Level-wise Logistic Regression

Hierarchy-aware top-down classifier that always emits a valid taxonomy path

← Back to index

Model overview

This baseline trains one logistic decision at each internal taxonomy node and traverses the training-fold taxonomy tree top-down at inference time. It is hierarchy-aware by construction and always returns a valid taxonomy leaf from the training tree.

5CV folds
57.69%Mean accuracy
0.5638Mean F1
0.10077504Mean p-adic loss
83.01%Prefix-2 accuracy
228.20Mean scoring ops

Cross-validation results

FoldAccuracyF1P-adic loss (mean)Prefix-2 accuracyMean scoring ops
058.04%0.56620.1120172183.20%256.27
157.02%0.55480.0992801282.66%218.17
259.27%0.57930.0808859584.04%219.09
356.64%0.55580.1092577681.87%231.90
457.46%0.56270.1024341683.31%215.56