public TensorOld PrepareTrain(TensorOld input)
{
PreparePredict(input);
BackwardOutput = input.GetSameShape();
FiltersGradient = Filters.GetSameShape();
BiasGradient = Bias.GetSameShape();
return(ForwardOutput);
}
/// <summary>
/// 训练前的准备工作，检查结构，分配内存等
/// </summary>
/// <param name="input">输入，主要使用结构信息</param>
/// <returns></returns>
public TensorOld PrepareTrain(TensorOld input)
{
if (input.Rank != 2)
{
throw new TensorShapeException("input tensor must have Rank=2");
}
ForwardOutput = new TensorOld(input.shape[0], UnitCount);
BackwardOutput = input.GetSameShape();
if (Weights == null)
{
SetWeights(TensorOld.RandGaussian(input.shape[1], UnitCount));
WeightsGradient = Weights.GetSameShape();
}
else
{
if (input.shape[1] != Weights.shape[0])
{
throw new TensorShapeException("input and weights are not match!");
}
}
if (Bias == null)
{
SetBias(TensorOld.Zeros(1, UnitCount));
BiasGradient = Bias.GetSameShape();
}
sampleStartIndex = new int[input.shape[0]];
errorStartIndex = new int[input.shape[0]];
for (int i = 0; i < sampleStartIndex.Length; i++)
{
sampleStartIndex[i] = i * input.shape[1];
errorStartIndex[i] = i * UnitCount;
}
return(ForwardOutput);
}