Using Redis for the data structure about classifier-reborn HOT 7 CLOSED

Redis backend implementation is done (except code review and documentation). How about MongoDB, SQLite, MySQL, and/or PostgreSQL backends (a single ORM style implementation would cover them all)? Because Redis, though persistent, is still memory based and would require a lot of memory for bigger datasets.

from classifier-reborn.

That sounds pretty compelling. I'd be happy to look at a pr.

from classifier-reborn.

I will try to give it a shot. However, looking at the code in bayes.rb, I feel that it would be a lot easier and quicker to abstract out certain portions of train, untrain and classifications methods as separate private methods and conditionally alias them (during initialization) for their corresponding Redis implementations. I earlier proposed overriding [] and []= methods, but after looking at the actual code, I think that would be too complected and unnecessary.

from classifier-reborn.

from classifier-reborn.

@parkr: Yeah, one common idiom is to create backends. So you have a MemoryBackend, a RedisBackend, and so forth. They all share a common interface but are separate so you can worry about one at a time.

I tried using this backend approach you suggested. Currently, it only supports the Memory backend, but the refactoring I did establishes the foundation and uniform API for other backends. Please find the changes in PR #84. I would really appreciate any code reviews and comments. I would note that I touched some Ruby code after a long time.

from classifier-reborn.

Hum. I'd say we might want to wait until we have some users that want to use a database backend. We have a pretty small number of contributors, so I don't like to add a lot of things that have to be maintained unless they're going to get used.

from classifier-reborn.

Closing this as this is now implemented by #84.

from classifier-reborn.