mi-911 / data-collection-api

data-collection-api

data-collection-api's Issues

Remove legacy last batch query from queries.py

Don't double filter

data-collection-api/serve.py

Line 36 in 8f61607

samples = samples[~samples.uri.isin(get_seen_entities())]

Query contextual information for actors, directors and subjects

Create queries such that we can show for an actor what movies they are known from, etc.

Dockerise the API

Find a Dockerfile for a simple Python app and dockerize it, installing everything from requirements.txt. Map ports 80 and 443 to the host ports. Remember to create a volume for the collected data.

Annotate data collections with exploration strategy

When collecting data for a user, find a nice way of annotating the data with a label so we can find e.g. all ratings from users who answered questions on entities in one-hop Movie exploration, random exploration, etc.

Try using particle filtering during the interview

Revise formula for movie sampling

Comment from Matteo:
Would it be easier to have both numbers normalized and then multiply by alpha and beta so to have like a linear combination where you can decide how much each part affects?

Conduct experiments on the number of particles and performance, measure time vs. precision

sample_relevant_neighbours should consider companies and decades

Exception on /final endpoint

[2020-04-16 10:32:45,672] ERROR in app: Exception on /api/final [POST] Traceback (most recent call last): File "/usr/local/lib/python3.7/site-packages/flask/app.py", line 2447, in wsgi_app response = self.full_dispatch_request() File "/usr/local/lib/python3.7/site-packages/flask/app.py", line 1953, in full_dispatch_request return self.finalize_request(rv) File "/usr/local/lib/python3.7/site-packages/flask/app.py", line 1968, in finalize_request response = self.make_response(rv) File "/usr/local/lib/python3.7/site-packages/flask/app.py", line 2131, in make_response " {rv.__class__.__name__}.".format(rv=rv) TypeError: The view function did not return a valid response. The return type must be a string, dict, tuple, Response instance, or WSGI callable, but it was a int.

Impose hard requirement on minimum number of DEs

Currently, a user can theoretically go through the entire interview providing only RE recommendations.

Create the final prediction

Make a final guess towards the user's preferences once the user has liked/disliked >= 50 entities.
Should be simple, something like personalised pagerank with liked/disliked nodes as sources.

Save timestamps for requests

Should also add an initial request to know when the user first visits the game.

Need to diversify a bit more

I often see the same two movies (brother bear and ted 2) when I like comedy, we should probably diversify it more. In addition, we should diversify the final screen.

Add sequel and prequel relations

Ensure that MindReader can only ask about decade once

Not always showing 9 questions with the new sampling strategy