void UpdateWeights(NLPFeatureDataModel Feature, double LRate)
{
for (int i = 0; i < layers.Count; i++)
{
double[] inputs = new double[Feature.PhraseFeatures.Length - 1];
for (int j = 0; j < Feature.PhraseFeatures.Length - 1; j++)
{
inputs[j] = Feature.PhraseFeatures[j]; //remove last feature
}
if (i != 0)
{
inputs = new double[layers[i - 1].Length];
for (int j = 0; j < layers[i - 1].Length; j++)
{
inputs[j] = layers[i - 1][j].output;
}
}
for (int j = 0; j < layers[i].Length; j++)
{
Neuron neuron = layers[i][j];
for (int k = 0; k < inputs.Length; k++)
{
layers[i][j].weight[k] += LRate * layers[i][j].delta * inputs[k];
}
layers[i][j].weight[layers[i][j].weight.Length - 1] += LRate * layers[i][j].delta;
}
}
}
public void GetInitilizationData()
{
List <NLPDataModel> rawTrainingData = new List <NLPDataModel>();
FileProcessing fileProcessing = new FileProcessing();
tokenizer = new Tokenizer();
trainingData.Clear();
rawTrainingData = fileProcessing.ReadCSVLines <NLPDataModel>("NLPBagofWords.csv", true).ToList();
tokenizer.Keywords = fileProcessing.Headings;
foreach (NLPDataModel data in rawTrainingData)
{
NLPFeatureDataModel feature = new NLPFeatureDataModel();
var properties = typeof(NLPDataModel).GetProperties();
feature.PhraseFeatures = new double[properties.Length];
for (int i = 0; i < properties.Length; i++)
{
NLPDataModel d = new NLPDataModel();
double f = -999;
properties[i].GetValue(d);
string token = (string)properties[i].GetValue(data);
Double.TryParse(token, out f);
feature.PhraseFeatures[i] = f;
}
int phaseType = -999;
int.TryParse(data.PhraseType, out phaseType);
feature.PhaseType = phaseType;
trainingData.Add(feature);
}
}
public void GuessSentence()
{
network.GetInitilizationData();
network.InitializeNeuralNetwork();
network.TrainNetwork(network.TrainingData, 0.3, (network.TrainingData.Count - 1), 3);
NLPFeatureDataModel inputData = network.NormalizeInput("Please create an SOW");
//inputData.PhraseFeatures[0] = 0;
//inputData.PhraseFeatures[1] = 0;
//inputData.PhraseFeatures[2] = 0;
network.ForwardPropigation(inputData.PhraseFeatures);
Assert.IsTrue(true);
}
public virtual async Task MessageReceivedAsync(IDialogContext context, IAwaitable <IMessageActivity> result)
{
var message = await result;
NLPFeatureDataModel inputData = network.NormalizeInput(message.Text);
double[] guesses = network.ForwardPropigation(inputData.PhraseFeatures);
string r = "";
r += "I do not know what you are asking: " + (guesses[0] * 100) + "%";
r += ", Create an SOW: " + (guesses[1] * 100) + "%";
r += ", Create a new Client: " + (guesses[2] * 100) + "%";
if ((guesses[1] * 100) > 90)
{
context.Call(new SOWForm(), ResumeAfterOptionDialog);
}
}
public NLPFeatureDataModel NormalizeInput(string InputString)
{
NLPFeatureDataModel feature = new NLPFeatureDataModel();
feature.PhraseFeatures = new double[typeof(NLPDataModel).GetProperties().Length - 1];
List <string> tokens = tokenizer.TokenizeString(InputString);
foreach (string token in tokens)
{
for (int i = 0; i < tokenizer.Keywords.Count; i++)
{
if (tokenizer.Keywords[i] == token)
{
feature.PhraseFeatures[i] = 1;
}
}
}
return(feature);
}
