Because we introduce AGLM usually without interaction columns, I decided to turn off use_interaction_columns as default.
Here I show RMSEs and elapsed times when using interaction columns and not using, and it can be observed that interaction columns reduce accuracy but increase computation time.
We choose simpler models as default from now.
with interaction columns:
system.time(source("~/aglm/examples/cross-validation.R"))
lambda.min: 0.06840858
RMSE: 4.721321
alpha.min: 1
lambda.min: 0.07507833
RMSE: 4.691021
user system elapsed
20.367 0.062 20.437
without interaction columns:
system.time(source("~/aglm/examples/cross-validation.R"))
lambda.min: 0.1484522
RMSE: 4.834732
alpha.min: 0.008
lambda.min: 8.415161
RMSE: 4.860114
user system elapsed
15.819 0.047 15.866
Because we introduce AGLM usually without interaction columns, I decided to turn off use_interaction_columns as default.
Here I show RMSEs and elapsed times when using interaction columns and not using, and it can be observed that interaction columns reduce accuracy but increase computation time.
We choose simpler models as default from now.
with interaction columns:
without interaction columns: