How Tensorboard Smooth

date_range 08/06/2018 00:47 infosorttoolslabeldeep learning
Currently, I find if I use a smooth window then I’ll get a better result than tensorflow smooth function. So I deep into the smooth function of tensorboard.
Tensorboard smooth function is explained here and used here.
It’s source code as follows
private resmoothDataset(dataset: Plottable.Dataset) {
let data = dataset.data();
const smoothingWeight = this.smoothingWeight;
let last = data.length > 0 ? data[0].scalar : NaN;
data.forEach((d) => {
if (!_.isFinite(last)) {
d.smoothed = d.scalar;
} else {
// 1storder IIR lowpass filter to attenuate the higher
// frequency components of the timeseries.
d.smoothed = last * smoothingWeight + (1  smoothingWeight) * d.scalar;
}
last = d.smoothed;
});
}
which can be rewrote in python as follows:
def smooth(scalars, weight): # 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
It’s a recursive function and we summarized its formula below,
where is our original data, is the iteration steps, and is the smooth weigths which is shown as a scroll widgets in tensorboard.