Which is true, but doesn't automatically mean that most of the code has to be developed in that compiled language. There is a reason why Python is a popular language for machine learning: the computation is done in optimized native libraries, but the logic can be developed in a more interactive language without a relevant performance penalty.
This also applies to a large amount of other problems, splitting the computation between a slower "high level language" which is suitable for rapid development and low level libraries in a fast compiled language is a good solution.
And of course, you could have fast compiling languages which allow interactive development - just naming Common Lisp as one. You can recompile individual functions witout rebuilding the executable. This could be achived with many languages, but unfortunately it seems that no one bothered to build such systems.
This also applies to a large amount of other problems, splitting the computation between a slower "high level language" which is suitable for rapid development and low level libraries in a fast compiled language is a good solution.
And of course, you could have fast compiling languages which allow interactive development - just naming Common Lisp as one. You can recompile individual functions witout rebuilding the executable. This could be achived with many languages, but unfortunately it seems that no one bothered to build such systems.