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
|
<!DOCTYPE html>
<html>
<head>
<meta http-equiv='content-type' value='text/html;charset=utf8'>
<meta name='generator' value='Ronn/v0.7.3 (http://github.com/rtomayko/ronn/tree/0.7.3)'>
<title>fitmle(1) - Fit a set of values with a power-law distribution</title>
<style type='text/css' media='all'>
/* style: man */
body#manpage {margin:0}
.mp {max-width:100ex;padding:0 9ex 1ex 4ex}
.mp p,.mp pre,.mp ul,.mp ol,.mp dl {margin:0 0 20px 0}
.mp h2 {margin:10px 0 0 0}
.mp > p,.mp > pre,.mp > ul,.mp > ol,.mp > dl {margin-left:8ex}
.mp h3 {margin:0 0 0 4ex}
.mp dt {margin:0;clear:left}
.mp dt.flush {float:left;width:8ex}
.mp dd {margin:0 0 0 9ex}
.mp h1,.mp h2,.mp h3,.mp h4 {clear:left}
.mp pre {margin-bottom:20px}
.mp pre+h2,.mp pre+h3 {margin-top:22px}
.mp h2+pre,.mp h3+pre {margin-top:5px}
.mp img {display:block;margin:auto}
.mp h1.man-title {display:none}
.mp,.mp code,.mp pre,.mp tt,.mp kbd,.mp samp,.mp h3,.mp h4 {font-family:monospace;font-size:14px;line-height:1.42857142857143}
.mp h2 {font-size:16px;line-height:1.25}
.mp h1 {font-size:20px;line-height:2}
.mp {text-align:justify;background:#fff}
.mp,.mp code,.mp pre,.mp pre code,.mp tt,.mp kbd,.mp samp {color:#131211}
.mp h1,.mp h2,.mp h3,.mp h4 {color:#030201}
.mp u {text-decoration:underline}
.mp code,.mp strong,.mp b {font-weight:bold;color:#131211}
.mp em,.mp var {font-style:italic;color:#232221;text-decoration:none}
.mp a,.mp a:link,.mp a:hover,.mp a code,.mp a pre,.mp a tt,.mp a kbd,.mp a samp {color:#0000ff}
.mp b.man-ref {font-weight:normal;color:#434241}
.mp pre {padding:0 4ex}
.mp pre code {font-weight:normal;color:#434241}
.mp h2+pre,h3+pre {padding-left:0}
ol.man-decor,ol.man-decor li {margin:3px 0 10px 0;padding:0;float:left;width:33%;list-style-type:none;text-transform:uppercase;color:#999;letter-spacing:1px}
ol.man-decor {width:100%}
ol.man-decor li.tl {text-align:left}
ol.man-decor li.tc {text-align:center;letter-spacing:4px}
ol.man-decor li.tr {text-align:right;float:right}
</style>
<style type='text/css' media='all'>
/* style: toc */
.man-navigation {display:block !important;position:fixed;top:0;left:113ex;height:100%;width:100%;padding:48px 0 0 0;border-left:1px solid #dbdbdb;background:#eee}
.man-navigation a,.man-navigation a:hover,.man-navigation a:link,.man-navigation a:visited {display:block;margin:0;padding:5px 2px 5px 30px;color:#999;text-decoration:none}
.man-navigation a:hover {color:#111;text-decoration:underline}
</style>
</head>
<!--
The following styles are deprecated and will be removed at some point:
div#man, div#man ol.man, div#man ol.head, div#man ol.man.
The .man-page, .man-decor, .man-head, .man-foot, .man-title, and
.man-navigation should be used instead.
-->
<body id='manpage'>
<div class='mp' id='man'>
<div class='man-navigation' style='display:none'>
<a href="#NAME">NAME</a>
<a href="#SYNOPSIS">SYNOPSIS</a>
<a href="#DESCRIPTION">DESCRIPTION</a>
<a href="#PARAMETERS">PARAMETERS</a>
<a href="#OUTPUT">OUTPUT</a>
<a href="#EXAMPLES">EXAMPLES</a>
<a href="#SEE-ALSO">SEE ALSO</a>
<a href="#REFERENCES">REFERENCES</a>
<a href="#AUTHORS">AUTHORS</a>
</div>
<ol class='man-decor man-head man head'>
<li class='tl'>fitmle(1)</li>
<li class='tc'>www.complex-networks.net</li>
<li class='tr'>fitmle(1)</li>
</ol>
<h2 id="NAME">NAME</h2>
<p class="man-name">
<code>fitmle</code> - <span class="man-whatis">Fit a set of values with a power-law distribution</span>
</p>
<h2 id="SYNOPSIS">SYNOPSIS</h2>
<p><code>fitmle</code> <var>data_in</var> [<var>tol</var> [TEST [<var>num_test</var>]]]</p>
<h2 id="DESCRIPTION">DESCRIPTION</h2>
<p><code>fitmle</code> fits the data points contained in the file <var>data_in</var> with a
power-law function P(k) ~ k<sup>-gamma</sup>, using the Maximum-Likelihood
Estimator (MLE). In particular, <code>fitmle</code> finds the exponent <code>gamma</code>
and the minimum of the values provided on input for which the
power-law behaviour holds. The second (optional) argument <var>tol</var> sets
the acceptable statistical error on the estimate of the exponent.</p>
<p>If <code>TEST</code> is provided, the program associates a p-value to the
goodness of the fit, based on the Kolmogorov-Smirnov statistics
computed on <var>num_test</var> sampled distributions from the theoretical
power-law function. If <var>num_test</var> is not provided, the test is based
on 100 sampled distributions.</p>
<h2 id="PARAMETERS">PARAMETERS</h2>
<dl>
<dt class="flush"><var>data_in</var></dt><dd><p> Set of values to fit. If is equal to <code>-</code> (dash), read the set from
STDIN.</p></dd>
<dt class="flush"><var>tol</var></dt><dd><p> The acceptable statistical error on the estimation of the
exponent. If omitted, it is set to 0.1.</p></dd>
<dt class="flush">TEST</dt><dd><p> If the third parameter is <code>TEST</code>, the program computes an estimate
of the p-value associated to the best-fit, based on <var>num_test</var>
synthetic samples of the same size of the input set.</p></dd>
<dt><var>num_test</var></dt><dd><p> Number of synthetic samples to use for the estimation of the
p-value of the best fit.</p></dd>
</dl>
<h2 id="OUTPUT">OUTPUT</h2>
<p>If <code>fitmle</code> is given less than three parameters (i.e., if <code>TEST</code> is
not specified), the output is a line in the format:</p>
<pre><code> gamma k_min ks
</code></pre>
<p>where <code>gamma</code> is the estimate for the exponent, <code>k_min</code> is the
smallest of the input values for which the power-law behaviour holds,
and <code>ks</code> is the value of the Kolmogorov-Smirnov statistics of the
best-fit.</p>
<p>If <code>TEST</code> is specified, the output line contains also the estimate of
the p-value of the best fit:</p>
<pre><code> gamma k_min ks p-value
</code></pre>
<p>where <code>p-value</code> is based on <var>num_test</var> samples (or just 100, if
<var>num_test</var> is not specified) of the same size of the input, obtained
from the theoretical power-law function computed as a best fit.</p>
<h2 id="EXAMPLES">EXAMPLES</h2>
<p>Let us assume that the file <code>AS-20010316.net_degs</code> contains the degree
sequence of the data set <code>AS-20010316.net</code> (the graph of the Internet
at the AS level in March 2001). The exponent of the best-fit power-law
distribution can be obtained by using:</p>
<pre><code> $ fitmle AS-20010316.net_degs
Using discrete fit
2.06165 6 0.031626 0.17
$
</code></pre>
<p>where <code>2.06165</code> is the estimated value of the exponent <code>gamma</code>, <code>6</code> is
the minimum degree value for which the power-law behaviour holds, and
<code>0.031626</code> is the value of the Kolmogorov-Smirnov statistics of the
best-fit. The program is also telling us that it decided to use the
discrete fitting procedure, since all the values in
<code>AS-20010316.net_degs</code> are integers. The latter information is printed
to STDERR.</p>
<p>It is possible to compute the p-value of the estimate by running:</p>
<pre><code> $ fitmle AS-20010316.net_degs 0.1 TEST
Using discrete fit
2.06165 6 0.031626 0.17
$
</code></pre>
<p>which provides a p-value equal to 0.17, meaning that 17% of the
synthetic samples showed a value of the KS statistics larger than that
of the best-fit. The estimation of the p-value here is based on 100
synthetic samples, since <var>num_test</var> was not provided. If we allow a
slightly larger value of the statistical error on the estimate of the
exponent <code>gamma</code>, we obtain different values of <code>gamma</code> and <code>k_min</code>,
and a much higher p-value:</p>
<pre><code> $ fitmle AS-20010316.net_degs 0.15 TEST 1000
Using discrete fit
2.0585 19 0.0253754 0.924
$
</code></pre>
<p>Notice that in this case, the p-value of the estimate is equal to
0.924, and is based on 1000 synthetic samples.</p>
<h2 id="SEE-ALSO">SEE ALSO</h2>
<p><span class="man-ref">deg_seq<span class="s">(1)</span></span>, <span class="man-ref">power_law<span class="s">(1)</span></span></p>
<h2 id="REFERENCES">REFERENCES</h2>
<ul>
<li><p>A. Clauset, C. R. Shalizi, and M. E. J. Newman. "Power-law
distributions in empirical data". SIAM Rev. 51, (2007), 661-703.</p></li>
<li><p>V. Latora, V. Nicosia, G. Russo, "Complex Networks: Principles,
Methods and Applications", Chapter 5, Cambridge University Press
(2017)</p></li>
</ul>
<h2 id="AUTHORS">AUTHORS</h2>
<p>(c) Vincenzo 'KatolaZ' Nicosia 2009-2017 <code><v.nicosia@qmul.ac.uk></code>.</p>
<ol class='man-decor man-foot man foot'>
<li class='tl'>www.complex-networks.net</li>
<li class='tc'>September 2017</li>
<li class='tr'>fitmle(1)</li>
</ol>
</div>
</body>
</html>
|
