internal feature importance estimators

[eilyashko]

am i wrong or one of the big limitations of using internal feature importance estimators is that they assume no NaN in the data (including exogenous variables?)

model_feature_selection = DecisionTreeRegressor(random_state=42)

plot_feature_relevance(
temp_ts,
relevance_table=ModelRelevanceTable(),
top_k=30,
relevance_params={"model": model_feature_selection},
)

[alex-hse-repository]

What do you mean under the "limitations"? This method works with NaNs in target/regressor columns, it simply drop them before fitting the model, the only thing is you get the warning about NaNs in your dataset