Missing mappings for $obs_loss

[jemus42]

I noticed that some measures like bbrier and logloss don't have an associated obs_loss

library(mlr3)

measure = msr("classif.bbrier")
measure$obs_loss
#> NULL

measure = msr("classif.logloss")
measure$obs_loss
#> NULL

While spelunking the measure-related internals I ended up here:

# Ok as expected
mlr3measures:::measures$mse$obs_loss
#> [1] "se"

# Should be "se" as well
mlr3measures:::measures$bbrier$obs_loss
#> [1] NA

# NA but should not be?
mlr3measures:::measures$logloss$obs_loss
#> [1] NA

For logloss I assumed something like Metrics:ll():

#> Metrics::ll
function (actual, predicted) 
{
    score <- -(actual * log(predicted) + (1 - actual) * log(1 - 
        predicted))
    score[actual == predicted] <- 0
    score[is.nan(score)] <- Inf
    return(score)
}