public override NDArray Backward(NDArray input, NDArray gradOutput, ModelMode mode)
{
DNN.ConvolutionBackwardData(bwdDataAlgo, cd, workspace, weight, gradOutput, gradInput);
DNN.ConvolutionBackwardFilter(bwdFilterAlgo, cd, workspace, input, gradOutput, gradWeight);
DNN.ConvolutionBackwardBias(cd, gradOutput, gradBias);
return(gradInput);
}
public Conv2Cudnn(IAllocator allocator, SeedSource seedSource, DType elementType, int batchSize, int inputWidth, int inputHeight, int nInputPlane, int nOutputPlane, ConvolutionDesc2d cd)
: base(allocator, seedSource, elementType, batchSize, inputWidth, inputHeight, nInputPlane, nOutputPlane, cd)
{
// Reshape weight and bias - CuDNN expects the dimensions to be structured slightly differently
this.weight = ViewReplace(this.weight, nOutputPlane, nInputPlane, cd.kH, cd.kW);
this.bias = ViewReplace(this.bias, 1, nOutputPlane, 1, 1);
this.gradWeight = ViewReplace(this.gradWeight, this.weight.Shape);
this.gradBias = ViewReplace(this.gradBias, this.bias.Shape);
var fwdWorkspace = DNN.GetConvolutionForwardWorkspaceSize(allocator, fwdAlgo, cd,
new TensorShape(elementType, new long[] { batchSize, nInputPlane, inputHeight, inputWidth }),
new TensorShape(weight),
new TensorShape(activation));
var bwdFilterWorkspace = DNN.GetConvolutionBackwardFilterWorkspaceSize(allocator, bwdFilterAlgo, cd,
new TensorShape(elementType, new long[] { batchSize, nInputPlane, inputHeight, inputWidth }),
new TensorShape(activation),
new TensorShape(weight));
var bwdFilterInputWorkspace = DNN.GetConvolutionBackwardDataWorkspaceSize(allocator, bwdDataAlgo, cd,
new TensorShape(weight),
new TensorShape(activation),
new TensorShape(elementType, new long[] { batchSize, nInputPlane, inputHeight, inputWidth }));
var workspaceSize = Math.Max(Math.Max(fwdWorkspace, bwdFilterWorkspace), bwdFilterInputWorkspace);
this.workspace = (CudaStorage)allocator.Allocate(DType.UInt8, workspaceSize);
}
static void Main(string[] args)
{
double[][][] d = IO.LoadAllText(@"C:\DNN\test.txt");
double[][] o = new DNN(@"C:\DNN\a.dnn").Compute(d[0]);
ShowText(d[0], d[1], o);
Console.ReadLine();
}
public override Tensor Forward(Tensor input, ModelMode mode)
{
using (var input4d = As4d(input))
using (var activation4d = As4d(activation))
{
DNN.SoftmaxForward(DNNSoftmaxAlgorithm.Log, DNNSoftmaxMode.Instance, input4d, activation4d);
}
return(activation);
}
public override Tensor Backward(Tensor input, Tensor gradOutput, ModelMode mode)
{
using (var activation4d = As4d(activation))
using (var gradInput4d = As4d(gradInput))
using (var gradOutput4d = As4d(gradOutput))
{
DNN.SoftmaxBackward(DNNSoftmaxAlgorithm.Log, DNNSoftmaxMode.Instance, activation4d, gradInput4d, gradOutput4d);
}
return(gradInput);
}
static void Main(string[] args)
{
double[][][] io = IO.LoadAllText(@"C:\DNN\train.txt");
DNN d = new DNN(@"C:\DNN\a.dnn");
new Task(() => { d.Train(io[0], io[1]); }).Start();
while (true)
{
Thread.Sleep(1000 * 10);
Console.WriteLine(d.e);
d.Save();
}
}
public override NDArray Backward(NDArray input, NDArray gradOutput, ModelMode mode)
{
DNN.PoolingBackward(poolingDesc, input, activation, gradInput, gradOutput);
return(gradInput);
}
public override NDArray Forward(NDArray input, ModelMode mode)
{
DNN.PoolingForward(poolingDesc, input, activation);
return(activation);
}
public override Tensor Backward(Tensor input, Tensor gradOutput, ModelMode mode)
{
DNN.PoolingBackward(poolingDesc, input, activation, gradInput, gradOutput);
return(gradInput);
}
public override Tensor Forward(Tensor input, ModelMode mode)
{
DNN.PoolingForward(poolingDesc, input, activation);
return(activation);
}
public static void ComputeImg(string s)
{
double[] d = IO.LoadImg(s);
double[][] o = new DNN(@"C:\DNN\a.dnn").Compute(new double[][] { d });
Console.WriteLine(IO.double2int(o[0]));
}
public override NDArray Forward(NDArray input, ModelMode mode)
{
DNN.ConvForward(fwdAlgo, cd, workspace, input, weight, activation);
DNN.AddTensor(bias, activation); // dims of bias with size = 1 are automatically broadcast over other dimensions
return(activation);
}
