Selecting Models

[rlagare]

System Information (please complete the following information):

• Model Builder
• Visual Studio Version (if applicable):

Describe the bug

• On which page of the tutorial did you run into an issue (URL): This is not a bug
• Clear description of the problem: Model builder automatically selects the best model for the loaded dataset. It also shows other models that it explored for that particular dataset. In some cases, the models have similar accuracies. 1. Is it possible to consume these other models instead of the one suggested automatically, and how do you do this?

2. Also, how do you inspect the parameters of the trained model? I can only see the training parameters (e.g., learning rate) of the selected model, but not the model parameters that are used to make predictions on testing data.

To Reproduce Steps to reproduce the behavior:

1. Go to '...'
2. Click on '....'
3. Scroll down to '....'
4. See error

Expected behavior

1. The ability to consume other models that are explored, and not just the model that was automatically suggested by Model builder. 2. The ability to inspect the parameters of the trained model-- this is useful for scientific reporting.

Screenshots If applicable, add screenshots to help explain your problem.

Additional context Add any other context about the problem here.

[torronen]

@rlagare The temp folder that is mentioned in the output window of Visual Studio has all the models as zip files. I have been succesful in copying them and running my own metrics on them with test data. Note that the name of the zip file and the number in the log file do not always exactly match. The params are lost as far as I know, i created an issue about it recently #1969 I agree, params would be useful for scientific reporting and also for practicioners to understand impact of parameters in their models.

Microsoft team may have some better info.

[torronen]

@rlagare LightGBM and FastTree .zip files include the hyperparams inside the zip file, so it might be possible to extract them throuhg .NET interfaces, or worst case, by reading directly from the Model.key file either somehow deserializing the binary file, or even reading as ASCII. Screenshots are from notepad

If we are lucky, we might be able to find the same from all model.zip files.

[rlagare]

Thanks for the advice @torronen. It seems like this aspect of ML.NET is not very user-friendly; I wonder if Microsoft team would be able to make parameter extraction/reading easier to do. @beccamc

[luisquintanilla]

Updated labels to enhancement / feature request inline with other #1591 work items.