Display graphs of results from UDP packet loss tester

hist.py

import matplotlib.pyplot as plt
from scipy import stats
import numpy as np

# parse CSV file
# format is <size>, <recv_time>, <send_time>
# recv_time and send_time may be on different clocks so we can only compare vs. mean
lines =[]
for l in open('client_data/0.csv'):
	parts = [float(x) for x in l.strip().split(',')]
	lines.append(parts)	

arr = np.array(lines)
sizes = arr[:,0]
diffs = (arr[:,2] - arr[:,1])
diffs = diffs - np.mean(diffs) # relative to mean
diffs *= 1000 # work in msec

# Perform linear least squares to get an equation that estimates relative transmission time time based on size
a = np.vstack([np.ones(len(lines)), sizes])
a = np.transpose(a)
b = diffs
x = np.linalg.lstsq(a,b)[0]
b,m = x
print 'msec = %f + %f x size' % (b,m)
print 'Derived transmission rate: %g kB/s' % ((1.0/m)/1024.0*1000)
# Get the times predicted based on the linear equation (any residues we will consider as jitter)
predicted_diffs = np.dot(a,x)

# Show some stats about jitter
import scipy
z = (diffs-predicted_diffs)
print 'Based on average transmission rate and packet size:'
print '95%% of packets arrive within %f ms of predicted time' % ((scipy.stats.scoreatpercentile(z, 97.5) - scipy.stats.scoreatpercentile(z, 2.5)) * 0.5)
print '99%% of packets arrive within %f ms of predicted time' % ((scipy.stats.scoreatpercentile(z, 99.5) - scipy.stats.scoreatpercentile(z, 0.5)) * 0.5)

# Do a plot of size vs. relative recv-send time, and a histogram of the residues of recv-send time vs. predicted recv-send time
plt.subplot(211)
plt.plot(sizes,diffs,'.')
s_sizes = np.sort(sizes)
plt.plot(s_sizes, (b+m*s_sizes),'r')
plt.subplot(212)
plt.hist((diffs-predicted_diffs), bins=50)

# Show plot
plt.show()