XGBoost Exercises in R: 20 Practice Problems
Twenty practice problems on XGBoost in R: regression, classification, tuning, early stopping, feature importance, SHAP. Hidden solutions.
RRun this once before any exercise
library(xgboost)
library(caret)
Exercise 1: Regression with xgb.train
Difficulty: Intermediate.
Show solution
RInteractive R
x <- as.matrix(mtcars[, -1]); y <- mtcars$mpg
d <- xgb.DMatrix(x, label = y)
fit <- xgb.train(params = list(objective = "reg:squarederror"),
data = d, nrounds = 100, verbose = 0)
Exercise 2: Binary classification
Difficulty: Intermediate.
Show solution
RInteractive R
x <- as.matrix(mtcars[, c("mpg","hp","wt")]); y <- mtcars$am
d <- xgb.DMatrix(x, label = y)
fit <- xgb.train(params = list(objective = "binary:logistic", eval_metric = "logloss"),
data = d, nrounds = 100, verbose = 0)
Exercise 3: Multiclass
Difficulty: Advanced.
Show solution
RInteractive R
x <- as.matrix(iris[, 1:4]); y <- as.integer(iris$Species) - 1
d <- xgb.DMatrix(x, label = y)
fit <- xgb.train(params = list(objective = "multi:softprob", num_class = 3),
data = d, nrounds = 50, verbose = 0)
Exercise 4: Set learning rate (eta)
Difficulty: Intermediate.
Show solution
RInteractive R
x <- as.matrix(mtcars[, -1]); y <- mtcars$mpg
d <- xgb.DMatrix(x, label = y)
xgb.train(list(objective = "reg:squarederror", eta = 0.05),
d, nrounds = 100, verbose = 0)
Exercise 5: max_depth
Difficulty: Intermediate.
Show solution
RInteractive R
xgb.train(list(objective = "reg:squarederror", max_depth = 3),
xgb.DMatrix(as.matrix(mtcars[,-1]), label = mtcars$mpg),
nrounds = 50, verbose = 0)
Exercise 6: Watchlist + early stopping
Difficulty: Advanced.
Show solution
RInteractive R
set.seed(1)
idx <- sample(seq_len(nrow(mtcars)), 22)
dtr <- xgb.DMatrix(as.matrix(mtcars[idx, -1]), label = mtcars$mpg[idx])
dte <- xgb.DMatrix(as.matrix(mtcars[-idx, -1]), label = mtcars$mpg[-idx])
fit <- xgb.train(list(objective = "reg:squarederror"),
dtr, nrounds = 200, watchlist = list(test = dte),
early_stopping_rounds = 10, verbose = 0)
fit$best_iteration
Exercise 7: Predict on new
Difficulty: Beginner.
Show solution
RInteractive R
x <- as.matrix(mtcars[, -1])
d <- xgb.DMatrix(x, label = mtcars$mpg)
fit <- xgb.train(list(objective = "reg:squarederror"), d, nrounds = 50, verbose = 0)
predict(fit, x[1:3, ])
Exercise 8: xgb.cv built-in CV
Difficulty: Advanced.
Show solution
RInteractive R
d <- xgb.DMatrix(as.matrix(mtcars[,-1]), label = mtcars$mpg)
xgb.cv(list(objective = "reg:squarederror"),
d, nrounds = 100, nfold = 5, verbose = 0)
Exercise 9: Feature importance
Difficulty: Intermediate.
Show solution
RInteractive R
d <- xgb.DMatrix(as.matrix(mtcars[,-1]), label = mtcars$mpg)
fit <- xgb.train(list(objective = "reg:squarederror"), d, nrounds = 100, verbose = 0)
xgb.importance(model = fit)
Exercise 10: Plot importance
Difficulty: Intermediate.
Show solution
RInteractive R
d <- xgb.DMatrix(as.matrix(mtcars[,-1]), label = mtcars$mpg)
fit <- xgb.train(list(objective = "reg:squarederror"), d, nrounds = 100, verbose = 0)
xgb.plot.importance(xgb.importance(model = fit))
Exercise 11: Regularization with lambda
Difficulty: Advanced.
Show solution
RInteractive R
d <- xgb.DMatrix(as.matrix(mtcars[,-1]), label = mtcars$mpg)
xgb.train(list(objective = "reg:squarederror", lambda = 1),
d, nrounds = 50, verbose = 0)
Exercise 12: subsample row sampling
Difficulty: Advanced.
Show solution
RInteractive R
d <- xgb.DMatrix(as.matrix(mtcars[,-1]), label = mtcars$mpg)
xgb.train(list(objective = "reg:squarederror", subsample = 0.7),
d, nrounds = 50, verbose = 0)
Exercise 13: colsample_bytree
Difficulty: Advanced.
Show solution
RInteractive R
d <- xgb.DMatrix(as.matrix(mtcars[,-1]), label = mtcars$mpg)
xgb.train(list(objective = "reg:squarederror", colsample_bytree = 0.7),
d, nrounds = 50, verbose = 0)
Exercise 14: caret tuning grid
Difficulty: Advanced.
Show solution
RInteractive R
set.seed(1)
train(mpg ~ ., data = mtcars, method = "xgbTree",
tuneGrid = expand.grid(nrounds = 50, max_depth = c(3, 5),
eta = c(0.05, 0.1), gamma = 0,
colsample_bytree = 0.7, min_child_weight = 1, subsample = 0.7),
trControl = trainControl(method = "cv", number = 5))
Exercise 15: Save model
Difficulty: Intermediate.
Show solution
RInteractive R
d <- xgb.DMatrix(as.matrix(mtcars[,-1]), label = mtcars$mpg)
fit <- xgb.train(list(objective = "reg:squarederror"), d, nrounds = 50, verbose = 0)
xgb.save(fit, "model.bin")
Exercise 16: Load model
Difficulty: Intermediate.
Show solution
RInteractive R
loaded <- xgb.load("model.bin")
Exercise 17: SHAP values
Difficulty: Advanced.
Show solution
RInteractive R
d <- xgb.DMatrix(as.matrix(mtcars[,-1]), label = mtcars$mpg)
fit <- xgb.train(list(objective = "reg:squarederror"), d, nrounds = 50, verbose = 0)
predict(fit, as.matrix(mtcars[,-1]), predcontrib = TRUE) |> head()
Exercise 18: Custom eval metric
Difficulty: Advanced.
Show solution
RInteractive R
mae <- function(preds, dtrain) {
labels <- getinfo(dtrain, "label")
list(metric = "mae", value = mean(abs(preds - labels)))
}
d <- xgb.DMatrix(as.matrix(mtcars[,-1]), label = mtcars$mpg)
xgb.train(list(objective = "reg:squarederror"), d, nrounds = 50,
feval = mae, watchlist = list(train = d), verbose = 0)
Exercise 19: Class imbalance scale_pos_weight
Difficulty: Advanced.
Show solution
RInteractive R
# pos = 100, neg = 900 -> scale = 9
xgb.train(list(objective = "binary:logistic", scale_pos_weight = 9),
xgb.DMatrix(matrix(rnorm(100), ncol = 10), label = sample(0:1, 10, replace = TRUE)),
nrounds = 10, verbose = 0)
Exercise 20: Compare to random forest
Difficulty: Advanced.
Show solution
RInteractive R
set.seed(1)
idx <- sample(seq_len(nrow(mtcars)), 22)
tr <- mtcars[idx, ]; te <- mtcars[-idx, ]
rf <- randomForest::randomForest(mpg ~ ., data = tr)
xgb_fit <- xgb.train(list(objective = "reg:squarederror"),
xgb.DMatrix(as.matrix(tr[,-1]), label = tr$mpg),
nrounds = 100, verbose = 0)
list(rf_rmse = sqrt(mean((te$mpg - predict(rf, te))^2)),
xgb_rmse = sqrt(mean((te$mpg - predict(xgb_fit, as.matrix(te[,-1])))^2)))
What to do next
- Random-Forest-Exercises (shipped), comparison.
- Machine-Learning-Exercises (shipped), broader practice.