Unconstrained Logistic Regression

Model overview

Unconstrained logistic regression classifier using L1 (Lasso) regularization to predict taxonomy IDs from product tags. Unlike parameter-constrained models, this classifier uses ALL available tags and relies on L1 regularization to achieve sparsity by driving unimportant coefficients to zero.

5 CV folds

53.16% Mean accuracy

0.5779 Mean F1

0.18551523 Mean p-adic loss

1,552 Avg non-zero params

Total train samples: 13,460
Total test samples: 3,365
Number of tags (input features): 1,627
Avg non-zero parameters: 1,552 / 345,462 (99.6% sparse)
L1 regularization (C): 1.0000

Cross-validation results

Fold	Accuracy	F1	P-adic loss (mean)	Non-zero params	Details
0	53.72%	0.5882	0.17259247	1,573	View
1	51.91%	0.5598	0.15714500	1,605	View
2	53.98%	0.5851	0.21760809	1,536	View
3	50.90%	0.5545	0.21425585	1,464	View
4	55.29%	0.6018	0.16597474	1,583	View