TensorBoard 标量图中 "smoothing" 参数背后的数学原理是什么?

What is the mathematics behind the "smoothing" parameter in TensorBoard's scalar graphs?

我认为它是某种移动平均线,但有效范围在 0 到 1 之间。

它被称为exponential moving average,下面是代码解释它是如何创建的。

假设所有 real 标量值都在一个名为 scalars 的列表中,平滑应用如下:

def smooth(scalars: List[float], weight: float) -> List[float]:  # Weight between 0 and 1
    last = scalars[0]  # First value in the plot (first timestep)
    smoothed = list()
    for point in scalars:
        smoothed_val = last * weight + (1 - weight) * point  # Calculate smoothed value
        smoothed.append(smoothed_val)                        # Save it
        last = smoothed_val                                  # Anchor the last smoothed value

    return smoothed

这是执行指数平滑的实际源代码片段,并在评论中解释了一些额外的去偏,以补偿零初始值的选择:

last = last * smoothingWeight + (1 - smoothingWeight) * nextVal

来源:https://github.com/tensorflow/tensorboard/blob/34877f15153e1a2087316b9952c931807a122aa7/tensorboard/components/vz_line_chart2/line-chart.ts#L714