Create get new task function and expose in API

[rufuspollock]

Location: /{api}/app/{id}/newtask/[{user-id}]

user-id is optional (may not be logged in!)

[lucasmation]

I have some ideas on how tasks should be atributed, which probably should be incorporated into PyBossa structure. Basically there are two ways people can colaborate:

• sequentially: with one volunteer building on the work of a previous volunteer. This is how a "wiki" works.
• in parallel: with the same task beeing done in parallel by many volunteers and the results compared for consistency. In the case of the table transcription we need both. Given a page (i) we can define a sej (j) of sequetitial translations (t). So a Job is described by: JOBi,j,t. Professional and academic table transcriptions using humans always use two parallel transcriptions, which are compared for consistency. Lets put some indexes as this could help in coding. So in this case, for translating page i=1, we would have:

Sequence j=1: JOB i=1,j=1,t=1 > JOB i=1,j=1,t=2 > JOB i=1,j=1,t=3 > ... > JOB i=1,j=1,t=n1 (when a user declares the transcription to be over.

Sequence j=2: JOB i=1,j=2,t=1 > JOB i=1,j=2,t=2 > JOB i=1,j=2,t=3 > ... > JOB i=1,j=2,t=n2

we would then have to compare (authomatically or by experienced users or project supervisors) the final iterations of each sequence: JOB i=1,j=1,t=n1 and JOB i=1,j=2,t=n2

@teleyinex , @rgrp : is this flexible sequential and/or parallel way of atributing tasks already build into PyBossa?

[nigini]

@teleyinex @rgrp This is maybe a Flask noob question, but I've searched the web and could not solve it yet! Can you try to help me with this?

I'm trying to create a new route at the API, but I can't stop receiving an 'HTTP 404'! As the route is different from the default ones (e.g. /app/1/newtask/1), I started creating a new 'MethodView', with a GET method. Then registered it with a different algorithm then 'register_api'. Is this the idea? The code is bellow (consider it a early^2 version).

class AppTaskAPI(MethodView):

    @jsonpify
    def get(self, app_id, command, user_id=None):
        bossa_app = model.Session.query(model.App).filter(model.App.id == app_id).one()
        if( bossa_app and not bossa_app.hidden ):
            #ToDo: This is a temporary solution!
            import random
            tasks = bossa_app.tasks
            task = tasks[random.randint(0,len(tasks))]
            return json.dumps(task)
        else:
            abort(404)

def register_apptask_api(view, endpoint, url, app='<int:app_id>', cmd='newtask', user='<int:user_id>'):
    view_func = view.as_view(endpoint)
    blueprint.add_url_rule('%s/%s/%s/%s' % (url, app, cmd, user),
        view_func=view_func,
        methods=['GET']
        )
    blueprint.add_url_rule('%s/%s/%s' % (url, app, cmd),
        view_func=view_func,
        methods=['GET']
        )

register_apptask_api(AppTaskAPI, 'api_apptask', '/app')

[nigini]

I'm making some tests with the Flickr demo... It appears that the "newtask" algorithm is working on the premise that TaskRuns are already existent; what is not true. I'll examine this now!

[nigini]

Yep! The "join(TaskRun)" at line 307 of new_task at model.py will crash if there is no TaskRun in the system yet! Yet! I could not capture this with a unit test (I'm learning how to use it!). @rgrp Can this like be commented until I finish a new algorithm for TaskRun creation? For the Flickr App it has no problem, but it may be used at other site!

[rufuspollock]

nigini no I would not comment this out. I'm also surprised it crashes if there are no TaskRuns (it just should not return anything). The real fix is probably a left join.

Aside: I'd also suggest not writing new core code until you are happy with unit tests as we should wherever possibly only add new code with tests ;-)

BTW: this issue is closed and can I assume you are not working on it?

[nigini]

OK! Issue closed. Not working on this, but yes on the algorithm to choose tasks.

BTW: I'm actually able to use unit tests, but never used it in a production environment in Python. Already got the idea of how it is used at PyBossa.

[rufuspollock]

But this task included the algorithm to choose tasks :-)

Would you like to update on what you want from algorithm to choose tasks so we can see if more work is needed on this issue?

At the moment the algorithm does crude randomization ...