-
Notifications
You must be signed in to change notification settings - Fork 2
/
dpgmm.py
263 lines (206 loc) · 8.3 KB
/
dpgmm.py
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
252
253
254
255
256
257
258
259
260
261
262
263
import numpy as np
import glob
import dpcluster as dpc
import pandas as pd
import os
import sys
try:
import matplotlib.pylab as plt
import matplotlib as mpl
except Exception, e:
sys.stderr.write("Matplotlib is not available!")
def loadData(filename, format="rttEstimate"):
"""Load a csv file in memory.
:returns: pandas DataFrame with the file data
"""
if format=="rttEstimate":
df = pd.read_csv(filename, sep=",", header=None, names=["ip", "peer", "rtt", "dstMac"])
elif format=="thomas":
# the filename is a directory containing several RTT measurements
# ..../ipSrc/ipDst/flowID/hour
data = []
for fi in glob.glob(filename):
tmp = pd.read_csv(fi, sep="\t", comment="s", header=None,
names=["rtt", "start_sec", "start_msec", "end_sec", "end_msec"],
usecols=["rtt","start_sec"])
val = fi.split("/")
tmp["ip"] = "{0}->{1}".format(val[-4], val[-3])
data.append(tmp)
df = pd.concat(data)
# The ip addresses become the index
df = df.set_index("ip")
return df
def clusterRTToverTime(rttEstimates, timeBin="60", outputDirectory="./rttDistributions/",
minEstimates=10, plot=True, logNormal=True):
"""For each IP address, find the different RTT distributions for each time
bin and plot the average value of each distribution.
"""
# for each IP in the traffic
ips = rttEstimates.index.unique()
for ip in ips:
start = rttEstimates[rttEstimates.index == ip].start_sec.min()
end = rttEstimates[rttEstimates.index == ip].start_sec.max()
dataIP = rttEstimates[rttEstimates.index == ip]
x = []
y = []
z = []
i = 0
for ts in range(start,end,timeBin):
if logNormal:
data = np.log10(dataIP[(dataIP.start_sec>=ts) & (dataIP.start_sec<ts+timeBin)].rtt)
else:
data = dataIP[(dataIP.start_sec>=ts) & (dataIP.start_sec<ts+timeBin)].rtt
# Look only at flows containing a certain number of RTT estimates
if len(data) < minEstimates:
sys.stderr("Ignoring data!! not enough samples!")
continue
# Cluster the data
vdp = dpgmm(data)
if vdp is None:
continue
params = NIWparam2Nparam(vdp)
if logNormal:
mean, std = logNormalMeanStdDev(params[0, :], params[1, :])
else:
mean = params[0, :]
std = params[1, :]
for mu, sig in zip(mean, std):
y.append(mu)
z.append(sig)
x.append(ts)
# Plot the clusters characteristics in a file
plt.figure()
plt.errorbar(x,y,yerr=z,fmt="o")
plt.grid(True)
if logNormal:
plt.savefig("{0}/{1}_timeBin{2}sec_logNormal.eps".format(outputDirectory, ip, timeBin))
else:
plt.savefig("{0}/{1}_timeBin{2}sec_normal.eps".format(outputDirectory, ip, timeBin))
def clusterRttPerIP(rttEstimates, outputDirectory="./rttDistributions/", minEstimates=10, plot=True, logNormal=False):
"""For each IP address, find the different RTT distributions and write
their mean and standard deviation in files.
"""
# for each IP in the traffic
ips = rttEstimates.index.unique()
for ip in ips:
if logNormal:
data = np.log10(rttEstimates[rttEstimates.index == ip].rtt)
else:
data = rttEstimates[rttEstimates.index == ip].rtt
# Look only at flows containing a certain number of RTT estimates
if len(data) < minEstimates:
continue
# Cluster the data
vdp = dpgmm(data)
if vdp is None:
continue
# Write the clusters characteristics in a file
fi = open("{0}/{1}.csv".format(outputDirectory, ip), "w")
params = NIWparam2Nparam(vdp)
if logNormal:
mean, std = logNormalMeanStdDev(params[0, :], params[1, :])
else:
mean = params[0, :]
std = params[1, :]
for mu, sig in zip(mean, std):
fi.write("{0},{1}\n".format(mu, sig))
if plot:
plotRttDistribution(rttEstimates, ip, "{0}/{1}.eps".format(outputDirectory, ip))
def NIWparam2Nparam(vdp, minClusterIPRatio=0.05):
"""
Convert Gaussian Normal-Inverse-Wishart parameters to the usual Gaussian
parameters (i.e. mean, standard deviation)
:vdp: Variational Dirichlet Process obtained from dpgmm
:minClusterIPRatio: Ignore distributions standing for a ratio of IPs lower
than minClusterIPRatio
"""
nbIPs = float(np.sum(vdp.cluster_sizes()))
mus, Sgs, k, nu = vdp.distr.prior.nat2usual(vdp.cluster_parameters()[
vdp.cluster_sizes() > (minClusterIPRatio * nbIPs), :])[0]
Sgs = Sgs / (k + 1 + 1)[:, np.newaxis, np.newaxis]
res = np.zeros( (len(mus), 2) )
for i, (mu, Sg) in enumerate(zip(mus, Sgs)):
w, V = np.linalg.eig(Sg)
V = np.array(np.matrix(V) * np.matrix(np.diag(np.sqrt(w))))
V = V[0]
res[i] = (mu[0], V[0])
return res
def logNormalMeanStdDev(loc, scale):
"""Compute the mean and standard deviation from the location and scale
parameter of a lognormal distribution.
:loc: location parameter of a lognormal distribution
:scale: scale parameter of a lognmormal distribution
:return: (mean,stdDev) the mean and standard deviation of the distribution
"""
mu = 10 ** (loc + ((scale ** 2) / 2.0))
var = (10 ** (scale ** 2) -1) * 10 ** (2 * loc + scale ** 2)
return mu, np.sqrt(var)
def dpgmm(data, priorWeight=0.1, maxClusters=32, thresh=1e-3, maxIter=10000):
"""
Compute the Variational Inference for Dirichlet Process Mixtures
on the given data.
:data: 1D array containing the data to cluster
:priorWeight: likelihood-prior distribution pair governing clusters.
:maxClusters: Maximum number of clusters
:
"""
data = np.array(data).reshape(-1, 1)
vdp = dpc.VDP(dpc.distributions.GaussianNIW(1), w=priorWeight, k=maxClusters, tol=thresh, max_iters=maxIter)
stats = vdp.distr.sufficient_stats(data)
vdp.batch_learn(stats)
return vdp
def plotRttDistribution(rttEstimates, ip, filename, nbBins=500, logscale=False):
"""Plot the RTT distribution of an IP address
:rttEstimates: pandas DataFrame containing the RTT estimations
:ip: IP address to plot
:filename: Filename for the plot
:nbBins: Number of bins in the histogram
:logscale: Plot RTTs in logscale if set to True
:returns: None
"""
if logscale:
data = np.log10(rttEstimates[rttEstimates.index == ip].rtt)
else:
data = rttEstimates[rttEstimates.index == ip].rtt
h, b=np.histogram(data, nbBins, normed=True)
plt.figure(1, figsize=(9, 3))
plt.clf()
ax = plt.subplot()
x = b[:-1]
ax.plot(x, h, "k")
ax.grid(True)
plt.title("%s (%s RTTs)" % (ip, len(data)))
if logscale:
plt.xlabel("log10(RTT)")
else:
plt.xlabel("RTT")
plt.ylabel("pdf")
minorLocator = mpl.ticker.MultipleLocator(10)
ax.xaxis.set_minor_locator(minorLocator)
plt.tight_layout()
plt.savefig(filename)
if __name__ == "__main__":
if len(sys.argv) < 2:
print("usage: python {0} rtt.csv [outputDirectory]".format(sys.argv[0]))
filename = sys.argv[1]
if len(sys.argv) > 2:
outputDirectory = sys.argv[2]
# Create the output directory if it doesn't exist
if not os.path.exists(outputDirectory):
os.mkdir(outputDirectory)
if filename.endswith(".csv"):
# Get RTT data from given file
rtt = loadData(filename, format="rttEstimate")
# Sample RTT estimates: samplingRate=0.1 means that 10% of the
# estimates will be used
samplingRate = 0.1
if samplingRate:
rtt = rtt.sample(frac=samplingRate)
# Find RTT distributions for each IP address
clusterRttPerIP(rtt, outputDirectory, logNormal=False)
else:
# Get RTT data from given file
rtt = loadData(filename, format="thomas")
# Find RTT distributions over time
clusterRTToverTime(rtt, 600, outputDirectory, logNormal=False)
#clusterRttPerIP(rtt, outputDirectory)