public async Task Test(string dataset, string filename)
{
ModelSerializer modelSerializer = new ModelSerializer();
model = modelSerializer.DeserializeCNN(filename);
fDataSet testdataset = new fDataSet();
testdataset.Deserializer(dataset);
int index = random.Next(0, testdataset.fData.Count);
AI.Core.fDataSet dataSet = new AI.Core.fDataSet();
dataSet.fData = testdataset.fData;
pixel = testdataset.fData[index].Pixel;
label = testdataset.fData[index].DecodeLabel;
//model.predict
probs = model.Predict(dataSet.fData[index]);
AI.ML.CNN.Trainers.ADAM adam = new AI.ML.CNN.Trainers.ADAM();
softmaxOutput = adam.Softmax(probs);
double maxValue = softmaxOutput.Max();
predict = softmaxOutput.ToList().IndexOf(maxValue);
//int lindex = random.Next(0, 5);
//predict = lindex == 2 ? random.Next(0, 9) : label;
result[0] = label;
result[1] = predict;
await Clients.All.SendAsync("ReceiveMessage", pixel);
await Clients.All.SendAsync("ReceiveLabel", result);
}
public async Task ConfigureModel(string configstring, string algorithm, string dataset, string filename)
{
await Clients.All.SendAsync("ReceiveMessage", "Configuring model...");
model = new AI.ML.CNN.Model();
model.Configure(configstring);
await Clients.All.SendAsync("ReceiveMessage", "Initializing dataset...");
switch (dataset)
{
case "MNIST":
case "mnist":
dataFileName = @"C:\files\data\mnist\train-images.idx3-ubyte";
labelFileName = @"C:\files\data\mnist\train-labels.idx1-ubyte";
dataList = (List <AI.Core.fData>)AI.Core.UByteLoader.ReadGrayImage(dataFileName, 1, 0.0, 1.0, labelFileName, 0.0, 1.0);
dataSet = new AI.Core.fDataSet();
dataSet.fData = dataList;
await Clients.All.SendAsync("SetCookie", dataset, "dataset");
break;
case "CIFAR":
case "cifar":
dataFileName = @"C:\files\data\cifar\cifar-10-batches-bin\data_batch_1.bin";
dataList = (List <AI.Core.fData>)AI.Core.UByteLoader.ReadColorImage(dataFileName, 1, 0.0, 1.0, 0.0, 1.0);
dataSet = new AI.Core.fDataSet();
dataSet.fData = dataList;
await Clients.All.SendAsync("SetCookie", dataset, "dataset");
break;
default:
dataSet = new AI.Core.fDataSet();
dataSet.Deserializer(dataset);
await Clients.All.SendAsync("SetCookie", dataset, "dataset");
break;
}
await Clients.All.SendAsync("ReceiveMessage", "Initializing training...");
//parse trainer string
AI.ML.CNN.Trainers.DeltaRule deltaRule = new AI.ML.CNN.Trainers.DeltaRule();
deltaRule.Configure <AI.ML.CNN.Lossfunc.CategoricalCrossEntropy>(model, epochs, dataSet, learningRate, momentum);
await Clients.All.SendAsync("ReceiveMessage", deltaRule.NextVerbose());
filename = @"C:\files\model\" + filename;
AI.Core.ModelSerializer modelSerializer = new AI.Core.ModelSerializer();
path = modelSerializer.Serialize(configstring, model, filename);
await Clients.All.SendAsync("SetCookie", path, "model");
await Clients.All.SendAsync("ReceiveMessage", "done");
}
public async Task Test(string dataset, string filename)
{
switch (dataset)
{
case "MNIST":
case "mnist":
dataFileName = @"C:\files\data\mnist\train-images.idx3-ubyte";
labelFileName = @"C:\files\data\mnist\train-labels.idx1-ubyte";
model = modelSerializer.DeserializeCNN(filename);
List <AI.Core.fData> dataList = (List <AI.Core.fData>)AI.Core.UByteLoader.ReadGrayImage(dataFileName, 1, 0.0, 1.0, labelFileName, 0.0, 1.0);
index = random.Next(0, dataList.Count);
dataSet.fData = dataList;
pixel = dataList[index].Pixel;
label = dataList[index].DecodeLabel;
//model.predict
probs = model.Predict(dataSet.fData[index]);
softmaxOutput = adam.Softmax(probs);
maxValue = softmaxOutput.Max();
predict = softmaxOutput.ToList().IndexOf(maxValue);
//int lindex = random.Next(0, 5);
//predict = lindex == 2 ? random.Next(0, 9) : label;
result[0] = label;
result[1] = predict;
await Clients.All.SendAsync("ReceiveMessage", pixel);
await Clients.All.SendAsync("ReceiveLabel", result);
break;
default:
dataSet = new AI.Core.fDataSet();
dataSet.Deserializer(dataset);
model = modelSerializer.DeserializeCNN(filename);
index = random.Next(0, dataSet.fData.Count);
pixel = dataSet.fData[index].Pixel;
label = dataSet.fData[index].DecodeLabel;
//model.predict
probs = model.Predict(dataSet.fData[index]);
softmaxOutput = adam.Softmax(probs);
maxValue = softmaxOutput.Max();
predict = softmaxOutput.ToList().IndexOf(maxValue);
//int lindex = random.Next(0, 5);
//predict = lindex == 2 ? random.Next(0, 9) : label;
result[0] = label;
result[1] = predict;
await Clients.All.SendAsync("ReceiveMessage", pixel);
await Clients.All.SendAsync("ReceiveLabel", result);
break;
}
}
public string Serialize(string config, object model, string saveas)
{
string modeltype = model.GetType().ToString();
switch (modeltype)
{
case "AI.ML.CNN.Model":
AI.ML.CNN.Model cnnmodel = (AI.ML.CNN.Model)model;
return(SerializeCNN(config, cnnmodel, saveas));
case "AI.ML.ANN.Model":
throw new NotImplementedException();
default:
throw new Exception("Model Serilizer not found");
}
throw new Exception();
}
public async Task Test(string dataset, string filename)
{
switch (dataset)
{
case "MNIST":
case "mnist":
dataFileName = @"C:\files\data\mnist\train-images.idx3-ubyte";
labelFileName = @"C:\files\data\mnist\train-labels.idx1-ubyte";
model = modelSerializer.DeserializeCNN(filename);
dataList = (List <AI.Core.fData>)AI.Core.UByteLoader.ReadGrayImage(dataFileName, 1, 0.0, 1.0, labelFileName, 0.0, 1.0);
index = random.Next(0, dataList.Count);
dataSet.fData = dataList;
pixel = dataList[index].Pixel;
label = dataList[index].DecodeLabel;
//model.predict
probs = model.Predict(dataSet.fData[index]);
softmaxOutput = adam.Softmax(probs);
maxValue = softmaxOutput.Max();
predict = softmaxOutput.ToList().IndexOf(maxValue);
size = (int)System.Math.Sqrt(dataSet.fData[0].Pixel.Length);
matImg = new Mat(size, size, DepthType.Cv8U, 3);
Gimg = matImg.ToImage <Gray, Byte>();
Gimage = setGrayPixels(Gimg, size, pixel);
input_size = new int[] { 256, 256 };
resizedPixel = Resize_Gray(Gimage, input_size);
//int lindex = random.Next(0, 5);
//predict = lindex == 2 ? random.Next(0, 9) : label;
result[0] = label;
result[1] = predict;
await Clients.All.SendAsync("ReceiveMessage", resizedPixel, 1);
await Clients.All.SendAsync("ReceiveLabel", result);
break;
case "CIFAR":
case "cifar":
dataFileName = @"C:\files\data\cifar\cifar-10-batches-bin\data_batch_1.bin";
model = modelSerializer.DeserializeCNN(filename);
dataList = (List <AI.Core.fData>)AI.Core.UByteLoader.ReadColorImage(dataFileName, 1, 0.0, 1.0, 0.0, 1.0);
index = random.Next(0, dataList.Count);
dataSet.fData = dataList;
pixel = dataSet.fData[index].Pixel;
label = dataSet.fData[index].DecodeLabel;
//model.predict
probs = model.Predict(dataSet.fData[index]);
softmaxOutput = adam.Softmax(probs);
maxValue = softmaxOutput.Max();
predict = softmaxOutput.ToList().IndexOf(maxValue);
size = (int)System.Math.Sqrt(dataSet.fData[0].Pixel.Length / 3);
matImg = new Mat(size, size, DepthType.Cv8U, 3);
Cimg = matImg.ToImage <Bgr, Byte>();
Cimage = setColorPixels(Cimg, size, pixel);
input_size = new int[] { 256, 256 };
resizedPixel = Resize_Color(Cimage, input_size);
//int lindex = random.Next(0, 5);
//predict = lindex == 2 ? random.Next(0, 9) : label;
cifarResult[0] = classes[label];
cifarResult[1] = classes[predict];
await Clients.All.SendAsync("ReceiveMessage", resizedPixel, 3);
await Clients.All.SendAsync("ReceiveLabel", cifarResult);
break;
default:
dataSet = new AI.Core.fDataSet();
dataSet.Deserializer(dataset);
pixel = dataSet.fData[index].Pixel;
model = modelSerializer.DeserializeCNN(filename);
index = random.Next(0, dataSet.fData.Count);
label = dataSet.fData[index].DecodeLabel;
//model.predict
probs = model.Predict(dataSet.fData[index]);
softmaxOutput = adam.Softmax(probs);
maxValue = softmaxOutput.Max();
predict = softmaxOutput.ToList().IndexOf(maxValue);
//int lindex = random.Next(0, 5);
//predict = lindex == 2 ? random.Next(0, 9) : label;
result[0] = label;
result[1] = predict;
if (model.Input.Output.Count == 3)
{
size = (int)System.Math.Sqrt(dataSet.fData[0].Pixel.Length / 3);
matImg = new Mat(size, size, DepthType.Cv8U, 3);
Cimg = matImg.ToImage <Bgr, Byte>();
Cimage = setColorPixels(Cimg, size, pixel);
input_size = new int[] { 256, 256 };
resizedPixel = Resize_Color(Cimage, input_size);
cifarResult[0] = classes[label];
cifarResult[1] = classes[predict];
await Clients.All.SendAsync("ReceiveMessage", resizedPixel, 3);
await Clients.All.SendAsync("ReceiveLabel", cifarResult);
}
else
{
size = (int)System.Math.Sqrt(dataSet.fData[0].Pixel.Length);
matImg = new Mat(size, size, DepthType.Cv8U, 3);
Gimg = matImg.ToImage <Gray, Byte>();
Gimage = setGrayPixels(Gimg, size, pixel);
input_size = new int[] { 256, 256 };
resizedPixel = Resize_Gray(Gimage, input_size);
//int lindex = random.Next(0, 5);
//predict = lindex == 2 ? random.Next(0, 9) : label;
result[0] = label;
result[1] = predict;
await Clients.All.SendAsync("ReceiveMessage", resizedPixel, 1);
await Clients.All.SendAsync("ReceiveLabel", result);
}
break;
}
}
public ML.CNN.Model DeserializeCNN(string filename)
{
CNNModel cnnmodel = JsonConvert.DeserializeObject <CNNModel>(File.ReadAllText(filename));
ML.CNN.Model Model = new AI.ML.CNN.Model();
Model.Configure(cnnmodel.config);
int count = 0;
int anncount = 0;
for (int i = 1; i < Model.Layers.Length; i++)
{
ML.CNN.Model.Unit lyr;
lyr = Model.Layers[i];
bool convtype = lyr.GetType().ToString() == "AI.ML.CNN.Layers.Convolution" ? true : false;
bool conctype = lyr.GetType().ToString() == "AI.ML.CNN.Layers.Connected" ? true : false;
switch (convtype)
{
case true:
ML.CNN.Layers.Convolution convLyr = (ML.CNN.Layers.Convolution)lyr;
AI.ML.CNN.Layers.Convolution.Kernel krn;
for (int j = 0; j < convLyr.Filters.Count; j++)
{
// j: indexing filters
for (int k = 0; k < ((AI.ML.CNN.Layers.Convolution.Filter)convLyr.Filters[j]).Kernels.Length; k++)
{
//k: indexing kernels
krn = (AI.ML.CNN.Layers.Convolution.Kernel)((AI.ML.CNN.Layers.Convolution.Filter)convLyr.Filters[j]).Kernels[k];
krn.Weights = cnnmodel.cnnWeights[count++].Weights;
}
}
break;
case false:
switch (conctype)
{
case true:
ML.CNN.Layers.Connected connLyr = (ML.CNN.Layers.Connected)lyr;
ML.ANN.Neuron n; ML.ANN.Synapse syn;
double?[] weights;
for (int j = 0; j < connLyr.Neurons.Length; j++)
{
n = connLyr.Neurons[j];
weights = cnnmodel.annWeights[anncount++].Weights;
for (int k = 0; k < n.Synapse.Count; k++)
{
syn = n.Synapse[k];
syn.W = weights[k];
}
}
break;
case false:
break;
}
break;
default:
break;
}
}
return(Model);
}
