Dealing with categorical features while training LightGBM model implies encoding them as integers and providing categorical_feature parameter with their indices or names.
LightGBM documentation says that:
Tree decision rule works best when categorical features are presented by consecutive integers starting from zero
all negative values will be treated as missing values
My question is: when zero_as_missing parameter is set, does it refer only to continuous features, and categorical will still be considered missing if their value is negative?
zero_as_missing 🔗︎, default = false, type = bool
set this to true to treat all zero as missing values (including the unshown values in LibSVM / sparse matrices)