Consilium

Consilium is a single page application that implements a movie recommender system by running machine learning algorithm(s) to make predictions on movie ratings.

[Consilium Live][live] [live]:https://consilium.herokuapp.com/

Instruction for Collaborators

Besides the standard,

$ bundle install
$ npm intall

Set up Redis in memory store by first installing Redis on your Mac or Linux OS

Mac OS X
$ brew install redis

Linux
$ sudo apt-get install redis-server

Setting up Redis on Rails (localhost)

In config/initializers/redis.rb, you will see how Redis is instantiated. The Redis instance will be stored in a global variable $redis and you may call this variable anywhere in the Rails app. For local host development, make sure to use

$redis = Redis.new(:host => 'localhost', :port => 6379)

and comment out the other line, because that one is for deployment to Heroku

Redis instance on Heroku

Redis URL can be found by running

$ heroku config | grep REDIS

Redis instance info can be found by running

$ heroku redis:info

Start running

First you need to start your Redis server by typing in

$ redis-server

Now you can start your Rails server

$ rails s

Don't forget webpack!

Database Design

New Design

Currently Consilium doesn't really have a database, it stores a giant hash map in Redis cache which contains all the necessary information for SVD matrix factorization algorithm. The values are computed off-line with some Python scripts. Now the next natural move is to actually keep things in a database. So here's the new schema for the upcoming back-end design

`HistoricalUser`

We will introduce a ActiveRecord model that is responsible for holding historical data from the giant data set from MovieLens. HistoricalUser has many movies through association

column name	key	data type	note
id	PK	Integer	Provided
preference		Arrays	Offline computed

HistoricalUser has many Ratings in our ActiveRecord associations

`Rating`

This is basically a join table with two foreign keys, pointing to HistoricalUser / User and Movie

column name	key	data type	note
id	PK	Integer	Auto-incremented
user_id	FK	Integer
movie_id	FK	Integer
value		Float

`Movie`

Movie has many ratings and many viewers through the Rating association

column name	key	data type	note
id	PK	Integer	Provided
title		String
year		Integer
imdb_rating		Float
imdb_id		String
feature		Arrays	Offline computed

`User`

This is a model for active user who login to our system through FB authentication. We will use Facebook ID as the primary key for this table. The detailed design will come later as we implement the actual FB authentication system on the front-end.

calvinfeng/consilium