public IntentsLoader(string fileName, SentenceVectorizer vectorizer)
{
this.FileName = fileName;
this.Vectorizer = vectorizer;
this.Initialize(FileName);
}
static void Main(string[] args)
{
var glovePath = Path.Combine(Directory.GetCurrentDirectory(), "data/glove.6B.300d.txt");
var dataPath = Path.Combine(Directory.GetCurrentDirectory(), "data/kc_data.json");
var stopwatch = new Stopwatch();
// double[][] trainData = null;
// double[][] testData = null;
// MakeTrainTest(Iris.Dataset(), out trainData, out testData);
Console.WriteLine("Loading GLoVE vectors from {0}", glovePath);
stopwatch.Start();
var reader = new WordVectorReader(glovePath);
stopwatch.Stop();
Console.WriteLine("Task finished in {0} ms", stopwatch.ElapsedMilliseconds);
stopwatch.Restart();
var vectorizer = new SentenceVectorizer(reader);
var loader = new DataLoader.IntentsLoader(dataPath, vectorizer);
var trainData = loader.GetData();
stopwatch.Stop();
Console.WriteLine("Task finished in {0} ms", stopwatch.ElapsedMilliseconds);
Console.WriteLine("Training started");
stopwatch.Restart();
var numInput = loader.InputDimensions;
var numHidden = 50;
var numOutput = loader.OutputDimensions;
var nn = new SparseNetwork(numInput, numHidden, numOutput);
var maxEpochs = 50;
double learnRate = 0.02;
nn.Train(trainData, maxEpochs, learnRate);
var trainAcc = nn.Accuracy(trainData);
Console.WriteLine("\nAccuracy on training data = " + trainAcc.ToString("F4"));
stopwatch.Stop();
Console.WriteLine("Task finished in {0} ms", stopwatch.ElapsedMilliseconds);
// var testAcc = nn.Accuracy(testData);
// Console.WriteLine("\nAccuracy on test data = " + testAcc.ToString("F4"));
// Console.WriteLine(nn.ToString());
// var test = new NeuralNetwork.Sparse.SparseMatrix<double>(10000000, 10000000);
// test[2, 4] = 1d;
// test[10000, 20] = 2d;
// test[10000, 30] = 4d;
// test[5000000, 5000000] = Math.PI;
// Console.WriteLine(test[2, 4]);
// Console.WriteLine(test[10000, 20]);
// Console.WriteLine(test[10000, 30]);
// Console.WriteLine(test[5000000, 5000000]);
// Console.WriteLine(test[10, 20]);
while (true)
{
Console.Write("Sentence for testing: ");
var sentence = Console.ReadLine();
if (sentence == "exit")
{
break;
}
stopwatch.Restart();
var sent_vector = vectorizer.SentenceToVector(sentence);
var result = nn.ComputeOutputs(sent_vector);
var resultIdx = SparseNetwork.MaxIndex(result);
stopwatch.Stop();
Console.WriteLine($"Result: {loader.Intents[resultIdx].name}, Confidence: {result[resultIdx]}");
Console.WriteLine($"Inference time: {stopwatch.ElapsedMilliseconds}");
}
}
