Closed pminervini closed 6 years ago
I think load_dir
is not a good idea here, because this is really just the path to a pretrained model. Maybe initialization_dir? Ultimately, the save_dir is also the load_dir of that specific reader.
load_dir
sounds good to me, for the path to a pretrained model
When you look at the configuration after training, I think this might be confusing, because of the duality of save_dir which is used to load a trained reader.
we should also change every falg with _folder
to _dir
.
Why is it confusing ?
Either _folder
or _dir
are good to me
Actually, they are really just training FLAGs anyway, so you are right. It might not make a difference after all. Both are never used after training.
I was struggling to understand:
Turns out it's
reader_dir
andpretrained_model
- thanks @dirkweissenborn !Can we change it to
save_dir
andload_dir
, or something along this line - for consistency and clarity ?Let's just agree on the naming, then I'll handle it.