[R-package] predict ignores predict_raw_score

[david-cortes]

After fitting a binary classification model, calling predict with predict_raw_score=TRUE or its aliases as explained in the docs will not output the raw score, only a probability as it would without this parameter:

library(lightgbm)
data("agaricus.train")
pool = lgb.Dataset(agaricus.train$data, label=agaricus.train$label)
params = list(
    objective_type="binary",
    force_row_wise=TRUE,
    seed=123
)
model = lgb.train(params, pool, verbose=0)
summary(predict(model, agaricus.train$data, params=list(predict_raw_score=TRUE)))

     Min.   1st Qu.    Median      Mean   3rd Qu.      Max. 
0.0000201 0.0000212 0.0000232 0.4821127 0.9999774 0.9999774 

summary(predict(model, agaricus.train$data))

     Min.   1st Qu.    Median      Mean   3rd Qu.      Max. 
0.0000201 0.0000212 0.0000232 0.4821127 0.9999774 0.9999774 

[jameslamb]

Thanks for raising this! This is an inconsistency we should resolve.

As documented in the R package docs or in ?lightgbm::predict.lgb.Booster, that function has a keyword argument rawscore which you can set to TRUE to get raw scores.

However, we should consider raising warnings on inconsistencies between those keyword arguments and the equivalent core-LightGBM parameters, and clearly document the expectation for which value "wins" when both are supplied.

[github-actions]

This issue has been automatically locked since there has not been any recent activity since it was closed. To start a new related discussion, open a new issue at https://github.com/microsoft/LightGBM/issues including a reference to this.