Corpus class: loading model separately from input data

[naktinis]

I would like to be able to pre-load the trained model into the Corpus class once (for performance reasons mostly, as it takes ~2 seconds) and reuse it with multiple inputs.

If I understand correctly, Corpus class's __init__ method loads both the model and the input data at once. One easy option without breaking backwards compatibility would be to move input loading into a separate method and check if dirname is passed into the constructor before running it.

For example, here's one way to implement this:

Do you think this is something that makes sense? Or maybe I am missing some functionality that already makes such model reuse possible.

[kylebgorman]

I've never wanted to do that yet but it is absolutely sensible if you'd use it. In general Corpus.init is way too long and should probably be broken into a bunch of method calls.

[kylebgorman]

While I agree this would be a good enhancement, I don't have the time or energy to implement it myself. Feel free to reopen if you generate a PR ;)