Random Forest, using Ranger • tidypredict

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 mtcars dataset:

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

model <- ranger(mpg ~ ., data = mtcars, num.trees = 5, max.depth = 2)

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
#> 1      0         1          2          8         gear     3.50
#> 2      1         3          4          2           hp   192.50
#> 3      2         5          6          4           wt     2.26
#> 4      3        NA         NA         NA         <NA>       NA
#> 5      4        NA         NA         NA         <NA>       NA
#> 6      5        NA         NA         NA         <NA>       NA
#>   terminal prediction
#> 1    FALSE         NA
#> 2    FALSE         NA
#> 3    FALSE         NA
#> 4     TRUE    17.4000
#> 5     TRUE    12.9000
#> 6     TRUE    29.4375

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

tidypredict_fit(model)
#> case_when(hp < 192.5 & gear < 3.5 ~ 17.4, hp >= 192.5 & gear < 
#>     3.5 ~ 12.9, wt < 2.26 & gear >= 3.5 ~ 29.4375, .default = 21.3625) + 
#>     case_when(vs < 0.5 & wt < 2.455 ~ 26, vs >= 0.5 & wt < 2.455 ~ 
#>         31.8, gear < 3.5 & wt >= 2.455 ~ 14.25, .default = 19.0666666666667) + 
#>     case_when(hp < 65.5 & disp < 120.65 ~ 31.275, hp >= 65.5 & 
#>         disp < 120.65 ~ 26.6333333333333, wt < 3.505 & disp >= 
#>         120.65 ~ 19.9533333333333, .default = 14.6857142857143) + 
#>     case_when(disp < 93.5 & cyl < 5 ~ 30.625, disp >= 93.5 & 
#>         cyl < 5 ~ 22.32, cyl < 7 & cyl >= 5 ~ 18.78, .default = 15.7769230769231) + 
#>     case_when(disp < 107.6 & cyl < 5 ~ 31.1333333333333, disp >= 
#>         107.6 & cyl < 5 ~ 23.6625, hp < 192.5 & cyl >= 5 ~ 18.65, 
#>         .default = 12.8625)

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

Use directly inside dplyr, mutate(iris, !! tidypredict_fit(model))
Use tidypredict_to_column(model) to a piped command set
Use tidypredict_to_sql(model) to retrieve the SQL statement

parsnip

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

library(parsnip)

parsnip_model <- rand_forest(mode = "regression", trees = 5) %>%
  set_engine("ranger", max.depth = 2) %>%
  fit(mpg ~ ., data = mtcars)

tidypredict_fit(parsnip_model)
#> case_when(disp < 450 & vs < 0.5 ~ 16.9684210526316, disp >= 450 & 
#>     vs < 0.5 ~ 10.4, drat < 4 & vs >= 0.5 ~ 24.2, .default = 30.6857142857143) + 
#>     case_when(wt < 2.3325 & hp < 131.5 ~ 31.6666666666667, wt >= 
#>         2.3325 & hp < 131.5 ~ 21.7571428571429, drat < 3.035 & 
#>         hp >= 131.5 ~ 12.1, .default = 16.3153846153846) + case_when(hp < 
#>     78.5 & cyl < 5 ~ 31.28, hp >= 78.5 & cyl < 5 ~ 24.4, wt < 
#>     4.747 & cyl >= 5 ~ 17.52, .default = 10.4) + case_when(carb < 
#>     2.5 & drat < 3.615 ~ 17.5875, carb >= 2.5 & drat < 3.615 ~ 
#>     13.4, wt < 2.23 & drat >= 3.615 ~ 28.8857142857143, .default = 20.4285714285714) + 
#>     case_when(wt < 2.2775 & disp < 266.9 ~ 31.375, wt >= 2.2775 & 
#>         disp < 266.9 ~ 22.4133333333333, wt < 4.747 & disp >= 
#>         266.9 ~ 16.5818181818182, .default = 10.4)