XGBoost 功能重要性显示重量而不是增益？

Question

I'm doing an XGBoost for a linear regression problem and the model works fine but is not printing out the feature importance (gain).

The results look like this:

But was expecting something like this:

Does anyone know why this is happening and how to fix it?

Answer 1

The gain, cover, and frequency metrics are only for the gbtree booster. The gblinear booster only gives weight. Perhaps you would prefer to fit the gbtree booster? That's the default option, and I think, what is most often used.

library(xgboost)

m1 <- xgboost(
  data  = as.matrix(mtcars[, -1]),
  label = mtcars[, 1],
  nrounds = 50,
  verbose = 0
)

xgb.importance(model = m1)
#>     Feature         Gain       Cover   Frequency
#>  1:     cyl 4.387140e-01 0.020018810 0.039711191
#>  2:      wt 3.033430e-01 0.112723364 0.133574007
#>  3:    disp 1.870484e-01 0.391643155 0.332129964
#>  4:      hp 4.358684e-02 0.112051592 0.126353791
#>  5:    qsec 1.397432e-02 0.192798603 0.211191336
#>  6:    drat 1.082512e-02 0.090420529 0.106498195
#>  7:    carb 2.487836e-03 0.035469569 0.019855596
#>  8:    gear 1.177536e-05 0.015047696 0.009025271
#>  9:      vs 7.260741e-06 0.025392987 0.016245487
#> 10:      am 1.413125e-06 0.004433696 0.005415162

m2 <- xgboost(
  data  = as.matrix(mtcars[, -1]),
  label = mtcars[, 1],
  nrounds = 50,
  verbose = 0,
  booster = "gblinear"
)

xgb.importance(model = m2)
#>     Feature       Weight
#>  1:      am  3.411794186
#>  2:      vs  1.866894841
#>  3:    gear  1.492013931
#>  4:    carb -1.169109583
#>  5:    drat  0.893951356
#>  6:      wt -0.591026664
#>  7:     cyl  0.216187149
#>  8:    qsec  0.150260374
#>  9:      hp -0.014555559
#> 10:    disp -0.004487043

^{Created on 2022-08-17 by the reprex package (v2.0.1)}