Closed pminervini closed 6 years ago
I think train just supports train right? We could make it a cli but it currently isn't as far as I understand it.
Anyway, this should not happen before v0.1
jack/train_reader.py
is how we interact with Jack from the command line - I think that to keep it separated from the rest of the code base helps wrt. clarity
So it might live in something like cli/train_reader.py
or bin/train_reader.py
. That would be ok with me
I mean, I just want to have a script that says train. I think we don't have too many cli use cases so we can as well make them explicit.
Ok! :) I'll handle this right after Issue #274
Generally, I like the idea of separating this. However, I think right now the train_reader
script has too much re-usable training logic inside to be just an "interface". What I would like to see is a train_reader
function in the framework that implements most of the training logic (early stopping etc), and which I call through an API. Something like
def train_reader(reader, train_data, test_data, dev_data, configuration:dict)
(with more type specifications). The CLI interface should then be a very light wrapper around this function (which itself still lives in the framework and can be accessed programmatically).
train_reader.py
is a CLI interface for Jack's functionalities - I think it can make sense to keep it separate from the framework.Do you think so as well?