Early Stopping Results
Results appear here after calculation.
Metric History
Stopping Decision
Epoch-by-Epoch Analysis
Best and stop rows are outlined.| Epoch | Training | Validation | Decision value | Improvement | Best so far | Wait | Status |
|---|
Settings Comparison
Compare patience and minimum improvement combinations using the same data.
| Patience | Minimum improvement | Stop epoch | Best epoch | Best metric | Epochs saved |
|---|
Copyable Report
Formula Used
Minimize mode: improvement occurs when current value < best value − minimum improvement.
Maximize mode: improvement occurs when current value > best value + minimum improvement.
Stopping rule: stop when consecutive non-improving monitored epochs reach the patience value.
Percentage mode scales the threshold using the magnitude of the current best value. Warm-up and starting epochs delay monitoring. Restored weights point to the best recorded epoch.
How to Use
- Choose the metric and optimization mode.
- Set patience, minimum improvement, and optional baseline.
- Paste data, import a CSV file, or enter rows.
- Choose smoothing and comparison settings.
- Select Calculate to review the best and stopping epochs.
Example Data Format
| Epoch | Training metric | Validation metric |
|---|---|---|
| 1 | 0.9200 | 0.9800 |
| 2 | 0.7700 | 0.8200 |
| 3 | 0.6500 | 0.7000 |
| 4 | 0.5600 | 0.6100 |
Frequently Asked Questions
What does patience mean?
Patience counts monitored epochs without sufficient improvement. Training stops when that count reaches the selected limit.
What is minimum improvement?
It is the smallest change accepted as meaningful. Smaller changes increase the waiting counter.
Should loss use minimize mode?
Yes, most loss and error metrics improve by decreasing. Automatic mode detects common loss names.
Should accuracy use maximize mode?
Yes, accuracy, precision, recall, F1, and AUC generally improve by increasing.
Why restore the best weights?
The final observed epoch may perform worse. Restoring weights returns the model state from the best monitored epoch.
When is smoothing useful?
Smoothing helps noisy validation metrics. It may prevent stopping because of one temporary fluctuation.
How is overfitting detected here?
The calculator checks whether training improves while validation worsens across consecutive epochs. This is an indicator, not a diagnosis.
What happens when no stop occurs?
The calculator reports no trigger within supplied data. More epochs may still activate the rule later.
Can I compare several settings?
Yes, enter comma-separated patience and minimum improvement values. The comparison table evaluates every combination.