MHNfs is a few-shot method trained on FS-Mol. The trained model is applied to the benchmark by using the training data as a support set without updating any model weigths.
| Test set | Target label | Metric | Score |
|---|---|---|---|
| test | CLS_RET | f1 | 0.2905982905982906 |
| test | CLS_RET | mcc | 0.10705453823777566 |
| test | CLS_RET | pr_auc | 0.41735935247372935 |
| test | CLS_RET | roc_auc | 0.758757435558493 |
| test | CLS_RET | accuracy | 0.2169811320754717 |
| test | CLS_RET | cohen_kappa | 0.022661630748722517 |
