public void TestTwoAttributeAxes()
{
List <DataPoint> dataPoints = new List <DataPoint> {
new DataPoint {
Class = 1,
Attributes = new float[] { 123, 987 },
Categories = new uint[] { },
Weight = 1.0f,
},
new DataPoint {
Class = 0,
Attributes = new float[] { 123, 876 },
Categories = new uint[] { },
Weight = 1.0f,
},
new DataPoint {
Class = 1,
Attributes = new float[] { 456, 765 },
Categories = new uint[] { },
Weight = 1.0f,
},
new DataPoint {
Class = 1,
Attributes = new float[] { 456, 543 },
Categories = new uint[] { },
Weight = 1.0f,
},
};
using (CudaManager cudaManager = Provider.CudaManagerPool.GetCudaManagerForThread(Provider.Logger))
using (DecisionLearner decisionLearner = new DecisionLearner(cudaManager, dataPoints)) {
IDataNode root = decisionLearner.FitDecisionTree().Node;
Assert.AreEqual(1.0, Accuracy(root, dataPoints));
}
}
static void Main()
{
cuda = new CUDA(true);
cublas = new CUBLAS(cuda);
//allocate vector on cuda device in main thread
CudaManager.CallMethod(AllocateVectors);
//changing first vector from other thread
Thread changeThread = new Thread(ChangeVectorOnDevice_ThreadRun)
{
IsBackground = false
};
changeThread.Start();
//wait for changeThread to finish
autoResetEvent.WaitOne();
//getting vector from device in another one thread
Thread getThread = new Thread(GetVectorFromDevice_ThreadRun)
{
IsBackground = false
};
getThread.Start();
//wait for getThread to finish
autoResetEvent.WaitOne();
Console.WriteLine("({0}, {1}, {2}, {3}, {4})", vector2[0], vector2[1], vector2[2], vector2[3], vector2[4]);
Console.ReadKey(true);
}
public DecisionLearnerContext(CudaManager cudaManager, IEnumerable <IDataPoint> dataPoints)
{
CudaManager = cudaManager;
DataPoints = dataPoints.ToList();
TotalWeight = DataPoints.Sum(dp => dp.Weight);
if (DataPoints.Count > 0)
{
NumAttributeAxes = DataPoints[0].Attributes.Length;
NumCategoricalAxes = DataPoints[0].Categories.Length;
}
}
public void TestSum()
{
int[] numbers = Enumerable.Range(999, 10000).ToArray();
using (CudaManager cudaManager = Provider.CudaManagerPool.GetCudaManagerForThread(Provider.Logger))
using (CudaArray <int> numbersBuffer = numbers)
using (CudaArray <int> outputBuffer = new[] { 0 }) {
KernelManager kernels = new KernelManager(cudaManager);
for (int nBlocks = 1; nBlocks <= 1024; nBlocks *= 2)
{
for (int threadsPerBlock = 1; threadsPerBlock <= 1024; threadsPerBlock *= 2)
{
kernels["setIntKernel"].Arguments(outputBuffer, 0).ExecuteTask();
Assert.AreEqual(0, outputBuffer.Read()[0]);
kernels["sumToOutputKernel"].Arguments(numbersBuffer, numbers.Length, outputBuffer, SharedBuffer.Ints(threadsPerBlock)).Execute(nBlocks * threadsPerBlock, threadsPerBlock);
Assert.AreEqual(numbers.Sum(), outputBuffer.Read()[0]);
}
}
}
}
public void TestSimpleAttributeSplit()
{
List <DataPoint> dataPoints = new List <DataPoint> {
new DataPoint {
Class = 0,
Attributes = new float[] { 123 },
Categories = new uint[] { },
Weight = 1.0f,
},
new DataPoint {
Class = 1,
Attributes = new float[] { 456 },
Categories = new uint[] { },
Weight = 1.0f,
},
};
using (CudaManager cudaManager = Provider.CudaManagerPool.GetCudaManagerForThread(Provider.Logger))
using (DecisionLearner decisionLearner = new DecisionLearner(cudaManager, dataPoints)) {
IDataNode root = decisionLearner.FitDecisionTree().Node;
AttributeSplit attributeSplit = (AttributeSplit)root;
Assert.AreEqual(0, attributeSplit.Axis);
Assert.AreEqual(456, attributeSplit.SplitValue);
DataLeaf left = (DataLeaf)attributeSplit.Left;
Assert.AreEqual(1, left.ClassDistribution[0]);
Assert.AreEqual(0, left.ClassDistribution[1]);
DataLeaf right = (DataLeaf)attributeSplit.Right;
Assert.AreEqual(0, right.ClassDistribution[0]);
Assert.AreEqual(1, right.ClassDistribution[1]);
Assert.AreEqual(1.0, Accuracy(root, dataPoints));
}
}
public void TestCategoryPlusAttributeSplit()
{
List <DataPoint> dataPoints = new List <DataPoint> {
new DataPoint {
Class = 1,
Attributes = new float[] { 99, 987 },
Categories = new uint[] { 0x3, 0x1 },
Weight = 1.0f,
},
new DataPoint {
Class = 0,
Attributes = new float[] { 99, 987 },
Categories = new uint[] { 0x3, 0x1 | 0x2 },
Weight = 1.0f,
},
new DataPoint {
Class = 1,
Attributes = new float[] { 99, 765 },
Categories = new uint[] { 0x3, 0x1 | 0x2 },
Weight = 1.0f,
},
new DataPoint {
Class = 1,
Attributes = new float[] { 99, 765 },
Categories = new uint[] { 0x3, 0x1 },
Weight = 1.0f,
},
};
using (CudaManager cudaManager = Provider.CudaManagerPool.GetCudaManagerForThread(Provider.Logger))
using (DecisionLearner decisionLearner = new DecisionLearner(cudaManager, dataPoints)) {
IDataNode root = decisionLearner.FitDecisionTree().Node;
Assert.AreEqual(1.0, Accuracy(root, dataPoints));
}
}
public DecisionLearner(CudaManager cudaManager, IEnumerable <IDataPoint> dataPoints)
{
Context = new DecisionLearnerContext(cudaManager, dataPoints);
_attributeSplitter = new AttributeSplitter(Context);
_categoricalSplitter = new CategoricalSplitter(Context);
}
public GPUPolicyLearner()
{
_cudaManager = Provider.CudaManagerPool.GetCudaManagerForThread(Provider.Logger);
}
private static void GetVectorFromDevice_ThreadRun()
{
CudaManager.CallMethod(GetVectorFromDevice);
//releasing main thread
autoResetEvent.Set();
}
private static void ChangeVectorOnDevice_ThreadRun()
{
CudaManager.CallMethod(ChangeVectorOnDevice);
//releasing main thread
autoResetEvent.Set();
}