[galib] Constant Statistics over Generations

Steven Bilton swb04 at doc.ic.ac.uk
Thu Jan 31 10:39:12 EST 2008


I've been running a GASteadyStateGA using a GARealGenome with 5 alleles 
and a custom fitness function. After any number of generations ( 5, 10, 
50 ), the statistics for the initial and current population are 
identical - which seems very odd, as though the population isn't 
changing in each generation. For what sorts of reasons would this occur, 
and does anybody have any advice for solving it? Thank in advance.

Steve

----------

The params are set up as follows:

GAParameterList params;
GASteadyStateGA::registerDefaultParameters(params);
params.set(gaNnGenerations, 5);
params.set(gaNpopulationSize, 50);
params.set(gaNflushFrequency, 1);

GASteadyStateGA ga(genome);
ga.parameters(params);   

----------

The full statistics log is included below:

20    # current generation
0    # current convergence
130    # number of selections since initialization
119    # number of crossovers since initialization
4    # number of mutations since initialization
125    # number of replacements since initialization
170    # number of genome evaluations since initialization
6    # number of population evaluations since initialization
102.76    # maximum score since initialization
84.8377    # minimum score since initialization
91.4827    # average of all scores ('on-line' performance)
102.76    # average of maximum scores ('off-line' performance)
89.84    # average of minimum scores ('off-line' performance)

89.7384    # mean score in initial population
102.76    # maximum score in initial population
84.8377    # minimum score in initial population
2.74352    # standard deviation of initial population
-1    # diversity of initial population (0=identical,-1=unset)

89.7384    # mean score in current population
102.76    # maximum score in current population
84.8377    # minimum score in current population
2.74352    # standard deviation of current population
-1    # diversity of current population (0=identical,-1=unset)

20    # how far back to look for convergence
100    # how often to record scores
1    # how often to write scores to file



More information about the galib mailing list