C++ Efficiently Calculating a Running Median [duplicate]

Answer 1

The Jeff McClintock running median estimate. Requires keeping only two values. This example iterates over an array of sampled values (CPU consumption). Seems to converge relatively quickly (about 100 samples) to an estimate of the median. The idea is at each iteration the median inches toward the input signal at a constant rate. The rate depends on what magnitude you estimate the median to be. I use the average as an estimate of the magnitude of the median, to determines the size of each increment of the median. If you need your median accurate to about 1%, use a step-size of 0.01 * the average.

float median = 0.0f;
float average = 0.0f;

// for each sample
{
    average += ( sample - average ) * 0.1f; // rough running average.
    median += _copysign( average * 0.01, sample - median );
}