiss army knife

baritsu

The main goal of baritsu is to implement wrappers for mlpack that allows formula as their argument. Also, baritsu provides parsnip engines of those wrappers, so they can be used with tidymodels workflows.

Installation

You can install the development version of baritsu from GitHub with:

remotes::install_github("paithiov909/baritsu")

Example

This is a basic example which shows you how to solve a common problem:

suppressPackageStartupMessages({
  require(tidymodels)
  require(baritsu)
})

data("penguins", package = "modeldata")

set.seed(1218)
data_split <- initial_split(penguins, strata = species)
penguins_train <- training(data_split)
penguins_test <- testing(data_split)

rec <-
  recipe(
    species ~ .,
    data = penguins_train
  ) |>
  step_impute_median(all_numeric_predictors()) |>
  step_impute_mode(all_nominal_predictors())

spec <-
  decision_tree(
    tree_depth = 0,
    min_n = 5
  ) |>
  set_engine("baritsu") |>
  set_mode("classification")

translate(spec)
#> Decision Tree Model Specification (classification)
#> 
#> Main Arguments:
#>   tree_depth = 0
#>   min_n = 5
#> 
#> Computational engine: baritsu 
#> 
#> Model fit template:
#> baritsu::decision_trees(formula = missing_arg(), data = missing_arg(), 
#>     x = missing_arg(), y = missing_arg(), tree_depth = 0, min_n = 5)

wf_fit <- workflow() |>
  add_recipe(rec) |>
  add_model(spec) |>
  fit(penguins_train)

pred <- augment(wf_fit, penguins_test) |>
  dplyr::select(species, .pred_class)

pred
#> # A tibble: 86 × 2
#>    species .pred_class
#>    <fct>   <fct>      
#>  1 Adelie  Adelie     
#>  2 Adelie  Adelie     
#>  3 Adelie  Adelie     
#>  4 Adelie  Adelie     
#>  5 Adelie  Adelie     
#>  6 Adelie  Adelie     
#>  7 Adelie  Adelie     
#>  8 Adelie  Adelie     
#>  9 Adelie  Adelie     
#> 10 Adelie  Adelie     
#> # ℹ 76 more rows

f_meas(pred, truth = species, estimate = .pred_class)
#> # A tibble: 1 × 3
#>   .metric .estimator .estimate
#>   <chr>   <chr>          <dbl>
#> 1 f_meas  macro          0.957