tidy xgboost

library(autostats)
library(workflows)
library(dplyr)
library(tune)
library(rsample)
library(hardhat)
library(broom.mixed)
library(Ckmeans.1d.dp)
library(igraph)

autostats provides convenient wrappers for modeling, visualizing, and predicting using a tidy workflow. The emphasis is on rapid iteration and quick results using an intuitive interface based off the tibble and tidy_formula.

Prepare data

Set up the iris data set for modeling. Create dummies and any new columns before making the formula. This way the same formula can be use throughout the modeling and prediction process.

set.seed(34)

 iris %>%
  dplyr::as_tibble() %>% 
  framecleaner::create_dummies(remove_first_dummy  = TRUE) -> iris1
#> 1 column(s) have become 2 dummy columns

 iris1 %>%
 tidy_formula(target = Petal.Length) -> petal_form
 
 petal_form
#> Petal.Length ~ Sepal.Length + Sepal.Width + Petal.Width + species_versicolor + 
#>     species_virginica
#> <environment: 0x56377a04b268>

Use the rsample package to split into train and validation sets.

iris1 %>%
  rsample::initial_split() -> iris_split

iris_split %>%
  rsample::analysis() -> iris_train

iris_split %>%
  rsample::assessment() -> iris_val

iris_split
#> <Training/Testing/Total>
#> <112/38/150>

Fit boosting models and visualize

Fit models to the training set using the formula to predict Petal.Length. Variable importance using gain for each xgboost model can be visualized.

xgboost with grid search hyperparameter optimization

auto_tune_xgboost returns a workflow object with tuned parameters and requires some postprocessing to get a trained xgb.Booster object like tidy_xgboost.

xgboost also can be tuned using a grid that is created internally using dials::grid_max_entropy. The n_iter parameter is passed to grid_size. Parallelization is highly effective in this method, so the default argument parallel = TRUE is recommended.

iris_train %>%
  auto_tune_xgboost(formula = petal_form, n_iter = 5L,trees = 20L, loss_reduction = 2, mtry = 3, tune_method = "grid", parallel = FALSE, counts = TRUE) -> xgb_tuned_grid
#> i Fold1: preprocessor 1/1
#> i Fold1: preprocessor 1/1 (prediction data)
#> i Fold1: preprocessor 1/1, model 1/1
#> i Fold1: preprocessor 1/1, model 1/1 (predictions)
#> i Fold2: preprocessor 1/1
#> i Fold2: preprocessor 1/1 (prediction data)
#> i Fold2: preprocessor 1/1, model 1/1
#> i Fold2: preprocessor 1/1, model 1/1 (predictions)
#> i Fold3: preprocessor 1/1
#> i Fold3: preprocessor 1/1 (prediction data)
#> i Fold3: preprocessor 1/1, model 1/1
#> i Fold3: preprocessor 1/1, model 1/1 (predictions)
#> i Fold4: preprocessor 1/1
#> i Fold4: preprocessor 1/1 (prediction data)
#> i Fold4: preprocessor 1/1, model 1/1
#> i Fold4: preprocessor 1/1, model 1/1 (predictions)
#> i Fold5: preprocessor 1/1
#> i Fold5: preprocessor 1/1 (prediction data)
#> i Fold5: preprocessor 1/1, model 1/1
#> i Fold5: preprocessor 1/1, model 1/1 (predictions)
#> i Fold1: preprocessor 1/1
#> i Fold1: preprocessor 1/1 (prediction data)
#> i Fold1: preprocessor 1/1, model 1/1
#> i Fold1: preprocessor 1/1, model 1/1 (predictions)
#> i Fold2: preprocessor 1/1
#> i Fold2: preprocessor 1/1 (prediction data)
#> i Fold2: preprocessor 1/1, model 1/1
#> i Fold2: preprocessor 1/1, model 1/1 (predictions)
#> i Fold3: preprocessor 1/1
#> i Fold3: preprocessor 1/1 (prediction data)
#> i Fold3: preprocessor 1/1, model 1/1
#> i Fold3: preprocessor 1/1, model 1/1 (predictions)
#> i Fold4: preprocessor 1/1
#> i Fold4: preprocessor 1/1 (prediction data)
#> i Fold4: preprocessor 1/1, model 1/1
#> i Fold4: preprocessor 1/1, model 1/1 (predictions)
#> i Fold5: preprocessor 1/1
#> i Fold5: preprocessor 1/1 (prediction data)
#> i Fold5: preprocessor 1/1, model 1/1
#> i Fold5: preprocessor 1/1, model 1/1 (predictions)
#> i Fold1: preprocessor 1/1
#> i Fold1: preprocessor 1/1 (prediction data)
#> i Fold1: preprocessor 1/1, model 1/1
#> i Fold1: preprocessor 1/1, model 1/1 (predictions)
#> i Fold2: preprocessor 1/1
#> i Fold2: preprocessor 1/1 (prediction data)
#> i Fold2: preprocessor 1/1, model 1/1
#> i Fold2: preprocessor 1/1, model 1/1 (predictions)
#> i Fold3: preprocessor 1/1
#> i Fold3: preprocessor 1/1 (prediction data)
#> i Fold3: preprocessor 1/1, model 1/1
#> i Fold3: preprocessor 1/1, model 1/1 (predictions)
#> i Fold4: preprocessor 1/1
#> i Fold4: preprocessor 1/1 (prediction data)
#> i Fold4: preprocessor 1/1, model 1/1
#> i Fold4: preprocessor 1/1, model 1/1 (predictions)
#> i Fold5: preprocessor 1/1
#> i Fold5: preprocessor 1/1 (prediction data)
#> i Fold5: preprocessor 1/1, model 1/1
#> i Fold5: preprocessor 1/1, model 1/1 (predictions)
#> i Fold1: preprocessor 1/1
#> i Fold1: preprocessor 1/1 (prediction data)
#> i Fold1: preprocessor 1/1, model 1/1
#> i Fold1: preprocessor 1/1, model 1/1 (predictions)
#> i Fold2: preprocessor 1/1
#> i Fold2: preprocessor 1/1 (prediction data)
#> i Fold2: preprocessor 1/1, model 1/1
#> i Fold2: preprocessor 1/1, model 1/1 (predictions)
#> i Fold3: preprocessor 1/1
#> i Fold3: preprocessor 1/1 (prediction data)
#> i Fold3: preprocessor 1/1, model 1/1
#> i Fold3: preprocessor 1/1, model 1/1 (predictions)
#> i Fold4: preprocessor 1/1
#> i Fold4: preprocessor 1/1 (prediction data)
#> i Fold4: preprocessor 1/1, model 1/1
#> i Fold4: preprocessor 1/1, model 1/1 (predictions)
#> i Fold5: preprocessor 1/1
#> i Fold5: preprocessor 1/1 (prediction data)
#> i Fold5: preprocessor 1/1, model 1/1
#> i Fold5: preprocessor 1/1, model 1/1 (predictions)
#> i Fold1: preprocessor 1/1
#> i Fold1: preprocessor 1/1 (prediction data)
#> i Fold1: preprocessor 1/1, model 1/1
#> i Fold1: preprocessor 1/1, model 1/1 (predictions)
#> i Fold2: preprocessor 1/1
#> i Fold2: preprocessor 1/1 (prediction data)
#> i Fold2: preprocessor 1/1, model 1/1
#> i Fold2: preprocessor 1/1, model 1/1 (predictions)
#> i Fold3: preprocessor 1/1
#> i Fold3: preprocessor 1/1 (prediction data)
#> i Fold3: preprocessor 1/1, model 1/1
#> i Fold3: preprocessor 1/1, model 1/1 (predictions)
#> i Fold4: preprocessor 1/1
#> i Fold4: preprocessor 1/1 (prediction data)
#> i Fold4: preprocessor 1/1, model 1/1
#> i Fold4: preprocessor 1/1, model 1/1 (predictions)
#> i Fold5: preprocessor 1/1
#> i Fold5: preprocessor 1/1 (prediction data)
#> i Fold5: preprocessor 1/1, model 1/1
#> i Fold5: preprocessor 1/1, model 1/1 (predictions)

xgb_tuned_grid %>%
  parsnip::fit(iris_train) %>% 
  parsnip::extract_fit_engine() -> xgb_tuned_fit_grid


xgb_tuned_fit_grid %>% 
  visualize_model()

xgboost with default parameters

iris_train %>%
  tidy_xgboost(formula = petal_form) -> xgb_base
#> accuracy tested on a validation set
#> # A tibble: 3 × 2
#>   .metric .estimate
#>   <chr>       <dbl>
#> 1 ccc         0.688
#> 2 rmse        1.18 
#> 3 rsq         0.528

xgboost with hand-picked parameters

iris_train %>% 
  tidy_xgboost(petal_form, 
               trees = 250L, 
               tree_depth = 3L, 
               sample_size = .5,
               mtry = .5,
               min_n = 2) -> xgb_opt
#> accuracy tested on a validation set
#> # A tibble: 3 × 2
#>   .metric .estimate
#>   <chr>       <dbl>
#> 1 ccc         0.960
#> 2 rmse        0.450
#> 3 rsq         0.954

predict on validation set

make predictions

Predictions are iteratively added to the validation data frame. The name of the column is automatically created using the models name and the prediction target.


xgb_base %>%
  tidy_predict(newdata = iris_val, form = petal_form) -> iris_val2
#> created the following column: Petal.Length_preds_xgb_base

xgb_opt %>% 
  tidy_predict(newdata = iris_val2, petal_form) -> iris_val3
#> created the following column: Petal.Length_preds_xgb_opt


iris_val3 %>% 
  names()
#> [1] "Sepal.Length"                "Sepal.Width"                
#> [3] "Petal.Length"                "Petal.Width"                
#> [5] "species_versicolor"          "species_virginica"          
#> [7] "Petal.Length_preds_xgb_base" "Petal.Length_preds_xgb_opt"

predictions with eval_preds

Instead of evaluationg these prediction 1 by 1, This step is automated with eval_preds. This function is specifically designed to evaluate predicted columns with names given from tidy_predict.

iris_val3 %>% 
  eval_preds() 
#> # A tibble: 6 × 5
#>   .metric .estimator .estimate model    target      
#>   <chr>   <chr>          <dbl> <chr>    <chr>       
#> 1 ccc     standard       0.979 xgb_base Petal.Length
#> 2 ccc     standard       0.975 xgb_opt  Petal.Length
#> 3 rmse    standard       0.350 xgb_base Petal.Length
#> 4 rmse    standard       0.376 xgb_opt  Petal.Length
#> 5 rsq     standard       0.972 xgb_base Petal.Length
#> 6 rsq     standard       0.964 xgb_opt  Petal.Length

get shapley values

tidy_shap has similar syntax to tidy_predict and can be used to get shapley values from xgboost models on a validation set.

xgb_base %>% 
  tidy_shap(newdata = iris_val, form = petal_form) -> shap_list
shap_list$shap_tbl
#> # A tibble: 38 × 6
#>    Sepal.Length Sepal.Width Petal.Width species_versicolor species_virginica
#>           <dbl>       <dbl>       <dbl>              <dbl>             <dbl>
#>  1      -0.430      -0.0284       -1.61             0.0597           -0.103 
#>  2      -0.430      -0.0284       -1.61             0.0597           -0.103 
#>  3       0.0385     -0.0531       -1.60             0.0585           -0.122 
#>  4      -0.430      -0.0284       -1.61             0.0597           -0.103 
#>  5      -0.426      -0.0284       -1.61             0.0573           -0.103 
#>  6      -0.430      -0.0284       -1.61             0.0597           -0.103 
#>  7      -0.454      -0.0284       -1.56             0.0597           -0.0950
#>  8      -0.430      -0.0284       -1.61             0.0597           -0.103 
#>  9      -0.430      -0.0284       -1.61             0.0597           -0.103 
#> 10      -0.410       0.153        -1.56             0.0597           -0.103 
#> # ℹ 28 more rows
#> # ℹ 1 more variable: `(Intercept)` <dbl>
shap_list$shap_summary
#> # A tibble: 6 × 5
#>   name                  cor     var     sum sum_abs
#>   <chr>               <dbl>   <dbl>   <dbl>   <dbl>
#> 1 (Intercept)        NA     0       139.     139.  
#> 2 Petal.Width         0.954 1.43      6.95    42.3 
#> 3 Sepal.Length        0.960 0.130     0.818   11.1 
#> 4 species_virginica   0.998 0.0181    0.754    4.93
#> 5 species_versicolor -0.935 0.0116    1.95     4.25
#> 6 Sepal.Width        -0.762 0.00238   0.134    1.11
shap_list$swarmplot

shap_list$scatterplots

understand xgboost with other functions from the original package

Overfittingin the base config may be related to growing deep trees.

 xgb_base %>% 
  xgboost::xgb.plot.deepness()

 xgb_base %>% 
  xgboost::xgb.plot.deepness()

Plot the first tree in the model. The small in terminal leaves suggests overfitting in the base model.

xgb_base %>% 
  xgboost::xgb.plot.tree(model = ., trees = 1)