xgboost what

Packages

Prepare data and fit the model.

data(mtcars)
xs <- xgb.DMatrix(data = mtcars %>% select(cyl:carb) %>% as.matrix,
            label = mtcars$mpg)
f <- xgb.train(params = list(booster = 'gblinear', objective = 'reg:linear')
               , data = xs
               , nrounds = 100)

R-squared, variance explained.

d <- tibble(pred = predict(f, newdata = xs)
            , obs = mtcars$mpg) %>% 
  mutate(resid = pred - obs,
         resid_sq = resid^2)
sstot <- sum((d$pred - mean(d$obs))^2)
ssresid <- sum(d$resid_sq)
sprintf("percent variance explained, R^2: %1.1f%%", 100 * (1 - ssresid / sstot))

## [1] "percent variance explained, R^2: 82.0%"

Density plot of residuals.

ggplot(d, aes(x = resid)) +
  geom_density(fill = 'grey50', color = 'white', alpha = 0.7) +
  theme_bw()

Observed and predicted.

ggplot(d, aes(x = obs, y = pred, size = resid)) +
  geom_point() +
  theme_bw()