testing-auto-training-model

Fine-tuned multi-label codelist classifier based on svercoutere/robbert-2023-dutch-base-abb.

Labels (4)

• A1.1
• A1.2
• A1.7
• A2

Held-out evaluation

• macro_average_precision: 0.8964
• mean_brier_score: 0.0323
• macro_f1: 0.8511
• micro_f1: 0.8764
• weighted_f1: 0.8760
• macro_precision: 0.8414
• macro_recall: 0.8623
• micro_precision: 0.8623
• micro_recall: 0.8909
• subset_accuracy: 0.8773
• hamming_loss: 0.0351

Validation global threshold: 0.8000. Per-label thresholds are stored in threshold.json.