JuliaAI / MLJTuning.jl

Hyperparameter optimization algorithms for use in the MLJ machine learning framework
MIT License
67 stars 12 forks source link

Tuned Model interface doesnt have class_weights #134

Closed vishalhedgevantage closed 2 years ago

vishalhedgevantage commented 3 years ago

The TunedModel interface supports parameter - weights but not class_weights. For classification problems with highly imbalanced classes, we need tuned model measures working with class weights.

OkonSamuel commented 3 years ago

Hello @vishalhedgevantage another approach to solving classification on highly imbalanced dataset is to predict based on a threshold. MLJ offers a wrapper model called BinaryThreshholdPredictor that does this. Just type

julia> using MLJ

julia> ?BinaryThreshholdPredictor`  

for more info.

ablaom commented 3 years ago

@vishalhedgevantage Thanks for flagging this good suggestion.