Suppress noisy logging when `verbosity=0`

[ablaom]

Here's an example of annoying output when evaluating an Imbalance model in a pipeline:

julia> evaluate!(mach, resampling=cv, measure=accuracy, verbosity=0)
julia> evaluate!(mach, resampling=cv, measure=accuracy, verbosity=0)
[ Info: After filtering, the mapping from each class to number of borderline points is ("colosa" => 21, "virginica" => 22).
[ Info: After filtering, the mapping from each class to number of borderline points is ("colosa" => 21, "virginica" => 22).
[ Info: After filtering, the mapping from each class to number of borderline points is ("colosa" => 17, "virginica" => 23).
[ Info: After filtering, the mapping from each class to number of borderline points is ("colosa" => 17, "virginica" => 23).
[ Info: After filtering, the mapping from each class to number of borderline points is ("colosa" => 20, "virginica" => 25).
[ Info: After filtering, the mapping from each class to number of borderline points is ("colosa" => 20, "virginica" => 25).
[ Info: After filtering, the mapping from each class to number of borderline points is ("colosa" => 11, "virginica" => 24).
[ Info: After filtering, the mapping from each class to number of borderline points is ("colosa" => 11, "virginica" => 24).
[ Info: After filtering, the mapping from each class to number of borderline points is ("colosa" => 19, "virginica" => 27).
[ Info: After filtering, the mapping from each class to number of borderline points is ("colosa" => 19, "virginica" => 27).
[ Info: After filtering, the mapping from each class to number of borderline points is ("colosa" => 16, "virginica" => 27).
[ Info: After filtering, the mapping from each class to number of borderline points is ("colosa" => 16, "virginica" => 27).

[EssamWisam]

Do you obviously see the root cause for evaluate not to pass verbosity internally? https://github.com/JuliaAI/Imbalance.jl/blob/dev/src/oversampling_methods/borderline_smote1/interface_mlj.jl

Because in the code for the main method we have:

verbosity > 0 && @info INFO_BORDERLINE_PTS(y1_stats)

[ablaom]

evaluate passes verbosity to the fit call. Can you please point me to:

(i) the MMI.fit call
(ii) the line of code you quote above; I don't see it in the file you referenced.

[EssamWisam]

Interesting:
(i) there is no fit as this is a static transformer
(ii) there you go https://github.com/JuliaAI/Imbalance.jl/blob/dev/src/oversampling_methods/borderline_smote1/borderline_smote1.jl

[ablaom]

Ah. Then that explains it. Then there is no interface point for suppressing logging from evaluate. The transform and predict operations are typically silent in MLJ. It seems, realising this you provided verbosity as a hyperparameter (which is pretty non-standard, but fine). Can you change the default to silent?

[ablaom]

By the way, there is a way to make byproducts of calling transform available in the machine report associated with a static transformer (instead of just logging it to stdout). This is the more MLJ-idiomatic way of exposing this information. Perhaps we can open an issue as a possible feature enhancement.