not a lot of code is public domain and thus not a lot of training data is available