stdev()'s standard deviation calculation is not correct

[Dieterbe]

maybe I'm missing something, but the formula used at https://github.com/graphite-project/graphite-web/blob/master/webapp/graphite/render/functions.py#L2054 seems incorrect
here's a simple test program to verify:

#!/usr/bin/env python2
import math

values = [5, 8, 7, 10]

s = sum(values)
print "sum", s

mean = s / len(values)
print "mean", mean

print "traditional:"
print math.sqrt(sum([(i - mean) ** 2 for i in values]) / len(values))

print "graphite:"
print math.sqrt(len(values) * sum([i ** 2 for i in values]) - s ** 2) / len(values)

output:

sum 30
mean 7
traditional:
1.73205080757
graphite:
1.80277563773

[Dieterbe]

i was trying to rewrite this algorithm and tune it, but then i wondered:
would it make sense to start doing more of this math using standard packages like numpy? might be faster too if numpy uses C modules.

[obfuscurity]

We've had conversations around numpy in the past but rejected the idea due to the dependencies.

[drawks]

I wouldn't call it a consensus, but... The last in depth conversation WRT inclusion of numpy seemed to conclude with wanting to modularize the rendering functions first and then it would be easy to provide numpy based functions for systems where the dependency isn't a problem. FWIW numpy is much faster than the inbuilt pure-python solutions for several of the core functions.

[pcn]

Has anyone taken a stab at the modularization?

[drawks]

@pcn I just added #988 to the milestones for 1.0.0

[stale]

This issue has been automatically marked as stale because it has not had recent activity. It will be closed if no further activity occurs. Thank you for your contributions.