static void Main(string[] args)
{
const int usersCount = 470758, moviesCount = 4500;
const double targetMin = 1, targetMax = 5;
const int VSize = 5;
const double learningRate = 0.02;
const double maxError = 1.08;
const int maxIterations = 4;
const int batchCount = 10000;
List <Rate> rates = NetflixDataHelper.ReadDataSet(@"combined_data_1.txt");
Console.WriteLine("Dataset read: " + rates.Count + " records");
List <Chunk> chunks = NetflixDataHelper.GetChunks(rates, usersCount, moviesCount, batchCount);
Console.WriteLine("Feature matrix created");
Vector <double> rmseChunks = Vector <double> .Build.Dense(chunks.Count);
for (int k = 0; k < chunks.Count; k++)
{
FactorizationMachine fm =
new FactorizationMachine(chunks[0].X.ColumnCount, VSize, targetMin, targetMax, 0.1);
double kRMSE = fm.Learn(chunks, k, learningRate, maxIterations, maxError);
rmseChunks[k] = kRMSE;
}
var rmseMean = rmseChunks.Mean();
var rmseStdDev = rmseChunks.StandardDeviation();
Console.WriteLine("RMSE mean: " + rmseMean + "+-" + rmseStdDev);
Console.ReadKey();
}
public double Learn(List <Chunk> trainData, int skipIndex, double learningRate, int itCount, double err = 1.08)
{
double rmse = 5;
for (int it = 0; it < itCount; it++)
{
int startTime = Environment.TickCount;
NetflixDataHelper.Shuffle(ref trainData);
for (int k = 0; k < trainData.Count; k++)
{
if (k == skipIndex)
{
continue;
}
Chunk chunk = trainData[k];
int n = chunk.X.RowCount;
while (n > 1)
{
n--;
int m = rnd.Next(n + 1);
this.GradDescent(chunk.X.Row(m), chunk.Y[m], learningRate);
}
}
Vector <double> testEval = this.Predict(trainData[skipIndex].X);
Vector <double> e = testEval - trainData[skipIndex].Y;
rmse = Math.Sqrt(e.PointwisePower(2).Sum() / e.Count);
var time = (Environment.TickCount - startTime) / 1000.0;
Console.WriteLine($"Test batch - {skipIndex}, iteration - {it}, rmse: {rmse}, time: {time}");
if (rmse < err)
{
break;
}
}
return(rmse);
}
