Function train 

pub fn train(
    dirs: &Dirs,
    trusted: &BTreeMap<String, Vec<String>>,
    options: &Options,
    progress: &mut dyn FnMut(&Progress),
    cancel: &Cancel,
) -> Result<Summary, LearnError>

Trains on the corpus as it is (see the module’s steps). The table is replaced only when no worse; trained.toml says what happened.