public void MnistTest()
{
string TestImages = "Assets/Resources//mnist/test/t10k-images.idx3-ubyte";
string TestLabels = "Assets/Resources//mnist/test/t10k-labels.idx1-ubyte";
var testingImageFile = new ImageFile(TestImages); // Testing Images
var testingLabelFile = new LabelFile(TestLabels); // Testing Labels
int correctGuessCount = 0;
for (int batchNum = 0; batchNum < 100; batchNum++)
{
int startPoint = batchNum;
int endPoint = startPoint + 1;
testingImageFile.ReadNextBatch(startPoint, endPoint);
testingLabelFile.ReadNextBatch(startPoint, endPoint);
neuralNet.FeedForward(testingImageFile.images[0]);
Debug.Log((int)ParseNetworkOutput(neuralNet.Outputs()));
Debug.Log(String.Join(",", testingLabelFile.labels[0]));
if (testingLabelFile.labels[0][(int)ParseNetworkOutput(neuralNet.Outputs())] > 0.0f)
{
correctGuessCount += 1;
}
}
Debug.Log((correctGuessCount / TestImages.Length) * 100);
}
// Note: Try convolutional NN?
public void MnistTrain()
{
string TrainImages = "Assets/Resources/mnist/train/train-images.idx3-ubyte";
string TrainLabels = "Assets/Resources/mnist/train/train-labels.idx1-ubyte";
var trainingImageFile = new ImageFile(TrainImages); // Training Images
var trainingLabelFile = new LabelFile(TrainLabels); // Training Labels
int batchSize = 8;
int totalBatchCount = (int)Math.Ceiling((double)trainingLabelFile.numberOfLabels / (double)batchSize);
Debug.Log("Number of Training Batches: " + totalBatchCount);
List <float> errors = new List <float>();
for (int batchNum = 0; batchNum < 100; batchNum++)
{
// Debug.Log("Doing batch: " + batchNum);
int startPoint = batchNum * batchSize;
int endPoint = startPoint + batchSize - 1;
if (endPoint >= trainingLabelFile.numberOfLabels)
{
endPoint = trainingLabelFile.numberOfLabels - 1;
}
trainingImageFile.ReadNextBatch(startPoint, endPoint);
trainingLabelFile.ReadNextBatch(startPoint, endPoint);
// Figure out how to slice arrays
// Maybe we modify the file I/O to only take it the batchSize number of image/label combos?
errors.Add(neuralNet.TrainUsingMiniBatch(trainingImageFile.images, trainingLabelFile.labels));
}
graph.ShowGraph(errors);
// MnistTest();
}
public IHttpActionResult UploadLabelFiles([FromUri] int orderDetailId)
{
// 将label文件反序列化，添加新的label文件，再序列化
var orderDetailInDb = _context.FBAOrderDetails.Find(orderDetailId);
var deList = DeserializeLabelFilesString(orderDetailInDb.LabelFiles);
var list = new List <LabelFile>();
list.AddRange(deList);
//从httpRequest中获取文件并写入磁盘系统
var filesGetter = new FilesGetter();
var filePathList = filesGetter.GetAndSaveMultipleFileFromHttpRequest(@"D:\Labels\");
if (filePathList.ToList().Count == 0)
{
throw new HttpResponseException(HttpStatusCode.BadRequest);
}
foreach (var f in filePathList)
{
var labelFile = new LabelFile();
labelFile.NameInSystem = f.Split('\\').Last();
labelFile.OriginalName = labelFile.NameInSystem.Split('-').Last();
labelFile.RootPath = f;
labelFile.UploadDate = DateTime.Now;
list.Add(labelFile);
}
orderDetailInDb.LabelFiles = SerializeLabelFiles(list);
_context.SaveChanges();
return(Ok(new { OrderDetailId = orderDetailId, list.Count }));
}
//We want to read every label file initially, so that each label file can be assigned to each file name that it lists, in the fileLookup dictionary.
//Loading of labels for actual use can then be done by calls to GetLabels.
public IniLabels()
{
fileLookup = new Dictionary <string, string>();
string labelDirectory = Path.Combine(MainForm.executableDir, "param_labels\\");
if (Directory.Exists(labelDirectory))
{
string[] files = Directory.GetFiles(labelDirectory);
foreach (string filePath in files)
{
if (Path.GetExtension(filePath).ToLowerInvariant() == ".ini")
{
LabelFile lf = new LabelFile(filePath);
foreach (string name in lf.fileNames)
{
fileLookup[name] = filePath;
}
}
}
}
}
