Table A5.
Hyperparameter values used during the hyperparameter tuning experiment with h: number of tested hyperparameter combinations, FC: feature class combinations with linear (l), quadratic (q), product (p) and hinge (h) feature classes, reg: regularization multiplier and iter: number of iterations
| h | FC | reg | iter |
|---|---|---|---|
| 75 | c(“lq”, “lh”, “lqp”, “lqh”, “lqph”) | seq(0.2, 3, 0.2) | 500 |
| 150 | c(“lq”, “lp”, “lh”, “lqp”, “lqh”, “lqph”) | seq(0.2, 5, 0.2) | 500 |
| 300 | c(“lq”, “lp”, “lh”, “lqp”, “lqh”, “lqph”) | seq(0.1, 5, 0.1) | 500 |
| 600 | c(“lq”, “lp”, “lh”, “lqp”, “lqh”, “lqph”) | seq(0.1, 5, 0.1) | c (500, 700) |
| 1200 | c(“lq”, “lp”, “lh”, “lqp”, “lqh”, “lqph”) | seq(0.1, 5, 0.1) | seq (300, 900, 200) |
The values are provided using the R code to generate them. In the optimizeModel function, in order to have consistent results, we set the seed argument to 186,546 (a randomly generated number).