Random, you're not so...

noiseDork.py

#!/bin/python

#Math is weird.
# I was having a discussion with a coworker who wanted to do site visit metrics, but not expose user information. His idea 
# was to have a very large user sample, but apply a large amount of noise to the individual signals. When averaged out, the
# noise cancels leaving a reasonably approximate metric. 
# I figured that the if the visit count is low, it would be literally lost in the noise. 
# Being suspiciously minded, I decided to test this theory.
# 
# Math is weird. 

import random


users = 1000000
sample = 0.0
realsample = 0.0

for x in xrange(0, users):
    maxvisits = 10
    #10% of users potentially visit "frequently"
    if random.randint(1, 100) < 10:
        maxvisits = 40
    # Make it noisy as hell.
    noise = random.randrange(-500, 500, 1)
    visits = random.randint(0, maxvisits)
    realsample += visits
    sample += visits + noise

print "Real Visits:" + repr(realsample) + \
        " reported Visits: " + repr(sample) + "\n"
print "Real Visit per user: " + repr(realsample / users) + \
      "Reported visits per user: " + repr(sample / users)
      
#sample output
# Real Visits:6351072.0 reported Visits: 6077616.0
# Real Visit per user: 6.351072Reported visits per user: 6.077616
# :boggles: