Model attributes¶

Input models¶

As part of the MTurk API, the following attributes are expected to exist on input models to tasks:

MTURK_PREFETCH¶

Tuple of strings. When preparing to show a task, MTURK_PREFETCH declares which foreign key relationships will be used, thereby using 1 database query instead of many.

For example, if you have a model PhotoPair that represents a pair of photos (photo1 and photo2) that you are going to show to a user, you can fetch both photos at the same time by specifying:

class PhotoPair(...):

    photo1 = ...
    photo2 = ...
    ...

    MTURK_PREFETCH = ('photo1', 'photo2')

Output models¶

The following attributes are expected to exist on output models for tasks. See server/intrinsic/models.py for example implementations.

mturk_submit(user, hit_contents, results, time_ms, time_active_ms, version, mturk_assignment=None, **kwargs)¶

This function is called when a MTurk HIT is submitted, and is in charge of storing the user’s response in the database. This function is free to perform slow operations, since it will be performed asynchronously by a celery worker.

Parameters:

user – mturk worker hit_contents – list of contents that the user was presented results – a dictionary, mapping the id of each content to the user’s response. This is returned by the javascript interface in the MTurk task. time_ms – total time taken by user, as a dictionary time_active_ms – only the time that the user was in the window, as a dictionary version – version string send by javascript interface mturk_assignment – mturk.models.MtAssignment instances

Returns:

a dictionary mapping contents to a list of objects created from that content. The dictionary key is of the form (ContentType id, object id), called a “content tuple”, i.e. { common.utils.get_content_tuple(content): [new_obj] } for content in hit_contents.

mturk_grade_test(test_content_wrappers, test_contents, results)¶

Process the results of a secret test (i.e. a subset of the objects have known answers). This only needs to be implemented if the experiment has test contents.

Parameters:

test_content_wrappers – a list (or QuerySet) of mturk.models.ExperimentTestContent instances. You should not need to use this list, since test_contents contains the actual test objects. test_contents – a list of the test objects shown to users. Entry i in the list corresponds to entry i of test_content_wrappers. results – a dictionary, mapping the id of each content to the user’s response. This is returned by the javascript interface in the MTurk task.

Returns:

(responses, responses_correct) where the \(i^\text{th}\) entry in responses is the extracted response from the \(i^\text{th}\) entry of test_contents, and the \(i^\text{th}\) entry of responses_correct is a bool indicating whether the \(i^\text{th}\) item in test_contents was correctly answered.