Repository navigation
ImageClassificationTrainer.EarlyStoppingMetric.Loss must have _bestMetricValue equal to Single.MaxValue in case of Loss metric #5938
Description
System Information:
- Windows 10
- ML.NET v1.6.0
- .NET 5.0
Bug
ImageClassificationTrainer.EarlyStopping doesn't work with ImageClassificationTrainer.EarlyStoppingMetric.Loss
It stops training immediately after epoch is reaching patience.
I expect reason of that bug in line 250 in ImageClassificationTrainer.cs
_bestMetricValue = 0.0f; must be equal to Single.MaxValue in case of Loss metric and 0.0f for Accuracy metric.
Activity