Spam detector

#2More performance!

    • status set to closed

    Training performance improved by: - Doing mass insert-ignores & updateWhere instead of forM upsertBy - Grouping by lemma (previously: per occurence +1 upsert, now: only one update)

    Evaluation performance improved by: - using selectList instead of dozens of getBy calls - Grouping by lemma (previously: per occurence a getBy, now: only one query and multiplication afterwards)

    This should be sufficient for now.