On my machine, quantgen with gurobi appears to produce fits with notably worse in-sample quantile loss on the highest and lowest quantile levels in the attached data set, with a 34% increase in loss on one and a 43% increase in loss in the other, relative to quantreg, with the following common settings:
including an intercept
no regularization
no noncrossing constraints
no standardization
Additionally, quantreg was much faster to produce results in this configuration, which, if I recall correctly, was not the case in the vignette example. quantgen with glpk was running the slowest, so I did not include it in the loss comparison above.
On my machine, quantgen with gurobi appears to produce fits with notably worse in-sample quantile loss on the highest and lowest quantile levels in the attached data set, with a 34% increase in loss on one and a 43% increase in loss in the other, relative to quantreg, with the following common settings:
Additionally, quantreg was much faster to produce results in this configuration, which, if I recall correctly, was not the case in the vignette example. quantgen with glpk was running the slowest, so I did not include it in the loss comparison above.
debug-quantgen-data-deparsed.txt