momentum

A trivial mini-package for computing the running univariate mean, variance, kurtosis and skew

• No dependencies ... not even numpy.
• No classes ... unless you want them.
• State is a dict, for trivial serialization.
• Tested against scipy, creme, statistics

For multivariate covariance updating, maybe see precise.

Install

pip install momentum

Usage: running mean, var

from momentum import var_init, var_update
from pprint import pprint

m = var_init()
for x in [5,3,2.4,1.0,5.0]:
    m = var_update(m,x)
pprint(m)

Usage: running mean, var, kurtosis and skew

from momentum import kurtosis_init, kurtosis_update

m = kurtosis_init()
for x in [5,3,2.4,1.0,5.0]:
    m = kurtosis_update(m,x)
pprint(m)

File an issue if you need more help using this.

Usage: running recency-weighted mean, var

from momentum import rvar_init, rvar_update
from pprint import pprint

m = rvar_init(rho=0.01,n=15)
for x in [5,3,2.4,1.0,5.0]:
    m = rvar_update(m,x)
pprint(m)

This will switch from running variance to a weighted variance after 15 data points.