As I understand label encoding is used for nominal data (not ordinal) in order to map them to numericals (e.g.: labels [apple, banana, carrot] are mapped to [0, 1, 2], but it does not matter which label is mapped to which numerical). Ordinal encoding does the same mapping but in a predefined order (e.g.: [small, medium, large] is mapped to [0, 1, 2].
My question is if I use label encoding on a feature in a dataset and ordinal encoding on another, will the label encoded feature be treated as ordinary aswell due to the distances the mapping initialized? Is there any rule
Some algorithm for example might see a closer proximity between apple (1) and banana (2) than beetween apple (1) and carrot (3), because the difference is smaller. The proximity is information which is only added due to the encoding, but does not actually exist.