public RecommenderSystemEngine(RankingDB db)
{
usr2usr.Add(PredictionMethod.Pearson, new SimilartyData(Pearson));
usr2usr.Add(PredictionMethod.Cosine, new SimilartyData(Cosine));
_db = db;
MinSimilarity = 0.7;
}
//Compute the hit ratio of all the methods in the list for a given train-test split (e.g. 0.95 train set size)
/*public Dictionary<string, double> ComputeHitRatio(List<PredictionMethod> lMethods, double dTrainSetSize)
* {
* Dictionary<string, double> res = new Dictionary<string, double>();
* foreach (var m in lMethods)
* {
* res.Add(m.ToString(), CalcHitRatio(m, dTrainSetSize));
* }
* return res;
* }*/
/*private int GetGrade(PredictionMethod m, RankingDB train, RankingDB test)
* {
* var engineTrain = SetEngine(train);
*
*
* int grade = 0;
* var allTest = test.GetAllData();
* foreach (var rankData in allTest)
* {
* double pretictedRating = Math.Round(engineTrain.PredictRating(m, rankData.ranker, rankData.item));
* if (pretictedRating == rankData.rank)
++grade;
* }
* return grade;
* }
*
* private RecommenderSystemEngine SetEngine(RankingDB train)
* {
* var engineTrain = new RecommenderSystemEngine(train);
* if (_engine.Mode == FilterMode.MaxUsers)
* engineTrain.MaxUsers = _engine.MaxUsers;
* else
* engineTrain.MinSimilarity = _engine.MinSimilarity;
* return engineTrain;
* }*/
// the train and test are initiliazed
private void DivideDB(RankingDB allRatings, double dTrainSetSize, RankingDB train, RankingDB test)
{
var users = allRatings.GetRankers();
//quick add and contains
HashSet <string> chosenUsers = new HashSet <string>();
int testSize = (int)Math.Round((1 - dTrainSetSize) * allRatings.NumOfRanks());
while (test.NumOfRanks() < testSize)
{
var currentUser = chooseNewRandomUser(users, chosenUsers);
chosenUsers.Add(currentUser);
IEnumerable <RankingDB.RankData> randomRanks = chooseSomeRandomRanks(currentUser, allRatings);
train.Add(randomRanks);
var other = otherRanks(currentUser, randomRanks, allRatings);
int otherCount = allRatings.NumOfRanks(currentUser) - randomRanks.Count();
int diff = otherCount - (testSize - test.NumOfRanks());
if (diff > 0) //other count > free space in test
{
train.Add(other.Take(diff));
test.Add(other.Skip(diff));
}
else
{
test.Add(other);
}
}
foreach (var usr in users)
{
if (!chosenUsers.Contains(usr))
{
train.AddUser(usr, allRatings.GetUserData(usr));
}
}
}
public void Load(string sFileName, double dTrainSetSize)
{
RankingDB db = Load(sFileName);
_test = new RankingDB();
_train = new RankingDB();
DivideDB(db, dTrainSetSize, _train, _test);
_trainEngine = new RecommenderSystemEngine(_train);
}
//load a dataset from a file
public RankingDB Load(string sFileName)
{
RankingDB db = new RankingDB();
//_engine = new RecommenderSystemEngine(_ratings);
var lines = File.ReadLines(sFileName);
//var lines = LoadPart(sFileName, 50000);
foreach (var line in lines)
{
var parts = line.Split(',');
db.Add(parts[0], parts[1], int.Parse(parts[2]));
}
return(db);
}
private IEnumerable <RankingDB.RankData> chooseSomeRandomRanks(string user, RankingDB allRatings)
{
//var usrRanks = allRatings.GetRanksAndItems(user);
var items = allRatings.getItems(user);
Random rand = new Random();
int k = rand.Next(1, items.Count() + 1);
RankingDB.RankData[] res = new RankingDB.RankData[k];
for (int i = 0; i < k; i++)
{
int index = rand.Next(items.Count());
string item = items.ElementAt(index);
res[i] = new RankingDB.RankData(user, item, allRatings.GetRank(user, item).Value);
}
return(res);
}
public void TrainBaseModel(int cFeatures)
{
double avg = (_train.SumRanks() + _test.SumRanks()) / (_train.NumOfRanks() + _test.NumOfRanks());
var svd = new SVD(avg, cFeatures);
RankingDB train = new RankingDB();
RankingDB validation = new RankingDB();
DivideDB(_train, 0.95, train, validation);
var ranks = train.GetAllData();
double RMSE = double.MaxValue; double LastRMSE = double.MaxValue;
while (RMSE <= LastRMSE)
{
LastRMSE = RMSE;
svd.Train(ranks);
_trainEngine.setSVD(svd);
RMSE = ComputeRMSE(PredictionMethod.SVD, _trainEngine, validation.GetAllData());
}
//_engine.setSVD(svd);
}
private double ComputeRMSE(PredictionMethod m, RankingDB db)
{
return(ComputeRMSE(m, new RecommenderSystemEngine(db), db.GetAllData()));
}
private IEnumerable <RankingDB.RankData> otherRanks(string currentUser, IEnumerable <RankingDB.RankData> ranks, RankingDB allRatings)
{
return(allRatings.GetRanksAndItems(currentUser).Except(ranks));
}
