[Model Builder] Canceling training should not discard current results

[Doggie52]

System information

• OS version/distro: Windows 10.0.18362 x64
• .NET Version (eg., dotnet --info): 2.1.507
• ML.NET Model Builder version: 16.0.1905.641
• Visual Studio 2017 version: 15.9.12

Issue

• What did you do? Ran Model Builder and started Training. After noticing that accuracy was not improving, I decided to cancel the training.
• What happened? Progress was canceled and I was unable to proceed to the Evaluate stage.
• What did you expect? I wanted to continue to the Evaluate stage to take the so far best performing model forward.

Possibly related to dotnet/machinelearning#844 although that issue I believe is for training one particular model, whereas this is for training multiple models.

Source code / logs

As you can see, no link to Evaluate was present beneath the results.

[rustd]

Thank you for reporting this. AutoML needs to support stopping a training session and give you best models it has so far.

[srsaggam]

We have this Item in the backlog. @CESARDELATORRE We should maybe consider this for next version

[JakeRadMSFT]

Related #233

[LittleLittleCloud]

#233

[vzhuqin]

This issue can't be repro on：

• Windows 10 Enterprise, Version 20H2
• ML.Net Model Builder (Preview): 16.5.21.2122301
• Microsoft Visual Studio Enterprise 2019: 16.9.4
• .Net: 5.0.202

Main branch:
https://privategallery.blob.core.windows.net/gallery/refs/heads/main/atom.xml

Steps：

1. Create new C# console app with .Net 5.0;
2. Add model builder by right click on the project;
3. Click "Stop training" after noticed the training results;
   
4. The message changed to "Cancelling">"x seconds remaining...">"Finalizing model...">"Training complete"