public static void MeanSquareError(TensorOld y, TensorOld yHat, TensorOld result)
{
//因为存在learning rate，所以梯度前面的系数不那么重要，但最好和损失函数一致，
TensorOld.Minus(yHat, y, result);
result.Multiple(2d / y.ElementCount);
}
public static void SquareError(TensorOld y, TensorOld yHat, TensorOld result)
{
TensorOld.Minus(y, yHat, result);
result.Apply(a => a * a);
}