Repository navigation
Data classification: Training failed with model builder error "Index was outside the bounds of the array". #2294
Description
System Information (please complete the following information):
Windows OS: Windows-11-Enterprise-21H2
ML.Net Model Builder 2022: 16.13.10.22513701 (Main Build)
Microsoft Visual Studio Enterprise: 2022(17.3.6)
.Net: 6.0
Describe the bug
- On which step of the process did you run into an issue:
Training failed with model builder error "Index was outside the bounds of the array" when choosing "SentimentText" as Label and using metric "Log-Loss Reduction".
Data Source
https://testpass.blob.core.windows.net/test-pass-data/wikipedia-detox-250-line-data.tsv
To Reproduce
Steps to reproduce the behavior:
- Select Create a new project from the Visual Studio 2022 start window.
- Choose the C# Console App (.NET Core6.0) project template.
- Add model builder by right click on the project.
- Select Data classification scenario.
- On Data page, select the data source and choose SentimentText column as label.
- On the Train page, choose metric "Log-Loss Reduction", then click "Start training" button.
- You will see that training failed with model builder error "Index was outside the bounds of the array".
Expected behavior
Training successful.
Screenshot
Additional context:
After the error occurs, .mbconfig file cannot be saved successfully with error "Value cannot be null".
Activity
zewditu commentedon Oct 20, 2022
This issue happens when The logloss- reduction is infinity.
@LittleLittleCloud how we are dealing with infinity values metrics?
luisquintanilla commentedon Dec 8, 2022
May be related to dotnet/machinelearning#6361
LittleLittleCloud commentedon Jan 11, 2023
Model Builder still uses mlnet 2.0 so the fix is not introduced yet
v-Hailishi commentedon Jan 12, 2023
The bug is still repro on the latest main build 16.14.2.2306101
LittleLittleCloud commentedon Jan 23, 2023
The fix should checked in now
v-Hailishi commentedon Jan 31, 2023
@LittleLittleCloud The training is still failed on the latest main build 16.14.3.2307701.
luisquintanilla commentedon Feb 16, 2023
This error was fixed. New issue created to track throwing a better error when no suitable model found. #2500