Add custom JSON adapter

[Martmists-GH]

The purpose of this adapter is to make use of different JSON parsing libraries, such as ujson or python-rapidjson.

Example usage:

chatbot = chatterbot.ChatBot(
    "Example bot",
    trainer='chatterbot.trainers.ChatterBotCorpusTrainer',
    storage_adapter={
        "import_path": "chatterbot.storage.CustomJsonFileStorageAdapter",
        "dump_func":"ujson.dumps",
        "load_func":"ujson.loads"
    }
)
chatbot.train("chatterbot.corpus.english")

As you can see, all that has to be done is to make storage_adapter a dict with an import_path, dump_func and load_func.
both dump and load functions default back to the stdlib json module. Note that dumps and loads WILL NOT WORK.

This has successfully been tested, however I do not know if I messed anything up by accident which could make it function less like it should.

Fixes #592

[Martmists-GH]

@gunthercox feel free to merge whenever

[gunthercox]

The JSON file database code is mostly contained in an external Python module (https://github.com/gunthercox/jsondb). I believe that some changes will need to happen there in order to make this work more elegantly. Ideally, the json dumps and loads methods could be set on the Database class from that library.

[Martmists-GH]

Noted. The reason I redid most of the I/O handling is because the current JSON adapter uses File I/O about 4 times for every filter call, causing it to slow down a lot (Not sure how much that matters on SSDs but I don't have any of those to test with). Once for every update call looked too much as well, so I set it up to save for every get_response. I will see if changing jsondb's source changes any of this, but I'm fairly sure it's the constant I/O that slows it down. At the moment even the default json.load and json.dump with the custom adapter are performing faster than jsondb, so I'll see what I can do there.

[Martmists-GH]

I made a PR to jsondb here. Using my PR as a test, I found that, despite it being a little faster due to it using ujson, it is still relatively slow. My guess is that the amount of File I/O calls is too high, which is why I propose to either make the amount of jsondb calls lower, or implement the custom adapter which makes less I/O calls.

[Martmists-GH]

Above two commits were made to support libraries such as rapidjson, which only supports dumps and loads calls.

[vkosuri]

@Martmists before pushing your commits try to runs tests your locally.

[Martmists-GH]

@vkosuri if you could explain to me how I will, but as said before I have no clue how to test

The weird thing with this one is that is somehow fails on py2 django 1.8 and 1.9, which I'm actually unsure of why this errors.

[vkosuri]

@Martmists I think this will help you http://chatterbot.readthedocs.io/en/stable/testing.html

[Martmists-GH]

I'll probably rewrite this storage adapter to use any dumping/loading method as to enable it to use for example the erlpack module.

[gunthercox]

Hi @Martmists, I apologize about the amount of time this pull request has remained open for. I'm going to close it off because I'm moving the project away from jsondb in favor of supporting popular, existing types of databases.

The JsonDatabaseAdapter in ChatterBot will be supported until the 0.7 version release but I'm not looking to add additional JSON type adapters anymore. I appreciate the work you put into this pull request and I'm sorry that I could not be merged into the project. If you still want to make it available for others to use you can always set it up as it's own Python package on PyPi.

[Martmists-GH]

I see, thanks for showing interest though!
I'm going to continue using this adapter myself, and I'll leave my fork of the repo up and up to date for people who want to use it as well.