Calling specific in-built objective function

[Sukanta Basu (Sukantabasu)]

Hi,

FLAML provides several in-built "metrics". However, I could not find an easy way to change the "objectives". For example, lightgbm provides several objectives (https://lightgbm.readthedocs.io/en/latest/Parameters.html). For regression problems, one can use rmse, mae, Huber, etc. Is there a way I can specify one of these pre-existing objectives? I am not referring to any custom objectives.

Thanks.

[Chi Wang (sonichi)]

Depending on whether the parameter is set in constructor or .fit(), use either
https://microsoft.github.io/FLAML/docs/Use-Cases/Task-Oriented-AutoML#a-shortcut-to-override-the-search-space
or
https://microsoft.github.io/FLAML/docs/Use-Cases/Task-Oriented-AutoML#extra-fit-arguments

[Sukanta Basu (Sukantabasu)]

Thanks a lot!

[Sukanta Basu (Sukantabasu)]

If I am not mistaken, "objective" should be defined via "custom_hp" option. However, I cannot figure out what should be the "domain".

[Chi Wang (sonichi)]

The value you want to use if you want to fix it.

[Sukanta Basu (Sukantabasu)]

I want to use "mae". So, the domain is "mae"?

[Sukanta Basu (Sukantabasu)]

If I try:
custom_hp = {
"lgbm": {
"objective":
{"domain": "mae",
}
},
}

I am getting the error:
1894 "period"
1895 ), # NOTE: this is after kwargs is updated to fit_kwargs_by_estimator
-> 1896 custom_hp=custom_hp and custom_hp.get(estimator_name),
1897 max_iter=max_iter / len(estimator_list) if self._learner_selector == "roundrobin" else max_iter,
1898 budget=self._state.time_budget,
1899 )
1900 logger.info("List of ML learners in AutoML Run: {}".format(estimator_list))
1901 self.estimator_list = estimator_list

AttributeError: 'str' object has no attribute 'get'