tensorboard.smooth

tensorboard.smooth(values, weight)

Smooth values the way TensorBoard’s Scalars dashboard does.

An exponential moving average, divided by 1 - weight ** n after the n-th finite value so that it starts at the first value rather than near zero. A value that is not finite is kept where it is and left out of the average.

Parameters

Name	Type	Description	Default
values	numpy.ndarray	The values, in step order.	required
weight	float	The smoothing weight, from 0 (none) up to but not including 1.	required

Returns

Name	Type	Description
	numpy.ndarray	The smoothed values, as many as values.