/// <summary>
/// Initialize the training session using the OrtEnv.
/// </summary>
/// <param name="env">Specifies the OrtEnv to use.</param>
public void Initialize(OrtEnv env)
{
NativeApiStatus.VerifySuccess(NativeMethodsTraining.OrtInitializeTraining(env.DangerousGetHandle(), m_param.DangerousGetHandle(), m_expectedInputs.DangerousGetHandle(), m_expectedOutputs.DangerousGetHandle()));
m_param.ExpectedInputs = getTensorDefs(m_expectedInputs);
m_param.ExpectedOutputs = getTensorDefs(m_expectedOutputs);
}
static void Main(string[] args)
{
string strModel;
int nBatch;
int nSteps;
if (!getArgs(args, out strModel, out nBatch, out nSteps))
{
return;
}
// Create the training session.
TrainingSession session = new TrainingSession();
// Set the error callback function called at the end of the forward pass.
session.Parameters.OnErrorFunction += OnErrorFunction;
// Set the evaluation callback function called after the error function on 'DISPLAY_LOSS_STEPS'
session.Parameters.OnEvaluationFunction += OnEvaluationFunction;
// Set the training and testing data batch callbacks called to get a new batch of data.
session.Parameters.OnGetTrainingDataBatch += OnGetTrainingDataBatch;
session.Parameters.OnGetTestingDataBatch += OnGetTestingDataBatch;
// Setup the training parameters.
session.Parameters.SetTrainingParameter(OrtTrainingStringParameter.ORT_TRAINING_MODEL_PATH, strModel);
session.Parameters.SetTrainingParameter(OrtTrainingStringParameter.ORT_TRAINING_INPUT_LABELS, "labels");
session.Parameters.SetTrainingParameter(OrtTrainingStringParameter.ORT_TRAINING_OUTPUT_LOSS, "loss");
session.Parameters.SetTrainingParameter(OrtTrainingStringParameter.ORT_TRAINING_OUTPUT_PREDICTIONS, "predictions");
session.Parameters.SetTrainingParameter(OrtTrainingStringParameter.ORT_TRAINING_LOG_PATH, "c:\\temp");
session.Parameters.SetTrainingParameter(OrtTrainingBooleanParameter.ORT_TRAINING_USE_CUDA, true);
session.Parameters.SetTrainingParameter(OrtTrainingBooleanParameter.ORT_TRAINING_SHUFFLE_DATA, false);
session.Parameters.SetTrainingParameter(OrtTrainingLongParameter.ORT_TRAINING_EVAL_BATCH_SIZE, nBatch);
session.Parameters.SetTrainingParameter(OrtTrainingLongParameter.ORT_TRAINING_TRAIN_BATCH_SIZE, nBatch);
session.Parameters.SetTrainingParameter(OrtTrainingLongParameter.ORT_TRAINING_NUM_TRAIN_STEPS, nSteps);
session.Parameters.SetTrainingParameter(OrtTrainingLongParameter.ORT_TRAINING_EVAL_PERIOD, 1);
session.Parameters.SetTrainingParameter(OrtTrainingLongParameter.ORT_TRAINING_DISPLAY_LOSS_STEPS, 400);
session.Parameters.SetTrainingParameter(OrtTrainingNumericParameter.ORT_TRAINING_LEARNING_RATE, 0.01);
session.Parameters.SetTrainingOptimizer(OrtTrainingOptimizer.ORT_TRAINING_OPTIMIZER_SGD);
session.Parameters.SetTrainingLossFunction(OrtTrainingLossFunction.ORT_TRAINING_LOSS_FUNCTION_SOFTMAXCROSSENTROPY);
session.Parameters.SetupTrainingParameters();
// Setup the training data information.
session.Parameters.SetupTrainingData(new List <string>()
{
"X", "labels"
});
// Load the MNIST dataset from file. See http://yann.lecun.com/exdb/mnist/ to get data files.
MnistDataLoaderLite dataLoader = new MnistDataLoaderLite("c:\\temp\\data");
dataLoader.ExtractImages(out m_rgTrainingData, out m_rgTestingData);
m_nTrainingDataIdx = 0;
m_nTestingDataIdx = 0;
// Setup the OnnxRuntime instance.
OrtEnv.SetLogLevel(LogLevel.Warning);
OrtEnv env = OrtEnv.Instance();
// Initialize the training session.
session.Initialize(env);
// Run the training session.
session.RunTraining();
session.EndTraining();
// Cleanup.
session.Dispose();
Console.WriteLine("Done!");
Console.WriteLine("press any key to exit...");
Console.ReadKey();
}
private void TestRegisterCustomOpLibrary()
{
using (var option = new SessionOptions())
{
string libName = "custom_op_library.dll";
string modelPath = "custom_op_test.onnx";
if (RuntimeInformation.IsOSPlatform(OSPlatform.Windows))
{
libName = "custom_op_library.dll";
}
else if (RuntimeInformation.IsOSPlatform(OSPlatform.Linux))
{
libName = "libcustom_op_library.so";
}
else if (RuntimeInformation.IsOSPlatform(OSPlatform.OSX))
{
libName = "libcustom_op_library.dylib";
}
string libFullPath = Path.Combine(Directory.GetCurrentDirectory(), libName);
Assert.True(File.Exists(libFullPath), $"Expected lib {libFullPath} does not exist.");
var ortEnvInstance = OrtEnv.Instance();
string[] providers = ortEnvInstance.GetAvailableProviders();
if (Array.Exists(providers, provider => provider == "CUDAExecutionProvider"))
{
option.AppendExecutionProvider_CUDA(0);
}
IntPtr libraryHandle = IntPtr.Zero;
try
{
option.RegisterCustomOpLibraryV2(libFullPath, out libraryHandle);
}
catch (Exception ex)
{
var msg = $"Failed to load custom op library {libFullPath}, error = {ex.Message}";
throw new Exception(msg + "\n" + ex.StackTrace);
}
using (var session = new InferenceSession(modelPath, option))
{
var inputContainer = new List <NamedOnnxValue>();
inputContainer.Add(NamedOnnxValue.CreateFromTensor <float>("input_1",
new DenseTensor <float>(
new float[]
{
1.1f, 2.2f, 3.3f, 4.4f, 5.5f,
6.6f, 7.7f, 8.8f, 9.9f, 10.0f,
11.1f, 12.2f, 13.3f, 14.4f, 15.5f
},
new int[] { 3, 5 }
)));
inputContainer.Add(NamedOnnxValue.CreateFromTensor <float>("input_2",
new DenseTensor <float>(
new float[]
{
15.5f, 14.4f, 13.3f, 12.2f, 11.1f,
10.0f, 9.9f, 8.8f, 7.7f, 6.6f,
5.5f, 4.4f, 3.3f, 2.2f, 1.1f
},
new int[] { 3, 5 }
)));
using (var result = session.Run(inputContainer))
{
Assert.Equal("output", result.First().Name);
var tensorOut = result.First().AsTensor <int>();
var expectedOut = new DenseTensor <int>(
new int[]
{
17, 17, 17, 17, 17,
17, 18, 18, 18, 17,
17, 17, 17, 17, 17
},
new int[] { 3, 5 }
);
Assert.True(tensorOut.SequenceEqual(expectedOut));
}
}
// Safe to unload the custom op shared library now
UnloadLibrary(libraryHandle);
}
}
