nojhan/eodev
Archived
1
0
Fork 0
Code Issues Pull requests Projects Releases Packages Wiki Activity

Checking in the name changes in selection procedures.

Browse source 
This impacts on many files, creating new entries in src (the old ones are moved
to obsolete dir), modifying t-eoSymreg.cpp and t-eobin.cpp in test,
as well as gprop.cc and mastermind.cc in app dir (not to mention almost all
files in tutorial:-(
This commit is contained in:
evomarc 2001-01-05 05:42:08 +00:00
commit a998ad0a41
36 changed files with 922 additions and 226 deletions
Download patch file Download diff file Expand all files Collapse all files

242
eo/tutorial/html/eoEngine.html
View file

@ -42,21 +42,163 @@ helper classes that are used within selection and replacement procedures
are presented.
<p>
<hr WIDTH="100%"><a NAME="selection"></a><b><font color="#000099"><font size=+2>Selection</font></font></b>
<br>&nbsp;
<br>&nbsp;
<p>The very beginning of the generation loop is the selection phase, where
some individuals from the population are chosen to become the parents,
to be later modified by the variation operators and become the offspring.
This is the first step of the <font color="#009900">artificial Darwinism</font>,
where the <i><font color="#FF6600">fittest individuals are allowed to reproduce</font></i>.
<br>Conceptually, there are two distinct ways to choose the lucky ones:
one by one from the very same population (i.e. with replacement), which
means that at the extreme the same individual can be chosen every time;
or as a whole, in some sort of batch procedure. Of course, repeated selection
of one individual results in a batch selection!
<p>There are hence two basic EO classes for selection: <font color="#009900">eoSelectOne</font>
and <font color="#009900">eoSelect</font>, with different interfaces.
<br>
<hr WIDTH="50%">
<br><b><font color="#000099">eoSelectOne: </font><font color="#FF0000">The
interface</font></b>
<p>The abstract class for selection of a single individual from a population
is <font color="#009900">eoSelectOne</font>, and the interface for its
<tt><font color="#993300">operator()</font></tt>
is
<center>
<p><b><tt><font color="#993300">const EOT &amp; operator()(const eoPop&lt;EOT>&amp;
_parents)</font></tt></b></center>
<p>which you could have guessed from the inheritance tree for class <b><font face="Arial,Helvetica"><font size=+1><a href="../../doc/html/class_eoselectone.html">eoSelectOne</a></font></font></b>.,
as you see there that <font color="#009900">eoSelectOne</font> derives
from <tt><font color="#993300">class eoUF&lt;const eoPop&lt;EOT>&amp;,
const EOT&amp;></font></tt>.
<br>This means that it takes <font color="#FF6600">1 population</font>
(without the right to modify it - see the <b><tt><font color="#993300">const</font></tt></b>
keyword in argument) and returns a const reference to an individual (again,
the <b><tt><font color="#993300">const</font></tt></b> keyword ensures
that nothing will happen to the individual in the population - remember
it returns a reference).
<br>
<hr WIDTH="50%">
<br><b><font color="#000099">eoSelectOne: </font><font color="#FF0000">Instances</font></b>
<ul>
<li>
<b><tt><font color="#009900">eoDetTournamentSelect</font></tt></b> uses
the <a href="#detTournament">(deterministic) tournament</a> to choose one
individual. Its constructor has one parameter, the tournament size (integer
>= 2).</li>
<li>
<b><tt><font color="#009900">eoStochTournamentSelect</font></tt></b> uses
the <a href="#stochTournament">binary stochastic tournament</a> to choose
one individual. Its constructor has one parameter, the tournament rate
(real in [0.5,1]).</li>
<li>
&nbsp;<b><tt><font color="#009900">eoProportionalSelect</font></tt></b>
is the original <font color="#FF6600">roulette wheel</font> selection:
each parent is selected with a probability proportional to its fitness.</li>
<li>
<b><tt><font color="#009900">eoRandomSelect</font></tt></b> is the <font color="#FF6600">random</font>
selection and should give bad results! At the moment, it selects one individual
uniformly, but it would be easy to use any probability distribution.</li>
</ul>
<hr WIDTH="50%">
<br><b><font color="#000099">eoSelect: </font><font color="#FF0000">The
interface</font></b>
<p>The abstract class for batch selection of a&nbsp; whole set of individuals
from a population is <font color="#009900">eoSelect</font>, and the interface
for its
<tt><font color="#993300">operator()</font></tt> is
<center>
<p><b><tt><font color="#993300">void operator()(const eoPop&lt;EOT>&amp;
_source, eoPop&lt;EOT>&amp; _dest)</font></tt></b></center>
<p>which you could have guessed from the inheritance tree for class <b><font face="Arial,Helvetica"><font size=+1><a href="../../doc/html/class_eoselect.html">eoSelect</a></font></font></b>.,
as you see there that <font color="#009900">eoSelect</font> derives from
<tt><font color="#993300">class
eoBF&lt;const eoPop&lt;EOT>&amp;, eoPop&lt;EOT>&amp;, void></font></tt>.
<br>This means that it takes <font color="#FF6600">2 populations</font>,
and fills the second one with individuals from the first one without modifying
it (see the <b><tt><font color="#993300">const</font></tt></b> keyword).
This raises two questions:
<ul>
<li>
How does it know how many individuals to select?</li>
<li>
How to use repeated selection of one individual (see above the <font color="#009900">eoSelectOne</font>
class)?</li>
</ul>
<hr WIDTH="50%">
<br><b><font color="#000099">eoSelect: </font><font color="#FF0000">HowMany</font></b>
<p>There are two ways an&nbsp; can derive the number of individuals it
has to select: either it is a fixed number, or it is some percentage of
the source population size (any positive real number). In both case, this
must be passed to the constructor. In most instances, however, the constructor
will accept a real number (double) and a boolean indicating whether this
real number should be used as an absolute integer or as a rate, thanks
to the <a href="#howmany">eoHowMany</a> class.
<p><b><font color="#FF0000">Note</font></b>: an <font color="#009900">eoSelect</font>
can select more individuals than there are in the original population.
It is the job of the <font color="#009900">replacement</font> to ensure
that the population size does not grow along the generations.
<p>
<hr WIDTH="50%"><b><font color="#000099">eoSelectMany: </font><font color="#FF0000">Encapsulating
eoSelectOne</font></b>
<p>It is clear that repeated selection of a single individual is a way
to do batch selection. This is why it is possible to encapsulate an object
of class <font color="#009900">eoSelectOne</font> into an object of class
<font color="#009900">eoSelect</font>
using the class <font color="#009900">eoSelectMany</font>. Class <font color="#009900">eoSelectMany</font>
is derived from class <font color="#009900">eoSelect</font> and takes in
its constructor an <font color="#009900">eoSelectOne</font> (plus the number
of individuals it should select, according to the <a href="#howmany">eoHowMany</a>
paradigm).
<p><b><font color="#FF0000">Note</font></b>: some procedures for selecting
a single individual require some pre-processing of the whole population
that takes place before any selection, and will be repeated identically
for every individual. The encapsulation of an&nbsp; into an&nbsp; allows
to call such technical processing only once through the use of method setup
of class . This method does nothing by default, but is mandatory
<br>
<hr WIDTH="50%">
<br><b><font color="#000099">eoSelect: </font><font color="#FF0000">Other
instances</font></b>
<ul>
<li>
<b><tt><font color="#009900">eoDetSelect</font></tt></b> selects individuals
<b><font color="#FF6600">deterministically</font></b>,
i.e. starting from the best ones down to the worse ones. If the total number
to select is less than the size of the source populations, the best individuals
are selected once. If more individuals are needed after reaching the bottom
of the population, then the selection starts again at top. It the total
number required is N times that of the source size, all individuals are
selected exactly N times.</li>
</ul>
No other instances of <font color="#009900">eoSelect</font> that are not
encapsualtions of <font color="#009900">eoSelectOne</font> procedures are
avaiable as of today (Jan. 4 2001).
<br>
<hr WIDTH="100%">
<br><a NAME="replacement"></a><b><font color="#000099"><font size=+2>Replacement</font></font></b>
<p>The replacement phase takes place <font color="#FF6600">after the birth
of all offspring</font> through variation operators. The idea is to close
the generation loop, i.e. to end up with a population of individuals that
will be the initial population of next generation. That population will
be <font color="#FF6600">built upon the old parents and the new-born offspring</font>.
In all algorithms that come up with EO, the <font color="#FF6600">population
size</font> is supposed to be <font color="#FF6600">constant</font> from
one generation to the next one - though nothing stops you from writing
an algorithm with varying population size.
<p><b><font color="#000099">Replacement: </font><font color="#FF0000">The
<p>The replacement phase takes plac<font color="#000000">e after the birth
of all offspring through </font>variation operators. This is the second
step of the <font color="#009900">artificial Darwinism</font>, where the
<i><font color="#FF6600">fittest
individuals are allowed to survive</font></i>.
<br>It can also be viewed on the algorithmic side as closing the generation
loop, i.e. building the population that will be the initial population
of next generation. That population will be <font color="#FF6600">built
upon the old parents and the new-born offspring</font>. In all algorithms
that come up with EO, the <font color="#FF6600">population size</font>
is supposed to be <font color="#FF6600">constant</font> from one generation
to the next one - though nothing stops you from writing an algorithm with
varying population size.
<br>
<hr WIDTH="50%">
<br><b><font color="#000099">Replacement: </font><font color="#FF0000">The
interface</font></b>
<p>The abstract class for replacement procedures is the functor class
<font color="#009900">eoReplacement</font>,
@ -75,7 +217,9 @@ void></font></tt>.
and is free to modify both, but the resulting population should be placed
in the first argument (usually called_parents) to close the loop and go
to next generation.
<p><b><font color="#000099">Replacement: </font><font color="#FF0000">Instances</font></b>
<br>
<hr WIDTH="50%">
<br><b><font color="#000099">Replacement: </font><font color="#FF0000">Instances</font></b>
<ul>
<li>
<b><tt><font color="#009900">eoGenerationalReplacement </font></tt></b>This
@ -86,22 +230,27 @@ that offspring and parents are of the same size (but does not check!).</li>
<br>&nbsp;
<li>
<b><tt><font color="#009900">eoMergeReduce</font></tt></b> This is one
the basic types of replacement in EO. It has two major steps, <font color="#FF6600">merging</font>
both populations of parents and offspring, and <font color="#FF6600">reducing</font>
this big population to the right size. It <font face="Arial,Helvetica"><font size=+1><a href="../../doc/html/class_eomegereduce.html">contains
<a NAME="SSGA"></a><b><tt><font color="#009900">eoMergeReduce</font></tt></b>
This is one the basic types of replacement in EO. It has two major steps,
<font color="#FF6600">merging</font> both populations of parents and offspring,
and <font color="#FF6600">reducing</font> this big population to the right
size. It <font face="Arial,Helvetica"><font size=+1><a href="../../doc/html/class_eomegereduce.html">contains
two objects</a></font></font> of respective types <b><font color="#009900"><a href="#merge">eoMerge</a></font></b>
and <b><font color="#009900"><a href="#reduce">eoReduce</a></font></b>
and you can probably guess what each of them actually does :-)</li>
<p><br>Available <font color="#FF6600">instances of eoMergeReduce</font>
replacement include
<br>&nbsp;
<p>&nbsp;
<p>Available <font color="#FF6600">instances of eoMergeReduce</font> replacement
include
<ul>
<li>
<b><tt><font color="#009900">eoCommaReplacement</font></tt></b>, one of
the two standard strategies in <font color="#000000">Evolution Strategies</font>,
<b><font color="#FF6600">selects the best offspring</font></b>. It is an
<b><tt><font color="#993300">eoMergeReduce(eoNoElitism, eoTruncate)</font></tt></b>.</li>
<b><font color="#FF6600">selects
the best offspring</font></b>. It is an
<b><tt><font color="#993300">eoMergeReduce(eoNoElitism,
eoTruncate)</font></tt></b>.</li>
<li>
<b><tt><font color="#009900">eoPlusReplacement</font></tt></b>, the other
@ -112,9 +261,8 @@ eoTruncate)</font></tt></b>.</li>
<li>
<b><tt><font color="#009900">eoEPReplacement</font></tt></b>, used in th<font color="#000000">e
Evolutionary
Programming historical </font>algorithm, does an EP stochastic tournament
among parents + offspring. It is an <b><tt><font color="#993300">eoMergeReduce(eoPlus,
Evolutionary Programming historical </font>algorithm, does an EP stochastic
tournament among parents + offspring. It is an <b><tt><font color="#993300">eoMergeReduce(eoPlus,
eoEPReduce)</font></tt></b> and its constructor requires as argument T,
the size of the tournament (unsigned int).</li>
</ul>
@ -152,7 +300,9 @@ Additional parameter (in the constructor) is the tournament rate, a <b><tt><font
<li>
<b><tt><font color="#009900">eoSurviveAndDie</font></tt></b> is</li>
</ul>
<b><font color="#000099">Replacement: </font><font color="#FF0000">Adding
<hr WIDTH="50%">
<br><b><font color="#000099">Replacement: </font><font color="#FF0000">Adding
(weak) elitism</font></b>
<p>You can add what is called <font color="#FF6600">weak elitism</font>
to any replacement by encapsulating it into an <b><tt><font color="#009900">eoWeakElitismReplacement</font></tt></b>
@ -170,11 +320,13 @@ can be any replacement object):
<br>and use now replace as your replacement procedure within your algorithm.
<p><font color="#FF0000">Note</font>: of course, adding weak elitism to
an elitist replacement makes no sense - but will not harm either :-)
<p><b><font color="#000099">Replacement: </font><font color="#FF0000">Test
<br>
<hr WIDTH="50%">
<br><b><font color="#000099">Replacement: </font><font color="#FF0000">Test
file</font></b>
<p>The file <b><tt><font color="#993300">t-eoReplacement</font></tt></b>
in the <b><font color="#FF6600">test directory</font></b> implements all
above replacmenet procedures within a very simple and easy-to-monitor Dummy
above replacement procedures within a very simple and easy-to-monitor Dummy
EO class.
<p>
<hr WIDTH="100%">
@ -239,10 +391,10 @@ parameter R should be in [0.5,1]. It is implemented in the <b><tt><font color="#
class, a sub-class of eoSelectOne, as well as in the <b><tt><font color="#009900">eoStochTournamentTruncate</font></tt></b>
class that repeatidly removes from the population the "winner" of the inverse
tournament.&nbsp; These objects use the C++ function determinitic_tournament
in&nbsp; <a href="../../doc/html/selectors_h-source.html">selectors.h</a>.<br>
<b><font color="#FF0000">Note</font></b>: A stochastic tournament with
rate 1.0 is strictly identical to a deterministic tournament of size 2.</li>
in&nbsp; <a href="../../doc/html/selectors_h-source.html">selectors.h</a>.</li>
<br><b><font color="#FF0000">Note</font></b>: A stochastic tournament with
rate 1.0 is strictly identical to a deterministic tournament of size 2.
<li>
<a NAME="EPTournament"></a><b><tt><font color="#009900">EP Tournament</font></tt></b>
of size T is a global tournament: it works by assigning a score to all
@ -252,12 +404,13 @@ Everytime I wins, its score in incremented by 1 (and by 0.5 for every draw).
The individuals are then selected deterministically based on their scores
from that procedure. The <b><tt><font color="#009900">EP Tournament</font></tt></b>
is implemented in the&nbsp; <b><tt><font color="#009900">eoEPReduce</font></tt></b>
truncation method used in some replacement procedures.&nbsp;<br>
<b><font color="#FF0000">Note</font></b>: whereas both the determinitic
truncation method used in the <b><tt><font color="#009900">eoEPReplacement</font></tt></b>
procedure.</li>
<br><b><font color="#FF0000">Note</font></b>: whereas both the determinitic
and the stochastic tournament select one individual, the EP tournament
is designed for batch selection. Of course it could be used to select a
single individual, but at a rather high computational cost.</li>
</ul>
single individual, but at a rather high computational cost.</ul>
<p><br>
<hr WIDTH="100%">
@ -266,7 +419,9 @@ populations</font></font></b>
<p>In replacement procedures, one frequently needs to merge two populations
(computed form old parents and new-born offspring). Classes derived from
the abstract class eoMerge are written for that purpose.
<p><b><font color="#000099">eoMerge</font></b>: <b><font color="#FF0000">interface</font></b>
<br>
<hr WIDTH="50%">
<br><b><font color="#000099">eoMerge</font></b>: <b><font color="#FF0000">interface</font></b>
<br>The abstract class for merging procedures is the functor class
<font color="#009900">eoMerge</font>,
and the interface for its <tt><font color="#993300">operator()</font></tt>
@ -282,7 +437,9 @@ void></font></tt>.
<br>This means that it takes <font color="#FF6600">2 populations</font>
and modifies the seond one by adding some individuals from the first one
(which is supposed to remain <b><tt><font color="#993300">const</font></tt></b>ant).
<p><b><font color="#000099">eoMerge</font></b>: <b><font color="#FF0000">instances</font></b>
<br>
<hr WIDTH="50%">
<br><b><font color="#000099">eoMerge</font></b>: <b><font color="#FF0000">instances</font></b>
<br>Available <font color="#FF6600">instances of eoMerge</font> objects
are <b><tt><font color="#009900">eoPlus</font></tt></b>, that simply adds
the parents to the offspring, or <b><tt><font color="#009900">eoElitism</font></tt></b>,
@ -296,7 +453,9 @@ populations</font></font></b>
<p>The other useful component of replacement procedures, <font color="#009900">eoReduce</font>,
<font color="#FF6600">kills
some individuals</font> from a given population.
<p><b><font color="#000099">eoReduce</font></b>: <b><font color="#FF0000">interface</font></b>
<br>
<hr WIDTH="50%">
<br><b><font color="#000099">eoReduce</font></b>: <b><font color="#FF0000">interface</font></b>
<br>The abstract class for reducing procedures is the functor class
<font color="#009900">eoReduce</font>,
and the interface for its <tt><font color="#993300">operator()</font></tt>
@ -311,7 +470,9 @@ as you see there that <font color="#009900">eoReduce</font> derives from
int, void></font></tt>.
<br>An <font color="#009900">eoReduce</font> shoud take a<font color="#FF6600">
population</font> and shrink it to the required size.
<p><b><font color="#000099">eoReduce</font></b>: <b><font color="#FF0000">instances</font></b>
<br>
<hr WIDTH="50%">
<br><b><font color="#000099">eoReduce</font></b>: <b><font color="#FF0000">instances</font></b>
<br>Available <font color="#FF6600">instances of eoReduce</font> are
<ul>
<li>
@ -347,7 +508,6 @@ the population. It requires the rate of the tournament (<b><tt><font color="#993
as parameter in the constructor (default is 0.75).</li>
</ul>
<p><br>
<hr WIDTH="100%">
<br><a NAME="howmany"></a><b><font color="#000099"><font size=+2>eoHowMany:
Choosing a number of individuals</font></font></b>
@ -366,7 +526,9 @@ It receives a <b><tt><font color="#993300">double</font></tt></b>, and
a <b><tt><font color="#993300">boolean</font></tt></b> indicating whether
that double is to be treated <b><font color="#FF6600">as a rate</font></b>
or as <b><font color="#FF6600">an absolute (unisgned) interger.</font></b>
<p><b><font color="#000099">eoHowMany</font></b>: <b><font color="#FF0000">interface</font></b>
<br>
<hr WIDTH="50%">
<br><b><font color="#000099">eoHowMany</font></b>: <b><font color="#FF0000">interface</font></b>
<br>The class interface for its <tt><font color="#993300">operator()</font></tt>
is
<center>