LinearTree does not fit well about linear-tree HOT 3 CLOSED

Hi,
The splits are computed until there is "some utility" and at least 3 points are available. In other words, if an improvement (a decrease in loss of 1e-5) is detected by adding a split the model continues to grow. If not, the model remains as is.
In the left part of your plot, these conditions are not met.
I suggest you to check also your model summary.
If you support the project don't forget to leave a star ;-)

from linear-tree.

And it makes any sense that there is no improvement? Just by wathing the chart. My objective is to overfiti it in max_depth. So it learn all the data.

I appreciate any comment

:)

from linear-tree.

Maybe yes, if you simply want to overfit them a simple DecisionTreeRegressor from sklearn suits well

from linear-tree.