Why do we use loc for pandas dataframes? it seems the following code with or without using loc both compiles and runs at a similar speed: