Random Forest, using Ranger

Function Works
tidypredict_fit(), tidypredict_sql(), parse_model() ✔
tidypredict_to_column() ✗
tidypredict_test() ✗
tidypredict_interval(), tidypredict_sql_interval() ✗
parsnip ✔

How it works

Here is a simple ranger() model using the iris dataset:

library(dplyr)
library(tidypredict)
library(ranger)

model <- ranger(Species ~ ., data = iris, num.trees = 100)

Under the hood

The parser is based on the output from the ranger::treeInfo() function. It will return as many decision paths as there are non-NA rows in the prediction field.

treeInfo(model) %>%
  head()
#>   nodeID leftChild rightChild splitvarID splitvarName splitval terminal
#> 1      0         1          2          3  Petal.Width     0.75    FALSE
#> 2      1        NA         NA         NA         <NA>       NA     TRUE
#> 3      2         3          4          3  Petal.Width     1.75    FALSE
#> 4      3         5          6          0 Sepal.Length     7.10    FALSE
#> 5      4        NA         NA         NA         <NA>       NA     TRUE
#> 6      5         7          8          3  Petal.Width     1.65    FALSE
#>   prediction
#> 1       <NA>
#> 2     setosa
#> 3       <NA>
#> 4       <NA>
#> 5  virginica
#> 6       <NA>

The output from parse_model() is transformed into a dplyr, a.k.a Tidy Eval, formula. The entire decision tree becomes one dplyr::case_when() statement

tidypredict_fit(model)[1]
#> [[1]]
#> case_when(Petal.Width < 0.75 ~ "setosa", Petal.Width >= 1.75 & 
#>     Petal.Width >= 0.75 ~ "virginica", Sepal.Length >= 7.1 & 
#>     Petal.Width < 1.75 & Petal.Width >= 0.75 ~ "virginica", Petal.Length >= 
#>     5.35 & Petal.Width < 1.65 & Sepal.Length < 7.1 & Petal.Width < 
#>     1.75 & Petal.Width >= 0.75 ~ "virginica", Sepal.Width < 2.75 & 
#>     Petal.Width >= 1.65 & Sepal.Length < 7.1 & Petal.Width < 
#>     1.75 & Petal.Width >= 0.75 ~ "virginica", Sepal.Width >= 
#>     2.75 & Petal.Width >= 1.65 & Sepal.Length < 7.1 & Petal.Width < 
#>     1.75 & Petal.Width >= 0.75 ~ "versicolor", Petal.Width < 
#>     1.45 & Petal.Length < 5.35 & Petal.Width < 1.65 & Sepal.Length < 
#>     7.1 & Petal.Width < 1.75 & Petal.Width >= 0.75 ~ "versicolor", 
#>     Petal.Length < 5 & Petal.Width >= 1.45 & Petal.Length < 5.35 & 
#>         Petal.Width < 1.65 & Sepal.Length < 7.1 & Petal.Width < 
#>         1.75 & Petal.Width >= 0.75 ~ "versicolor", Sepal.Length < 
#>         6.15 & Petal.Length >= 5 & Petal.Width >= 1.45 & Petal.Length < 
#>         5.35 & Petal.Width < 1.65 & Sepal.Length < 7.1 & Petal.Width < 
#>         1.75 & Petal.Width >= 0.75 ~ "versicolor", Sepal.Length >= 
#>         6.15 & Petal.Length >= 5 & Petal.Width >= 1.45 & Petal.Length < 
#>         5.35 & Petal.Width < 1.65 & Sepal.Length < 7.1 & Petal.Width < 
#>         1.75 & Petal.Width >= 0.75 ~ "virginica")

From there, the Tidy Eval formula can be used anywhere where it can be operated. tidypredict provides three paths:

parsnip

tidypredict also supports ranger model objects fitted via the parsnip package.

library(parsnip)

parsnip_model <- rand_forest(mode = "classification") %>%
  set_engine("ranger") %>%
  fit(Species ~ ., data = iris)

tidypredict_fit(parsnip_model)[[1]]
#> case_when(Petal.Width < 0.7 ~ "setosa", Petal.Length < 4.85 & 
#>     Petal.Width >= 0.7 ~ "versicolor", Petal.Width < 1.75 & Petal.Length >= 
#>     4.85 & Petal.Width >= 0.7 ~ "versicolor", Petal.Width >= 
#>     1.75 & Petal.Length >= 4.85 & Petal.Width >= 0.7 ~ "virginica")