Author: Ludvig R. Olsen ( r-pkgs@ludvigolsen.dk )
The ultimate goal of training machine learning models is to generalize to new, unseen data. This package contains tools for measuring model performance across multiple datasets via cross-dataset-validation (aka. leave-one-dataset-out).
Under development!
- Not generalized enough for general usage (ironic, I know)
- Relies on an old version of scikit-learn, needs updating
- Linear regression is not currently working
- Help strings are likely not up-to-date
| Function | Description |
|---|---|
nested_cross_validate() |
Run (repeated) nested cross-validation. |
train_full_model() |
Train model on all data and save to disk. |
evaluate_univariate_models() |
Evaluate prediction potential of every predictor separately. |
PipelineDesigner |
Design a scikit-learn pipeline for use in cross-validation. |
ROCCurve, ROCCurves |
ROC curve containers with various utility methods. |
select_samples() |
Utility for selecting samples based on (collapsed) labels. |