Skip to content

Function: emaSmooth()

ts
function emaSmooth(values, weight?): Float64Array;

Defined in: ml/metrics.ts:189

TensorBoard-style debiased EMA smoothing of a noisy training curve. weight in [0,1) is the momentum (0 = raw, →1 = very smooth). Non-finite inputs pass through untouched and don't advance the average.

Parameters

ParameterTypeDefault value
valuesArrayLike<number>undefined
weightnumber0.6

Returns

Float64Array

MIT licensed. WebGL2 required.