Custom objective for multi-output regression with multi_strategy=multi_output_tree

[sergun]

How does custom objective work for multi-output regression with multi_strategy=multi_output_tree?

Custom loss function is calculated on each output separately and then summed up?

Or is it called once, so it is possible e.g. to define loss function as weighted sum of losses applied to each output?

[trivialfis]

Hi, the multi_output_tree is working in progress, we don't recommend using a custom objective with it since we haven't done extensive tests yet.

[trivialfis]

Closing in favor of #9043

[RawabiAlwanin]

how gradients are combined in multi-target models using built in objectives like reg:squarederror? are they summed up? and
in which file can find this gradients combination?