Just a thought -- logging as an issue so I don't forget / can solicit feedback.

At the moment we have stat.Mean(x, weights []float64) float64 which computes the mean of a set of data. The thought is to also provide something like stat/running

package running

type Mean struct {
    count float64
    mean float64
}

func (m *Mean) Mean() float64 {
    return m.mean
}

func (m *Mean) Count() float64 {
    return m.count 
}

func (m *Mean) Add(x float64) {
      m.mean = x + m.mean * m.count / (m.count + 1)
      m.count++
}

func (m *Mean) AddWeighted(x, weight float64){ .... }

func (m *Mean) Reset() { ... }

and there could be similar for Variance/StdDev, DiscountedMean (where the count is discounted by a constant, etc.). These could be relatively easily combined with a shim:

type MeanStd struct {
    running.Mean()
    running.StdDev()
}

func (m *MeanStd) Add(x float64) {
    m.Mean.Add(x)
    m.StdDev.Add(x)
}

This would help both with streams that are too large to want to put in a []float64 and data that comes from some source (say, a channel) and so cannot be put into a float64.

I don't think Add is the right verb, but I'm not sure what the right verb is (Accumulate seems too heavy, Append also seems wrong ...)