You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
One special example to consider: option --probabilities does not change the way a model is trained, but it could be useful in training because it reports multi-class logistic loss (in addition to zero-one loss). The main purpose of --probabilities is for testing to convert the predicted values (-p) to probabilities.
There may be several models, some trained with --probabilities, some without and I may want to test all of them and predict probabilities. Currently, it is not possible (I need to provide the --probabilities option only for those models which were trained without it).
For me this is a low-priority issue - I have not encountered the problem above yet. But for newcomers, the error may be counter-intuitive (it says "more than once", but they see the option just once on the command line, they are not aware that it is saved in the model).
In #1082 (comment) @arielf wrote
I think this should be discussed in a separate issue (here).
I like @arielf 's suggestion
The text was updated successfully, but these errors were encountered: