Always predicting the most common taxonomy (baseline for comparison)
The dummy classifier is a baseline model that always predicts the most common taxonomy class from the training data. This provides a simple benchmark to compare against more sophisticated models. Any model that performs worse than this baseline is essentially useless.
For each fold, the model identifies the most frequent taxonomy in the training set and predicts it for every test example, regardless of tags or other features.
| Fold | Predictions | Accuracy | Mean p-adic loss |
|---|---|---|---|
| Fold 0 | 1,220 | 4.59% | 0.0005 |
| Fold 1 | 1,233 | 5.76% | 0.0005 |
| Fold 2 | 1,244 | 5.14% | 0.0005 |
| Fold 3 | 1,243 | 4.02% | 0.0005 |
| Fold 4 | 1,241 | 5.08% | 0.0005 |