We need to extend and clean up the functionality of the two new AgeML commands: model_feature_influence and age_model_vs_logistic_regression.
Ordering
Add option for ascending/descending ordering.
Optimization
Run optimization of feature addition one by one, based on a combination of both metrics AUC and MAE for example doing AUC * MAE
Try all permutations of features → evaluate errors and classification, find best metrics.
Analysis functions
Add function to compute how age prediction error and classification change with number of subjects used for training (start with 1 subject and increase logarithmically in 10 steps to maximum for example).
Inspect and expose model weights for each run (add to AgeML).
Flexibility
Ensure commands can run with different systems and covariates.
Consider classification with multiple systems simultaneously obtaining multiple deltas and optimising each.
Define which models should be available in age_model_vs_logistic_regression.
Code Style / Refactor
Order should return feature names directly (no need to call feature extractor separately).
feature_extractor(df, features) should return the correct subset (also allow specifying which features to use).
fit currently returns predicted labels – it shouldn’t.
Currently results are not saved in age_model_vs_logistic_regression. Fix this.
Encapsulate argument parsing into individual reusable functions (avoid repetition; each command should only need one line to add).
We need to extend and clean up the functionality of the two new AgeML commands: model_feature_influence and age_model_vs_logistic_regression.
Ordering
Add option for ascending/descending ordering.
Optimization
Run optimization of feature addition one by one, based on a combination of both metrics AUC and MAE for example doing AUC * MAE
Try all permutations of features → evaluate errors and classification, find best metrics.
Analysis functions
Add function to compute how age prediction error and classification change with number of subjects used for training (start with 1 subject and increase logarithmically in 10 steps to maximum for example).
Inspect and expose model weights for each run (add to AgeML).
Flexibility
Ensure commands can run with different systems and covariates.
Consider classification with multiple systems simultaneously obtaining multiple deltas and optimising each.
Define which models should be available in age_model_vs_logistic_regression.
Code Style / Refactor
Order should return feature names directly (no need to call feature extractor separately).
feature_extractor(df, features) should return the correct subset (also allow specifying which features to use).
fit currently returns predicted labels – it shouldn’t.
Currently results are not saved in age_model_vs_logistic_regression. Fix this.
Encapsulate argument parsing into individual reusable functions (avoid repetition; each command should only need one line to add).