Make word_align_corpus train lazily

[ddaspit]

• Properly cleans up the created model

---

This change is 

[Enkidu93]

This isn't used anywhere in silnlp, I don't believe. Is there anywhere else we should check / anyone else we should inform about this change in behavior?

@Enkidu93 reviewed 3 files and all commit messages, and made 1 comment.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on pmachapman).

[ddaspit]

Transductive support was just added. I don't think anyone is using this function yet.

@ddaspit made 1 comment.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on pmachapman).

[Enkidu93]

I'm probably missing something: Now when you call word_align_corpus(), the model will potentially be trained every time you call get_rows() on the returned corpus rather than once, right? Won't that affect callers who are using word_align_corpus() already?

@Enkidu93 made 1 comment.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on pmachapman).

[ddaspit]

Yes, you are correct. It is a change in behavior, but it wasn't really "correct" before, because there was no way to clean up the model. I don't think any callers would be calling it multiple times anyway.

@ddaspit made 1 comment.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on pmachapman).

[Enkidu93]

OK, right. I confirmed it isn't used like this in silnlp - just wanted to make sure there wasn't anything else we should check. But yes, a caller really shouldn't be calling get_rows() repeatedly anyways.

@Enkidu93 made 1 comment.
Reviewable status: complete! all files reviewed, all discussions resolved (waiting on pmachapman).