bitwise_weight_value BitwiseWeight::guided_mutation( bitwise_weight_value elite_value, bitwise_weight_value new_value, RandomNumberGenerator& rng )
{
int i;
bitwise_weight_value weight;
weight.integer = 0;
int elite = elite_value.integer, newv = new_value.integer;
bool is_neg;
if( rng.getRandom() > config->gm_prob_of_mutation )
is_neg = elite < 0;
else
is_neg = newv < 0;
if( elite < 0 ) elite *= -1;
if( newv < 0 ) newv *= -1;
int num = 0x1;
for( i = 0; i < number_of_elements - 1; i++ )
{
bool gm = rng.getRandom() < config->gm_prob_of_mutation;
if( ( gm && elite % 2 )
|| (!gm && newv % 2 ) )
weight.integer += num;
num <<= 1;
elite >>= 1;
newv >>= 1;
}
if( is_neg )
weight.integer *= -1;
weight.decimal = get_decimal_value( weight.integer );
return weight;
}
void Experiment::run( DataCollector& dc, RandomNumberGenerator& rng )
{
// INITIALIZE
double prob_learning_type; // ELITE or BEST OF GENERATION
int i;
for( int run = 0; run < config->number_of_runs; run++ )
{
std::cout << "" << (run+1) << ",";
std::cout.flush();
dc.start_new_run( run );
prob_learning_type = 1.;
dc.run << "Run seed = " << rng.getCurrentSeed() << endline;
//INITIALIZE THE MODEL
Model model;
model.set_data_collector( dc );
// RUN FOR N GENERATIONS
for( i = 0; i < config->number_of_generations; i++ )
{
//std::cout << (run+1) << "." << i << "," << std::flush;
// ELITIST LEARNING
if( config->structural_learning == PIPE
&& prob_learning_type < config->pipe_prob_elitist_learn )
{
model.is_elitist_learning = true;
model.adapt( model.get_elite() );
}
else
{
// BEST OF LEARNING
// GENERATION BASED LEARNING
model.sample( rng );
if( config->fitness_test == RETINA_SWITCHING
&& ( i % config->retina_switch_after ) == 0
&& i != 0 )
retina_test_switch();
// MEASURE FITNESS + INDEX BEST and ELITE
model.measure_fitness( dc.run );
// ADAPT PPC TOWARD BEST
model.adapt( model.get_best() );
}
// MUTATE
model.mutate( rng );
prob_learning_type = rng.getRandom();
dc.log_generation( run, i, model );
dc.flush_debug();
if( ( model.stop_condition_met() && config->stop_on_target_reached )
|| ( i + 1 ) == config->number_of_generations )
{
dc.log_run( run, i, model );
break;
}
}
// RANDOM FITNESS TEST (to confirm findings)
dc.log_random_fitness( model );
}
dc.log_experiment();
}
