Eraya · MSI Classifier
Gen5 Binary MANTIS
pure binary MSI-H vs MSS run using H-optimus-1 bags; MANTIS >= 0.65 is MSI-H, otherwise MSS
Live Gen5 Training Scores
Updates from history.csv and fold_metrics.json every 3 seconds. Metrics appear as each fold exports predictions.
Live Epoch History
Train loss, validation loss, and multiclass AUROC from the latest fold histories.
Fold Scores And Loss
Latest 20 active/completed folds, using best AUROC, best AUPRC, and best validation loss when available.
| Seed | Fold | Status | Epochs | Best AUROC | Best AUPRC | Best Loss | F1 Macro |
|---|
| Macro AUROC | - |
| Macro AUPRC | - |
| Accuracy | - |
| Balanced Accuracy | - |
| F1 Macro | - |
| F1 Weighted | - |
| Log Loss | - |
| Brier Multiclass | - |
| ECE Multiclass | - |
| MSI-H AUROC / AUPRC | - / - |
| MSI-L AUROC / AUPRC | - / - |
| MSS AUROC / AUPRC | - / - |
Pooled Confusion Matrix
Rows are true class, columns are predicted class, pooled from completed fold metrics.
Best Thresholds Vs Default 0.50
AUROC/AUPRC do not change with threshold. This shows best operating threshold, F1 gain, and ROC/PRC operating points vs default 0.50.
| Class | Metric | Best T | Best | Default | Delta | ROC best | ROC def | PR best | PR def |
|---|---|---|---|---|---|---|---|---|---|
| Waiting for threshold_sweep.csv | |||||||||
| Seed | Fold | Class | Metric | Best T | Best | Default | Delta | ROC best | ROC def | PR best | PR def | TP/TN/FP/FN | Def TP/TN/FP/FN |
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Waiting for completed folds to export threshold records. | |||||||||||||
Live TFRecord Progress
0/611 · 0%
Live H-optimus-1 Bag Progress
0/611 actual · 0% live
Live VM Storage
- GB free
Gen5 Analytics And Config
| Cohort slides | 611 |
| Classes | MSI-H / MSI-L / MSS |
| Extractor | H-optimus-1 |
| QC | otsu |
| Raw SVS cleanup | delete after matching TFRecords are ready |
| Bag / val bag | 1024 / 1024 |
| Batch / epochs | 1 / 40 |
| GPU folds / CPU workers | 1 / 6 |
| Macro AUROC | - |
| Macro AUPRC | - |
| Accuracy | - |
| Balanced Accuracy | - |
| F1 Macro | - |
| F1 Weighted | - |
| Log Loss | - |
| Brier Multiclass | - |
| ECE Multiclass | - |
Metrics will appear after the first completed fold exports predictions.parquet and fold_metrics.json.
Live Log
/home/pardeep/gen5/logs/train_binary_mantis065.log
No Gen5 log yet.
4BaseCare · Eraya · gen3-msi-run-001 · H-Optimus-0 + bistro.transformer · 265 training slides · 1360 unseen cohort