Random Forest

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 randomForest() model using the mtcars dataset:

library(dplyr)
library(tidypredict)
library(randomForest)

model <- randomForest(mpg ~ ., data = mtcars, ntree = 5, proximity = TRUE)

Under the hood

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

getTree(model, labelVar = TRUE) %>%
  head()
#>   left daughter right daughter split var split point status prediction
#> 1             2              3      carb       1.500     -3   20.12813
#> 2             4              5      gear       3.500     -3   28.82222
#> 3             6              7      disp     221.700     -3   16.72609
#> 4             0              0      <NA>       0.000     -1   18.10000
#> 5             8              9        hp      65.500     -3   31.88571
#> 6            10             11        wt       3.295     -3   21.38333

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(gear < 3.5 & carb < 1.5 ~ 18.1, hp < 65.5 & gear >= 
#>     3.5 & carb < 1.5 ~ 33.9, wt < 3.295 & disp < 221.7 & carb >= 
#>     1.5 ~ 21.82, wt >= 3.295 & disp < 221.7 & carb >= 1.5 ~ 19.2, 
#>     wt >= 4.66 & disp >= 221.7 & carb >= 1.5 ~ 10.4, qsec < 19.185 & 
#>         hp >= 65.5 & gear >= 3.5 & carb < 1.5 ~ 27.3, qsec >= 
#>         19.185 & hp >= 65.5 & gear >= 3.5 & carb < 1.5 ~ 32.4, 
#>     qsec < 16.96 & wt < 4.66 & disp >= 221.7 & carb >= 1.5 ~ 
#>         14.94, hp < 162.5 & qsec >= 16.96 & wt < 4.66 & disp >= 
#>         221.7 & carb >= 1.5 ~ 15.2, carb < 2.5 & hp >= 162.5 & 
#>         qsec >= 16.96 & wt < 4.66 & disp >= 221.7 & carb >= 1.5 ~ 
#>         19.2, .default = 16.34) + case_when(hp < 65.5 & hp < 
#>     95 ~ 31.5666666666667, wt < 2.23 & hp >= 65.5 & hp < 95 ~ 
#>     26.325, wt >= 2.23 & hp >= 65.5 & hp < 95 ~ 22.8, qsec < 
#>     17.62 & hp >= 190 & hp >= 95 ~ 14.325, qsec >= 17.62 & hp >= 
#>     190 & hp >= 95 ~ 10.4, hp < 130 & drat < 3.385 & hp < 190 & 
#>     hp >= 95 ~ 19.2, hp >= 130 & drat < 3.385 & hp < 190 & hp >= 
#>     95 ~ 17.025, gear >= 4.5 & drat >= 3.385 & hp < 190 & hp >= 
#>     95 ~ 19.7, vs < 0.5 & gear < 4.5 & drat >= 3.385 & hp < 190 & 
#>     hp >= 95 ~ 21, .default = 21.5) + case_when(qsec < 18.755 & 
#>     disp < 142.9 ~ 22.8, qsec >= 18.755 & disp < 142.9 ~ 28.74, 
#>     disp < 163.8 & wt < 3.49 & disp >= 142.9 ~ 20.1333333333333, 
#>     qsec >= 17.71 & wt >= 3.49 & disp >= 142.9 ~ 10.4, gear < 
#>         3.5 & disp >= 163.8 & wt < 3.49 & disp >= 142.9 ~ 19.2, 
#>     gear >= 3.5 & disp >= 163.8 & wt < 3.49 & disp >= 142.9 ~ 
#>         17.4, drat < 2.915 & hp < 212.5 & qsec < 17.71 & wt >= 
#>         3.49 & disp >= 142.9 ~ 15.5, drat >= 2.915 & hp < 212.5 & 
#>         qsec < 17.71 & wt >= 3.49 & disp >= 142.9 ~ 18.25, carb < 
#>         6 & hp >= 212.5 & qsec < 17.71 & wt >= 3.49 & disp >= 
#>         142.9 ~ 13.7, .default = 15) + case_when(cyl < 5 & am < 
#>     0.5 ~ 22.55, wt < 2.2775 & am >= 0.5 ~ 29.78, wt >= 2.2775 & 
#>     am >= 0.5 ~ 18.42, hp >= 192.5 & cyl >= 5 & am < 0.5 ~ 13.84, 
#>     hp < 136.5 & hp < 192.5 & cyl >= 5 & am < 0.5 ~ 18.5, drat < 
#>         3.11 & hp >= 136.5 & hp < 192.5 & cyl >= 5 & am < 0.5 ~ 
#>         17.075, .default = 17.3) + case_when(cyl >= 5 & hp < 
#>     116.5 ~ 21.1, qsec >= 18.14 & hp >= 116.5 ~ 18.2666666666667, 
#>     drat >= 4.165 & cyl < 5 & hp < 116.5 ~ 32.15, hp < 77.5 & 
#>         drat < 4.165 & cyl < 5 & hp < 116.5 ~ 24.4, disp < 337.9 & 
#>         hp < 192.5 & qsec < 18.14 & hp >= 116.5 ~ 16.76, disp >= 
#>         337.9 & hp < 192.5 & qsec < 18.14 & hp >= 116.5 ~ 19.2, 
#>     drat < 3.105 & hp >= 192.5 & qsec < 18.14 & hp >= 116.5 ~ 
#>         10.4, drat >= 3.105 & hp >= 192.5 & qsec < 18.14 & hp >= 
#>         116.5 ~ 13.975, disp < 114.05 & hp >= 77.5 & drat < 4.165 & 
#>         cyl < 5 & hp < 116.5 ~ 22.8, .default = 21.7))/5L

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

parsnip

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

library(parsnip)

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

tidypredict_fit(parsnip_model)
#> (case_when(wt < 2.0675 & disp < 107.6 ~ 30.5, wt >= 2.0675 & 
#>     disp < 107.6 ~ 32.4, drat >= 4.175 & wt < 3.16 & disp >= 
#>     107.6 ~ 26, hp < 96 & drat < 4.175 & wt < 3.16 & disp >= 
#>     107.6 ~ 22.8, hp >= 96 & drat < 4.175 & wt < 3.16 & disp >= 
#>     107.6 ~ 20.925, qsec < 18.6 & cyl < 7 & wt >= 3.16 & disp >= 
#>     107.6 ~ 19.2, qsec >= 18.6 & cyl < 7 & wt >= 3.16 & disp >= 
#>     107.6 ~ 17.9, wt < 3.4375 & hp < 177.5 & cyl >= 7 & wt >= 
#>     3.16 & disp >= 107.6 ~ 15.2, drat < 3.14 & hp >= 177.5 & 
#>     cyl >= 7 & wt >= 3.16 & disp >= 107.6 ~ 16.55, drat >= 3.14 & 
#>     hp >= 177.5 & cyl >= 7 & wt >= 3.16 & disp >= 107.6 ~ 15.05, 
#>     drat < 2.92 & wt >= 3.4375 & hp < 177.5 & cyl >= 7 & wt >= 
#>         3.16 & disp >= 107.6 ~ 15.5, .default = 18.8) + case_when(hp < 
#>     79.5 ~ 29.8, disp < 142.9 & cyl < 7 & hp >= 79.5 ~ 22.125, 
#>     disp >= 456 & cyl >= 7 & hp >= 79.5 ~ 10.4, hp < 116.5 & 
#>         disp >= 142.9 & cyl < 7 & hp >= 79.5 ~ 21.1333333333333, 
#>     drat >= 3.115 & disp < 456 & cyl >= 7 & hp >= 79.5 ~ 15.34, 
#>     qsec < 18.6 & hp >= 116.5 & disp >= 142.9 & cyl < 7 & hp >= 
#>         79.5 ~ 19.5, qsec >= 18.6 & hp >= 116.5 & disp >= 142.9 & 
#>         cyl < 7 & hp >= 79.5 ~ 17.8, drat < 3.075 & drat < 3.115 & 
#>         disp < 456 & cyl >= 7 & hp >= 79.5 ~ 15.725, .default = 19.2) + 
#>     case_when(wt < 3.3125 & drat < 3.75 ~ 22.05, wt < 2.23 & 
#>         drat >= 3.75 ~ 29.6, drat < 3.035 & wt >= 3.3125 & drat < 
#>         3.75 ~ 10.4, gear >= 4.5 & wt >= 2.23 & drat >= 3.75 ~ 
#>         15.8, carb < 2.5 & drat >= 3.035 & wt >= 3.3125 & drat < 
#>         3.75 ~ 16.3666666666667, vs < 0.5 & gear < 4.5 & wt >= 
#>         2.23 & drat >= 3.75 ~ 21, hp >= 290 & carb >= 2.5 & drat >= 
#>         3.035 & wt >= 3.3125 & drat < 3.75 ~ 15, qsec < 0 & vs >= 
#>         0.5 & gear < 4.5 & wt >= 2.23 & drat >= 3.75 ~ 24.4, 
#>         qsec >= 0 & vs >= 0.5 & gear < 4.5 & wt >= 2.23 & drat >= 
#>             3.75 ~ 22.8, disp < 395 & hp < 290 & carb >= 2.5 & 
#>             drat >= 3.035 & wt >= 3.3125 & drat < 3.75 ~ 13.68, 
#>         .default = 14.7) + case_when(wt >= 2.26 & cyl < 5 ~ 22.56, 
#>     qsec < 16.8 & wt < 2.26 & cyl < 5 ~ 26, qsec >= 16.8 & wt < 
#>         2.26 & cyl < 5 ~ 31.5, qsec >= 19.83 & disp < 266.9 & 
#>         cyl >= 5 ~ 18.1, qsec >= 17.71 & disp >= 266.9 & cyl >= 
#>         5 ~ 12.8, hp < 116.5 & qsec < 19.83 & disp < 266.9 & 
#>         cyl >= 5 ~ 21.08, hp >= 116.5 & qsec < 19.83 & disp < 
#>         266.9 & cyl >= 5 ~ 19.2, drat >= 3.19 & qsec < 17.71 & 
#>         disp >= 266.9 & cyl >= 5 ~ 15.25, drat < 3.075 & drat < 
#>         3.19 & qsec < 17.71 & disp >= 266.9 & cyl >= 5 ~ 16.4, 
#>     .default = 18.95) + case_when(hp < 80.5 ~ 29.9, gear >= 4.5 & 
#>     carb < 2.5 & hp >= 80.5 ~ 30.4, hp < 192.5 & carb >= 2.5 & 
#>     hp >= 80.5 ~ 15.725, disp < 182.9 & gear < 4.5 & carb < 2.5 & 
#>     hp >= 80.5 ~ 21.7, hp < 222.5 & hp >= 192.5 & carb >= 2.5 & 
#>     hp >= 80.5 ~ 10.4, wt < 3.3275 & disp >= 182.9 & gear < 4.5 & 
#>     carb < 2.5 & hp >= 80.5 ~ 21.4, drat < 3.88 & hp >= 222.5 & 
#>     hp >= 192.5 & carb >= 2.5 & hp >= 80.5 ~ 14.925, drat >= 
#>     3.88 & hp >= 222.5 & hp >= 192.5 & carb >= 2.5 & hp >= 80.5 ~ 
#>     15.8, cyl < 7 & wt >= 3.3275 & disp >= 182.9 & gear < 4.5 & 
#>     carb < 2.5 & hp >= 80.5 ~ 18.1, disp < 380 & cyl >= 7 & wt >= 
#>     3.3275 & disp >= 182.9 & gear < 4.5 & carb < 2.5 & hp >= 
#>     80.5 ~ 18.7, .default = 19.2))/5L