public static void Main(string[] args)
{
// Used to read the text files.
StreamReader movieStream = new StreamReader(new FileStream("netflix/movie_titles.txt", FileMode.Open));
//StreamReader userTestingStream = new StreamReader(new FileStream("netflix/TestingRatings.txt", FileMode.Open));
//StreamReader userTrainingStream = new StreamReader(new FileStream("netflix/TrainingRatings.txt", FileMode.Open));
StreamReader userTestingStream = new StreamReader(new FileStream("netflix/reduced/TestingRatings-new.txt", FileMode.Open));
StreamReader userTrainingStream = new StreamReader(new FileStream("netflix/reduced/TrainingRatings-new.txt", FileMode.Open));
// Container objects
Movies movies = new Movies();
Users trainingUsers = new Users();
Users testingUsers = new Users();
Computation computation = new Computation();
// stores the weights (correlation) between two users.
Dictionary <Tuple <uint, uint>, float> weight;
Console.WriteLine("Populating movie database...");
while (!movieStream.EndOfStream)
{
string[] movieData = movieStream.ReadLine().Split(new char[] { ',' }, 3);
uint movieID = uint.Parse(movieData[0]);
uint movieYear = uint.Parse(movieData[1]);
string movieTitle = movieData[2];
movies.AddMovie(movieID, new MovieInfo(movieYear, movieTitle));
}
Console.WriteLine("Populating training database...");
while (!userTrainingStream.EndOfStream)
{
string[] trainingData = userTrainingStream.ReadLine().Split(',');
uint movieID = uint.Parse(trainingData[0]);
uint userID = uint.Parse(trainingData[1]);
float rating = float.Parse(trainingData[2]);
// Check to see if user already exists in the dictionary.
if (trainingUsers.UserExists(userID))
{
trainingUsers.GetUser(userID).AddRating(movieID, rating);
}
else
{
// Add user and rating
trainingUsers.AddUser(userID, new UserInfo());
trainingUsers.GetUser(userID).AddRating(movieID, rating);
}
}
Console.WriteLine("Populating testing database...");
while (!userTestingStream.EndOfStream)
{
string[] testingData = userTestingStream.ReadLine().Split(',');
uint movieID = uint.Parse(testingData[0]);
uint userID = uint.Parse(testingData[1]);
float rating = float.Parse(testingData[2]);
// Check to see if user already exists in the dictionary.
if (testingUsers.UserExists(userID))
{
testingUsers.GetUser(userID).AddRating(movieID, rating);
}
else
{
// Add user and rating
testingUsers.AddUser(userID, new UserInfo());
testingUsers.GetUser(userID).AddRating(movieID, rating);
}
}
Console.WriteLine("\nCommands:\nerror - computes the " +
"error on each instance of the test set on the training set." +
"\nexit - quits the program." +
"\ncommand - displays all available commands." +
"\nquery - allows you to perform queries as a user on a particular year.");
string input = "";
do
{
Console.Write(">>>");
input = Console.ReadLine();
switch (input)
{
case "command":
Console.WriteLine("\nCommands:\nerror - computes the " +
"error on each instance of the test set on the training set." +
"\nexit - quits the program." +
"\ncommand - displays all available commands." +
"\nquery - allows you to perform queries as a user on a particular year.");
break;
case "error":
weight = new Dictionary <Tuple <uint, uint>, float>();
Console.WriteLine("\nComputing the error with regards to the test set. This may take a few minutes.");
float differenceRating = 0.0f;
float differenceRatingSquared = 0.0f;
float numRatings = 0.0f;
foreach (KeyValuePair <uint, UserInfo> users in testingUsers.GetDataset())
{
weight = new Dictionary <Tuple <uint, uint>, float>();
// ensure the user exists in the training data as well.
// otherwise we cannot predict their rating.
if (trainingUsers.UserExists(users.Key))
{
// Computes the average rating of the active user.
float averageRatingOfActive = computation.WeightedSumOfUser(trainingUsers.GetUser(users.Key));
// compute weights.
foreach (KeyValuePair <uint, UserInfo> otherUsers in trainingUsers.GetDataset())
{
float correlation = computation.Correlation(movies, trainingUsers.GetUser(users.Key), otherUsers.Value);
// exclude any zero correlations, and some numbers are just too small.
if (correlation != 0.0f && !float.IsNaN(correlation))
{
weight.Add(new Tuple <uint, uint>(users.Key, otherUsers.Key), correlation);
}
}
float sumWeight = computation.SumWeight(weight);
foreach (KeyValuePair <uint, float> ratings in users.Value.GetDataset())
{
float predictedRating = 0.0f;
float cumulativeSum = 0.0f;
foreach (KeyValuePair <Tuple <uint, uint>, float> weights in weight)
{
if (trainingUsers.GetUser(weights.Key.Item2).RatingExists(ratings.Key))
{
cumulativeSum += weights.Value *
(trainingUsers.GetRatingForUser(weights.Key.Item2, ratings.Key) -
computation.WeightedSumOfUser(trainingUsers.GetUser(weights.Key.Item2)));
}
else
{
cumulativeSum += weights.Value;
}
}
// odd occurrence when there is no other weights.
if (sumWeight != 0.0f)
{
predictedRating = averageRatingOfActive + cumulativeSum / sumWeight;
}
else
{
predictedRating = averageRatingOfActive;
}
// Logging.
Console.WriteLine("==============================");
Console.WriteLine("UserID: " + users.Key);
Console.WriteLine("Predicted Rating of " + movies.GetMovie(ratings.Key).Title + " is " + predictedRating);
Console.WriteLine("Actual Rating of " + movies.GetMovie(ratings.Key).Title + " is " + users.Value.GetRating(ratings.Key));
// These simply return the difference in ratings and adds them up.
// we will divide by the number of ratings to get the actual mean and squared error.
differenceRating += computation.MeanAbsoluteError(predictedRating, users.Value.GetRating(ratings.Key));
differenceRatingSquared += computation.MeanAbsoluteError(predictedRating, users.Value.GetRating(ratings.Key)) *
computation.MeanAbsoluteError(predictedRating, users.Value.GetRating(ratings.Key));
numRatings++;
}
}
}
float meanAbsoluteError = differenceRating / numRatings;
float rootMeanSquaredError = (float)Math.Sqrt(differenceRatingSquared / numRatings);
Console.WriteLine("MAE : " + meanAbsoluteError);
Console.WriteLine("RMSE : " + rootMeanSquaredError);
break;
case "query":
weight = new Dictionary <Tuple <uint, uint>, float>();
Console.Write("Enter a userID: ");
uint userID = uint.Parse(Console.ReadLine());
Console.Write("Enter a particular year: ");
uint year = uint.Parse(Console.ReadLine());
List <Dictionary <float, MovieInfo> > predictedRatings = new List <Dictionary <float, MovieInfo> >();
if (trainingUsers.UserExists(userID))
{
Console.WriteLine("\nMovies you have watched:");
foreach (KeyValuePair <uint, float> movie in trainingUsers.GetUser(userID).GetDataset())
{
if (movies.MovieExists(movie.Key))
{
Console.WriteLine(movie.Key + " - " + movies.GetMovie(movie.Key).Title + ", " +
movies.GetMovie(movie.Key).Year + ": " + movie.Value);
}
}
// Gather the movies only produced in that particular year.
Movies particularMovies = new Movies();
foreach (KeyValuePair <uint, MovieInfo> movie in movies.GetDataset())
{
if (year == movie.Value.Year)
{
particularMovies.AddMovie(movie.Key, movie.Value);
}
}
// Computes the average rating of the active user.
float averageRatingOfActive = computation.WeightedSumOfUser(trainingUsers.GetUser(userID));
// Compute the weights of the users and add them to the dictionary.
Console.WriteLine("Computing weights for users...");
foreach (KeyValuePair <uint, UserInfo> user in trainingUsers.GetDataset())
{
if (user.Key != userID)
{
// Compute their correlation accross all movies in the database rather than the particular movies.
float correlation = computation.Correlation(movies, trainingUsers.GetUser(userID), user.Value);
if (correlation != 0.0f && !float.IsNaN(correlation))
{
weight.Add(new Tuple <uint, uint>(userID, user.Key), correlation);
}
}
}
float sumWeight = computation.SumWeight(weight);
// Now we perform analysis on each of these movies.
foreach (KeyValuePair <uint, MovieInfo> movie in particularMovies.GetDataset())
{
float predictedRating = 0.0f;
float cumulativeSum = 0.0f;
foreach (KeyValuePair <Tuple <uint, uint>, float> weights in weight)
{
if (trainingUsers.GetUser(weights.Key.Item2).RatingExists(movie.Key))
{
cumulativeSum += weights.Value *
(trainingUsers.GetRatingForUser(weights.Key.Item2, movie.Key) -
computation.WeightedSumOfUser(trainingUsers.GetUser(weights.Key.Item2)));
}
else
{
cumulativeSum += weights.Value;
}
}
// odd occurrence when there is no other weights.
if (sumWeight != 0.0f)
{
predictedRating = averageRatingOfActive + cumulativeSum / sumWeight;
}
else
{
predictedRating = averageRatingOfActive;
}
Console.WriteLine("Predicted Rating of " + movie.Value.Title + " is " + predictedRating);
}
}
else
{
Console.WriteLine("Invalid userID");
}
break;
case "exit":
Console.WriteLine("Exiting database...");
break;
default:
Console.WriteLine("Unknown value has been input.");
break;
}
} while(input != "exit");
}
