File: documentation.xml

package info (click to toggle)
constraint 0.3.0-5
  • links: PTS
  • area: main
  • in suites: etch, etch-m68k
  • size: 312 kB
  • ctags: 501
  • sloc: python: 2,736; xml: 165; makefile: 43; sh: 1
file content (206 lines) | stat: -rw-r--r-- 10,803 bytes parent folder | download | duplicates (2)
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
<?xml version="1.0" encoding="ISO-8859-1"?>

<!DOCTYPE article SYSTEM "/home/logilab/docbook/dtd/docbookx.dtd" >
<article id='constraint-propagation-manual' lang='en'>
  <artheader>
    <title>Constraint Propagation in Python</title>
    <author>
      <firstname>Alexandre</firstname>
      <surname>Fayolle</surname>
    </author>
    <abstract>
      <simpara>This manual presents how to use the constraint package to solve
	constraint propagation problems in Python. It also presents the
	the extension mechanisms available in the package</simpara>
    </abstract>
    <copyright><year>2002-2005</year><holder>Logilab</holder>
    </copyright>
    <revhistory>
      <revision>
	<revnumber>$Revision: 1.6 $</revnumber>
	<date>$Date: 2005/09/05 15:56:26 $</date>
	<authorinitials>$Author: alf $</authorinitials>
      </revision>
    </revhistory>
  </artheader>


  <sect1 id='using'><title>Using the constraint package</title>

    <sect2><title>Sample problem</title> 

      <para>Let's take a real-life problem to get started. You're organising a
      an international Python Programming event, and you have to schedule
      conferences. There are 10 different conferences, you have 3 conference
      rooms available during 2 days. Each conference is 4 hours long, so you
      can organize at most 2 conferences per day in a given room.</para> 

      <para>Conferences 3, 4, 5 and 6 have to take place in room C, because
      it's the only one with Internet access.</para>

      <para>Some of the speakers are not available on both days, so
      conferences 1, 5 and 10 have to take place on day 1, and conferences 2,
      3 4 and 9 on day 2.</para>

      <para>You have made a quick poll over the python mailing list, and it
      turns out that people attending some of the conferences are likely to be
      attending some other conferences too, so you want to make sure that such
      conferences are not scheduled at the same time. A careful statistical
      study has found 4 groups of potential attendees. The first group want to
      attend conferences 1, 2, 3 and 10, the second conferences 2, 6, 8 and 9
      the third group conferences 3, 5, 6 and 7, and the last group
      conferences 1, 3, 7 and 8.</para>

      <para>You've tried to put this on a whiteboard, but this quickly proved
      to be tedious, so you thought about using the constraint solving
      package.</para>

    </sect2>

    <sect2><title>Variables, Domains and Constraints</title>

      <para>The first thing to find out in order to use the constraint package
      is what the variables are, what their domains are and what the
      constraints between variables are.</para>

      <para>If we look at our problem, the variables are the conferences' room
      and time slot, and for each conference, the domain is the cross product
      of the set of available rooms with the set of available time slots. We
      could say that the conferences which require Internet access have a
      different domain, because they need to be in room C. This is perfectly
      valid. However, we will model this as a constraint.</para>

      <para>Variables are manipulated as names, stored in character
      strings. Domains are instances of the fd.FiniteDomain class, which is
      instantiated with a list of values. Domains are manipulated through a
      dictionnary mapping a variable to its domain. Do not use the same domain
      instance for several variables, because in the current implementation,
      this is guaranteed to break. The code looks like this:
<programlisting role='python' >
<emphasis role='comment'># import Repository class and fd module, </emphasis>
<emphasis role='keyword'>from</emphasis> logilab.constraint <emphasis role='keyword'>import</emphasis> *
variables = (<emphasis role='string'>'c01'</emphasis>,<emphasis role='string'>'c02'</emphasis>,<emphasis role='string'>'c03'</emphasis>,<emphasis role='string'>'c04'</emphasis>,<emphasis role='string'>'c05'</emphasis>,<emphasis role='string'>'c06'</emphasis>,<emphasis role='string'>'c07'</emphasis>,<emphasis role='string'>'c08'</emphasis>,<emphasis role='string'>'c09'</emphasis>,<emphasis role='string'>'c10'</emphasis>)
values = [(room,slot) 
          <emphasis role='keyword'>for</emphasis> room <emphasis role='keyword'>in</emphasis> (<emphasis role='string'>'room A'</emphasis>,<emphasis role='string'>'room B'</emphasis>,<emphasis role='string'>'room C'</emphasis>) 
          <emphasis role='keyword'>for</emphasis> slot <emphasis role='keyword'>in</emphasis> (<emphasis role='string'>'day 1 AM'</emphasis>,<emphasis role='string'>'day 1 PM'</emphasis>,<emphasis role='string'>'day 2 AM'</emphasis>,<emphasis role='string'>'day 2 PM'</emphasis>)]
domains = {}
<emphasis role='keyword'>for</emphasis> v <emphasis role='keyword'>in</emphasis> variables:
    domains[v]=fd.FiniteDomain(values)
</programlisting>
      </para>

      <para>Constraints, like domains are objects. So far the only class that
      can be used is fd.Expression and fd.BinaryExpression. We use the
      fd.make_expression factory function to build an instance of the right
      class, depending on the number of variables that is passed. This
      function takes a list of affected variables and a python expression that
      evaluates to true if the constraint is satisfied. </para>

      <para>We have several constraints on our variables. First some
	conferences need to take place in room C:
<programlisting role='python' >
constraints = []
<emphasis role='keyword'>for</emphasis> conf <emphasis role='keyword'>in</emphasis> (<emphasis role='string'>'c03'</emphasis>,<emphasis role='string'>'c04'</emphasis>,<emphasis role='string'>'c05'</emphasis>,<emphasis role='string'>'c06'</emphasis>):
    constraints.append(fd.make_expression((conf,),
                                          "%s[0] == 'room C'"%conf))
</programlisting >
      </para>

      <para>Availability of the speakers impose some more constraints:
<programlisting role='python' >
<emphasis role='keyword'>for</emphasis> conf <emphasis role='keyword'>in</emphasis> (<emphasis role='string'>'c01'</emphasis>,<emphasis role='string'>'c05'</emphasis>,<emphasis role='string'>'c10'</emphasis>):
    constraints.append(fd.make_expression((conf,),
                                          "%s[1].startswith(<emphasis role='string'>'day 1'</emphasis>)"%conf))
<emphasis role='keyword'>for</emphasis> conf <emphasis role='keyword'>in</emphasis> (<emphasis role='string'>'c02'</emphasis>,<emphasis role='string'>'c03'</emphasis>,<emphasis role='string'>'c04'</emphasis>,<emphasis role='string'>'c09'</emphasis>):
    constraints.append(fd.make_expression((conf,),
                                          "%s[1].startswith(<emphasis role='string'>'day 2'</emphasis>)"%conf))
</programlisting>
      </para>

      <para>Then we want to say that some of the conferences should not be
	scheduled at the same time:
<programlisting role='python'>
groups = ((<emphasis role='string'>'c01'</emphasis>,<emphasis role='string'>'c02'</emphasis>,<emphasis role='string'>'c03'</emphasis>,<emphasis role='string'>'c10'</emphasis>),
          (<emphasis role='string'>'c02'</emphasis>,<emphasis role='string'>'c06'</emphasis>,<emphasis role='string'>'c08'</emphasis>,<emphasis role='string'>'c09'</emphasis>),
          (<emphasis role='string'>'c03'</emphasis>,<emphasis role='string'>'c05'</emphasis>,<emphasis role='string'>'c06'</emphasis>,<emphasis role='string'>'c07'</emphasis>),
          (<emphasis role='string'>'c01'</emphasis>,<emphasis role='string'>'c03'</emphasis>,<emphasis role='string'>'c07'</emphasis>,<emphasis role='string'>'c08'</emphasis>))
<emphasis role='keyword'>for</emphasis> g <emphasis role='keyword'>in</emphasis> groups:
    <emphasis role='keyword'>for</emphasis> conf1 <emphasis role='keyword'>in</emphasis> g:
        <emphasis role='keyword'>for</emphasis> conf2 <emphasis role='keyword'>in</emphasis> g:
            <emphasis role='keyword'>if</emphasis> conf2 > conf1:
                constraints.append(fd.make_expression((conf1,conf2),
                                                      <emphasis role='string'>'%s[1] != %s[1]'</emphasis>%\
                                                        (conf1,conf2)))
</programlisting>
      </para>

      <para>Finally, no two conferences can be scheduled in the same room at
	the same time:
<programlisting role='python'>
<emphasis role='keyword'>for</emphasis> conf1 <emphasis role='keyword'>in</emphasis> variables:
    <emphasis role='keyword'>for</emphasis> conf2 <emphasis role='keyword'>in</emphasis> variables:
        <emphasis role='keyword'>if</emphasis> conf2 > conf1:
            constraints.append(fd.make_expression((conf1,conf2),
                                                  <emphasis role='string'>'%s != %s'</emphasis>%(conf1,conf2)))
</programlisting>
      </para>

    </sect2>

    <sect2><title>The Repository class</title>
      <para>Repository objects are used to hold the variables, domains and
	constraints describing the problem. A Solver object can solve the
	problem described by a Repository.</para>

      <para>Here's the final touch:
<programlisting role='python'>
r = Repository(variables,domains,constraints)
solutions = Solver().solve(r)
<emphasis role='keyword'>print</emphasis> solutions
</programlisting>
      </para>
    </sect2>

    <para>The code is available in the file conferences.py in the examples
      directory of the distribution. It finds 64 possible schedules in a
      couple of seconds on my machine.</para>

    <sect2><title>Performance considerations</title>
      <para>There is still a lot of things to be worked on with this
	package. Here are a few tips that can help you to use it in its
	current state:
	<itemizedlist>
	  <listitem><para>Try to avoid constraints with a lot of variables. It
	  is better to have several binary constraints than one big N-ary
	  constraint with several 'and' conditions</para></listitem>

	  <listitem><para>If you cannot avoid constraint with lots of
	  variable, put the constraints with less variables first in the list,
	  because they will get evaluated before, will take less time to
	  process, and will hopefully reduce the domains of the variables
	  playing a role in your big constraints</para>
	    </listitem>
	</itemizedlist>
      </para>
    </sect2>

  </sect1>
  <sect1 id='extending'><title>Extending the constraint package</title>

    <para>WRITE ME!</para>
<!--
    <sect2><title>The interfaces</title>
      <para></para>
    </sect2>

    <sect2><title>The abstract implementations</title>
      <para></para>
    </sect2>

    <sect2><title>AbstractConstraint Narrowing algorithm</title>
      <para></para>
    </sect2>
-->

  </sect1>
</article>