[English name suggestion] "FeatureImportanceCalculator", "[.]Estimator", "[.]Indicator" or what?

Specific Domains Machine Learning
, ,
1

For a model that returns the ranking of the variables based on some “variable importance metric” to be used as:

m = ModelName(a_ML_model,options)
fit!(m,x,y)
var_importance_ranking = predict(m)
mda_losses = info(m)["mda"]
sobol_vars = info(m)["sobol"]

and that is going to join other existing models on this list, what would you use as “ModelName” ?

  • FeatureImportanceEstimator
  • FeatureImportanceIndicator
  • FeatureImportanceCalculator
  • VariableImportanceEstimator
  • VariableImportanceIndicator
  • VariableImportanceCalculator
  • Other ? (please specify…)
0 voters

Merci :upside_down_face:

2

Not trying to sway the vote here but I think “Feature Importance” is a fairly well established name for this, and I’d use “Estimator” over “Calculator” as calculator implies an operation with a known correct result - 1+1=2 - while here we are estimating a (meta) parameter in a statistical model that doesn’t really have a “true” value.

2 Likes
3

My “Other” vote would be FeatureRanking (or maybe FeatureRanker)

2 Likes
4

“Indicator” sounds weird to my ears but all the other options sound reasonable. I think the best “active noun” would be FeatureRanker. If you are going to break the active noun convention, I’d just go with FeatureImportance.

I’m not familiar with your API, but a pretty common application of the model will be to select only the top n features. So perhaps you think of this as a transformer, regarding the actual importances as “byproducts” of training you can inspect if you want to. In that case, something like FeatureImportanceSelector might work??

1 Like