Hyperparemetres optimizaton SGT Regressor

[XaviMartiCamarasa]

Hi all
I'm trying to use SGTRegressor, in an nowcasting experiment. For others like LinearRegression I found "optimal parameters", hopefully not overfitting(how could overcome this in river?)..

But for SGTRegressor with the same dataset is almost impossible a good prediction. At this point I do not know if it is because my "ignorance", because I do not find the optimal parametres or because simpy SGT Regressor does not perform well on Nowcasting..

Any idea will be welcome!

Many thanks, as usual, for you time

[smastelini]

@XaviMartiCamarasa, "there is no free lunch." So we expect no algorithm is good on everything.

However, without some reproducible code is hard to pinpoint the possible root causes or even check if the problem is a bad choice of hyperparameters or a learning bias problem.

[XaviMartiCamarasa]

Hi. I was just thinking in loud voice. No coding here.
How could avoid overfitting when doing a regression task?

Regards

[smastelini]

This is fairly philosophical, but when you use progressive validation, you can get up-to-date estimates of how your model performs on data it has never seen before. The performance measurements are reliable sources of information.

On a more technical side, regularization is one of the strategies to avoid overfitting. In the specific case of SGTs, they have a regularization parameter that can be used to control the model's complexity.

[XaviMartiCamarasa]

Many thanks!