How to get model builder's efficiency in my code

[ClaraGard]

System Information (please complete the following information):

• OS & Version: e.g. Windows 10
• ML.NET Version: v2.0.1
• .NET Version: .NET 6.0

Describe the bug
I am trying to make a regression model using autoML, when I am using the model builder it goes through a lot of models quickly (about 300 in 100 seconds) and when I am using an autoML experiment it tries about 10 models in 100 seconds, resulting in a much less efficient model and I would like to use autoML experiements to get several models automaticaly

To Reproduce
Steps to reproduce the behavior:

1. Load data.csv with Model2.cs
2. use the pipeline I made
3. It is slow and goes through 10 models, R² = 0.02
4. Load the data with model builder
5. create a regression model and train it
6. it is fast and goes through 300 models R²=0.06

Additional context
FirstMLProject.zip

[ClaraGard]

oops I forgot the mbconfig file sorry
FirstMLProject.zip

[LittleLittleCloud]

@ClaraGard You probably be using an old version of model builder, where subsampling training dataset is enabled. That's why model builder seems to be training much faster than AutoML.

However, subsampling is a double-edge sword. It makes training faster in the cost of potentially worse metric because it uses only partial of training dataset. This can be observed on your dataset as well, where the best rsquare (closer to 0 the better) from model builder is 0.06 and best metric from AutoML is 0.02.

The subsampling feature is also available in AutoML since #6714 via enabling SubSampling flag. We haven't release the package with that feature yet but you should be able to give it a try on our nightly build

Let me know if you have any other questions