Main Page   Class Hierarchy   Compound List   File List   Compound Members   File Members  

MillerEncoding Class Reference

This is a rather generic version of the direct encoding proposed by Miller, Todd and Hedge (1989). More...

#include <miller.h>

Inheritance diagram for MillerEncoding:

ANNEncoding List of all members.

Public Methods

 MillerEncoding (const GeneticID &name, const StringMap &params)
virtual void copy (const Genstruct &other)
virtual bool execute (const GeneticMsg &msg) const
virtual void addPrivateGenes (Gentainer &g, const StringMap &params)
virtual void init ()

Detailed Description

This is a rather generic version of the direct encoding proposed by Miller, Todd and Hedge (1989).

It works on freely feed-forward topologies.

The basic idea in direct encoding is to have one binary gene for each potential connection in the neural network that tells whether or not that connection actually exists.

In most implementations the connection genes are initialized (in the initial generation) with a certain probability of connection p(c). Often p(c) is simply 0.5, but can vary.

Our implementation of the encoding method is somewhat different from the original method by Miller et al.

Most important modification in our implementation is the initialization of the genomes in the first generation. Instead of using a fixed probability of connection p(c), we initialize the different genomes with different p(c). The p(c) of a certain genome is determined from a normally distributed random value, controlled by the pcVariance and pcAverage parameters (see below).

Definition at line 62 of file miller.h.

Constructor & Destructor Documentation

MillerEncoding (  const GeneticID &    name,
const StringMap &    params
) 
 

Standard constructor.

Parameters:
params  ["pruneInputs"] Should the input neurons be encoded in the genome also. [Default=1]
params  ["pcVariance"] Variance distribution of the connection probability of the initial neural networks. [Default=0]
params  ["pcAverage"] Average connection probability for initial neural networks. [Default=0.5]
name  Gene name, usually "brainplan".

Definition at line 46 of file miller.cc.

Member Function Documentation

void addPrivateGenes (  Gentainer &    g,
const StringMap &    params
)  [virtual]
 

Implementation for Genstruct.

Reimplemented from ANNEncoding.

Definition at line 66 of file miller.cc.

void copy (  const Genstruct &    o )  [virtual]
 

Implementation for Genstruct.

Reimplemented from ANNEncoding.

Definition at line 58 of file miller.cc.

References ANNEncoding::copy(), mPcAverage, mPcVariance, and mPruneInputs.

bool execute (  const GeneticMsg &    msg )  const [virtual]
 

Implementation for Genstruct.

Definition at line 100 of file miller.cc.

void init (    )  [virtual]
 

Implementation for Genstruct.

Definition at line 84 of file miller.cc.

The documentation for this class was generated from the following files:
Generated on Thu Feb 10 20:21:26 2005 for Annalee by doxygen1.2.18