NullReferenceException when training data

[ivan-pelly]

System Information (please complete the following information):

• OS & Version: Windows 10
• ML.NET Version: Microsoft.ML 2.0.0, Microsoft.ML.AutoML 0.20.0
• .NET Version: NET 6.0

Describe the bug
Training following the example at https://github.com/jwood803/MLNetExamples/tree/master/MLNetExamples/AutoML throws a System.AggregateException

To Reproduce
Steps to reproduce the behavior:

1. Load the attached project
2. Edit line 6 in Program.cs to point to the path of the provided data file on your system
3. Run the code
4. See error

Expected behavior
The code executes to completion.

NDML.zip

[ivan-pelly]

Additional error info attached.

[michaelgsharp]

@LittleLittleCloud looks like this is using AutoML. Any ideas?

@luisquintanilla are you familiar with that sample by Jon?

[ivan-pelly]

Thanks for giving this a look. Happy to provide any additional info as needed.

[luisquintanilla]

Hi @ivan-pelly,

Thanks for your question. This looks like it's using the old AutoML API. We recommend you use the new AutoML API. Check out this guide for more information. You can also find accompanying samples.

Closing this issue for now. If you run into issues with the new AutoML API, please feel free to open a new issue.

[ivan-pelly]

Hi @luisquintanilla - thanks for the info. My project is using Microsoft.ML.AutoML version 0.20.0, which it reports as the latest stable version. Is there a more recent version I'm not seeing, or are you referring to a different package? Any guidance appreciated. Thank you!

[ivan-pelly]

[luisquintanilla]

@ivan-pelly 0.20.0 is the latest version. However, the old APIs are still part of that package. For example, your call to CreateRegressionExperiment and Progress<T> are the old APIs. I would recommend looking at the how-to guide and samples which show the new API.

[ivan-pelly]

Got it - thanks @luisquintanilla ! So the old and new api's are in the same package? Is there a way to distinguish which is which, or would just need to follow the docs and hope I don't call something old? Cheers.

[luisquintanilla]

@ivan-pelly

So the old and new api's are in the same package

Correct. If you were using the APIs prior to 0.20.0, your solutions should still work. However, we recommend you use the versions in 0.20.0 and later.

Is there a way to distinguish which is which, or would just need to follow the docs and hope I don't call something old?

Unfortunately not at this time and the best way is to consult the docs and samples.