/// <summary>
/// Constructs the behavior vector at the end of an individual evaluation.
/// </summary>
List<double> IBehaviorCharacterization.calculate(SimulatorExperiment exp, instance_pack ip)
{
// initialize the BC vector
BehaviorVector = new List<double>();
// If the robot never Stopped, set the max evaluation time as the end tick
if (endTick == 0)
endTick = exp.evaluationTime;
// Adjust end tick by the fraction we are sampling
endTick = (int)(endTick * 0.5);
// Calculate when to perform an update
int numBehaviorChunks = VectorLength / 2;
chunkSize = Convert.ToInt32(Math.Floor((double)endTick / (double)numBehaviorChunks)) * 2;
float x, y;
for (int chunkNum = 1; chunkNum < numBehaviorChunks + 1; chunkNum++)
{
// Take bc samples from the internal Trajectory store
x = Trajectory[chunkNum * chunkSize - 2];
x = (x - ip.env.AOIRectangle.Left) / ip.env.AOIRectangle.Width;
BehaviorVector.Add(x);
y = Trajectory[chunkNum * chunkSize - 1];
y = (y - ip.env.AOIRectangle.Top) / ip.env.AOIRectangle.Height;
BehaviorVector.Add(y);
}
return BehaviorVector;
}
/// <summary>
/// Constructs the behavior vector at the end of an individual evaluation.
/// </summary>
List <double> IBehaviorCharacterization.calculate(SimulatorExperiment exp, instance_pack ip)
{
// initialize the BC vector
BehaviorVector = new List <double>();
// If the robot never Stopped, set the max evaluation time as the end tick
if (endTick == 0)
{
endTick = exp.evaluationTime;
}
// Adjust end tick by the fraction we are sampling
endTick = (int)(endTick * 0.5);
// Calculate when to perform an update
int numBehaviorChunks = VectorLength / 2;
chunkSize = Convert.ToInt32(Math.Floor((double)endTick / (double)numBehaviorChunks)) * 2;
float x, y;
for (int chunkNum = 1; chunkNum < numBehaviorChunks + 1; chunkNum++)
{
// Take bc samples from the internal Trajectory store
x = Trajectory[chunkNum * chunkSize - 2];
x = (x - ip.env.AOIRectangle.Left) / ip.env.AOIRectangle.Width;
BehaviorVector.Add(x);
y = Trajectory[chunkNum * chunkSize - 1];
y = (y - ip.env.AOIRectangle.Top) / ip.env.AOIRectangle.Height;
BehaviorVector.Add(y);
}
return(BehaviorVector);
}
/// <summary>
/// Records the individual's location at each tick of the simulation.
/// </summary>
void IBehaviorCharacterization.update(SimulatorExperiment exp, instance_pack ip)
{
if (ip.robots[0].Stopped)
{
// If this is the first update after the robot has Stopped,
// send the endpoint to be the current simulation tick
if (endTick == 0)
{
endTick = Convert.ToInt32(ip.timeSteps * exp.timestep);
}
}
// initialize the Trajectory list
if (!Initialized)
{
// initialize the sensor value sampling/storage components
Trajectory = new List <int>(exp.evaluationTime * 2);
Initialized = true;
}
// update the Trajectory at every tick
Trajectory.Add(Convert.ToInt16(ip.robots[0].Location.X));
Trajectory.Add(Convert.ToInt16(ip.robots[0].Location.Y));
}
/// <summary>
/// Records the individual's location at each tick of the simulation.
/// </summary>
void IBehaviorCharacterization.update(SimulatorExperiment exp, instance_pack ip)
{
if (ip.robots[0].Stopped)
{
// If this is the first update after the robot has Stopped,
// send the endpoint to be the current simulation tick
if (endTick == 0)
endTick = Convert.ToInt32(ip.timeSteps * exp.timestep);
}
// initialize the Trajectory list
if (!Initialized)
{
// initialize the sensor value sampling/storage components
Trajectory = new List<int>(exp.evaluationTime * 2);
Initialized = true;
}
// update the Trajectory at every tick
Trajectory.Add(Convert.ToInt16(ip.robots[0].Location.X));
Trajectory.Add(Convert.ToInt16(ip.robots[0].Location.Y));
}
public static void Main(string[] args)
{
string folder = "";
string environment_name = null; // "hallway.xml";
string substrate_name = null; // "substrate.xml";
string filename = null; // "seedGenome2x.xml";
string to_eval = null;
bool homogenous = true; // = true;
bool overrideHomogenousSetting = false;
int generations = 1000, populationSize = 0; //was 500
bool novelty = false;
bool eval = false;
bool evolveSubstrate = false;
bool leo_setting = false;
bool overrideLeo = false;
bool overrideInputDensity = false;
bool overrideHiddenDensity = false;
bool overrideFitnessFunction = false;
bool overrideBehaviorCharacterization = false;
bool seed_population = false;
bool overrideAgentCount = false;
bool multiobjective = false;
int input_density_override = -1;
int hidden_density_override = -1;
int agent_count_override = -1;
string fitness_function_override = "";
string behavior_override = "";
string experimentName = null;
bool benchmark = false;
bool overrideES = false;
if (args.Length != 0 && args[0] == "-help")
{
showHelp();
return;
}
//evolve -experiment multiagent_exp.xml -seed seedGenome2x.xml
if (!(args.Length == 0) && args[0] == "evolve")
{
for (int j = 1; j < args.Length; j++)
{
if (j <= args.Length - 2)
{
switch (args[j])
{
case "-benchmark":
benchmark = true;
break;
case "-multiobjective":
multiobjective = true;
break;
case "-es":
evolveSubstrate = Convert.ToBoolean(args[++j]);
overrideES = true;
Console.WriteLine("Evolvable-substrate: " + evolveSubstrate);
break;
case "-leo":
leo_setting = Convert.ToBoolean(args[++j]);
overrideLeo = true;
break;
case "-experiment":
experimentName = args[++j];
break;
case "-homogenous":
homogenous = Convert.ToBoolean(args[++j]);
overrideHomogenousSetting = true;
break;
case "-populationSize": if (!int.TryParse(args[++j], out populationSize))
{
Console.WriteLine("Invalid number of runs specified.");
}
break;
case "-generations": if (!int.TryParse(args[++j], out generations))
{
Console.WriteLine("Invalid number of generations specified.");
}
break;
case "-agent_count":
overrideAgentCount = true;
agent_count_override = Convert.ToInt32(args[++j]);
break;
case "-novelty":
novelty = true;
Console.WriteLine("Novelty search enabled...");
//TODO j++;
break;
case "-fitness_function":
overrideFitnessFunction = true;
fitness_function_override = args[++j];
Console.WriteLine("Setting fitness function to " + fitness_function_override);
break;
case "-behavior_characterization":
overrideBehaviorCharacterization = true;
behavior_override = args[++j];
Console.WriteLine("Setting behavior characterization to " + behavior_override);
break;
case "-hidden_density":
overrideHiddenDensity = true;
hidden_density_override = Convert.ToInt32(args[++j]);
Console.WriteLine("Setting hidden density to " + hidden_density_override);
break;
case "-input_density":
overrideInputDensity = true;
input_density_override = Convert.ToInt32(args[++j]);
Console.WriteLine("Setting rangefinder input density to " + input_density_override);
break;
case "-rng_seed": int seed = Convert.ToInt32(args[++j]);
Utilities.random.Reinitialise(seed);
SharpNeatLib.Maths.RouletteWheel.random = new Random(seed);
Console.WriteLine("Using RNG seed " + seed);
break;
case "-environment": environment_name = args[++j];
Console.WriteLine("Using environment " + environment_name);
break;
case "-substrate": substrate_name = args[++j];
Console.WriteLine("Using substrate " + substrate_name);
break;
case "-eval": to_eval = args[++j];
eval = true;
Console.WriteLine("Attempting to evaluate file " + to_eval);
break;
case "-seed": filename = args[++j];
Console.WriteLine("Attempting to use seed from file " + filename);
break;
case "-seed_pop": filename = args[++j];
seed_population = true;
Console.WriteLine("Attempting to use seed population from file " + filename);
break;
case "-folder": folder = args[++j];
Console.WriteLine("Attempting to output to folder " + folder);
break;
}
}
}
if (!homogenous && evolveSubstrate)
{
Console.WriteLine("Evolvalbe-Substrate not supported for heterogenous teams");
return;
}
if (experimentName == null)
{
Console.WriteLine("Missing [experimentName].");
Console.WriteLine("See help \"-help\"");
return;
}
ExperimentWrapper wr = ExperimentWrapper.load(experimentName);
SimulatorExperiment experiment = wr.experiment;
if (populationSize != 0)
{
experiment.populationSize = populationSize;
}
else
{
populationSize = experiment.populationSize;
}
if (overrideHomogenousSetting)
{
if (experiment is MultiAgentExperiment)
{
((MultiAgentExperiment)experiment).homogeneousTeam = homogenous;
}
}
if (environment_name != null)
{
experiment.environmentName = environment_name;
experiment.environmentList.Add(Engine.Environment.loadEnvironment(environment_name));
}
//change sensor density if requested
if (overrideInputDensity)
{
experiment.rangefinderSensorDensity = input_density_override;
}
if (overrideAgentCount)
{
if (experiment is MultiAgentExperiment)
{
((MultiAgentExperiment)experiment).numberRobots = agent_count_override;
}
}
if (substrate_name != null)
{
experiment.substrateDescription = new SubstrateDescription(substrate_name);
}
// se.substrateDescription = substrate;
if (filename != null)
{
if (!seed_population)
{
experiment.loadGenome(filename);
}
}
if (eval)
{
experiment.eval = true;
Console.WriteLine("Time to evaluate");
experiment.loadGenome(to_eval);
experiment.initialize();
if (benchmark)
{
experiment.timesToRunEnvironments = 6; //was 25
MultiAgentExperiment exp = (MultiAgentExperiment)experiment;
exp.benchmark = true;
}
Engine.NetworkEvaluator x = new Engine.NetworkEvaluator(experiment);
// Schrum: Loop used for testing: See how consistent evals are
// for (int i = 0; i < 3; i++) {
SharpNeatLib.BehaviorType behavior;
SharpNeatLib.NeuralNetwork.INetwork network = experiment.genome.Decode(null);
Console.WriteLine("Links: " + network.NumLinks);
Console.WriteLine("Output Modules: " + network.NumOutputModules);
Console.WriteLine("Fitness score:" + x.EvaluateNetwork(network, out behavior));
// schrum2: Prevent this from crashing in domains with no behavior measure defined
if (behavior.behaviorList != null)
{
Console.Write("Behavior: ");
foreach (double d in behavior.behaviorList)
{
Console.Write(d + " ");
}
Console.WriteLine();
}
// Schrum: This will actually print out different objective scores.
// In the case of FourTasks, there is an objective for each environment.
if (behavior != null)
{
Console.Write("MO: ");
foreach (double d in behavior.objectives)
{
Console.Write(d + " ");
}
Console.WriteLine();
}
// } // Schrum: End of test loop commented out above.
return;
}
Console.WriteLine("Experiment initialize");
experiment.initialize();
Console.WriteLine("Experiment done initializing");
if (experiment.adaptableANN && evolveSubstrate)
{
Console.WriteLine("Right now ES-HyperNEAT does not support plastic ANNs. Disable adaptation or use the fixed-substrate HyperNEAT");
return;
}
if (overrideLeo)
{
experiment.substrateDescription.useLeo = leo_setting;
}
Console.WriteLine("Override fitness?");
//change fitness function if requested
if (overrideFitnessFunction)
{
experiment.setFitnessFunction(fitness_function_override);
}
if (overrideBehaviorCharacterization)
{
experiment.setBehavioralCharacterization(behavior_override);
}
if (overrideES)
{
experiment.evolveSubstrate = evolveSubstrate;
}
//make sure substrate's input density matches # of inputs in environment
uint number_of_sensors = Convert.ToUInt32(experiment.rangefinderSensorDensity);
uint dx = 0, dy = 0;
/*
* experiment.substrateDescription.getNeuronDensity(0, out dx, out dy);
*
* //TODO maybe updateInputDensity of AgentBrain needs to be called
*
* if (dx != number_of_sensors)
* {
* Console.WriteLine("Input density forced to match number of sensors.");
* Console.WriteLine("Input Density beforehand: " + dx.ToString() + " " + dy.ToString());
* experiment.substrateDescription.setNeuronDensity(0, number_of_sensors, 1);
* experiment.substrateDescription.getNeuronDensity(0, out dx, out dy);
* Console.WriteLine("Input Density afterwards: " + dx.ToString() + " " + dy.ToString());
* }
*/
if (overrideHiddenDensity)
{
experiment.substrateDescription.getNeuronDensity(1, out dx, out dy);
Console.WriteLine("Hidden Density beforehand: " + dx.ToString() + " " + dy.ToString());
experiment.substrateDescription.setNeuronDensity(0, Convert.ToUInt32(hidden_density_override), 1);
experiment.substrateDescription.getNeuronDensity(0, out dx, out dy);
Console.WriteLine("Hidden Density afterwards: " + dx.ToString() + " " + dy.ToString());
}
Console.WriteLine("New evolver");
HyperNEATEvolver evolve = new HyperNEATEvolver(experiment);
if (novelty)
{
evolve.enableNoveltySearch(true);
}
if (multiobjective)
{
evolve.experiment.DefaultNeatParameters.multiobjective = true;
}
Console.WriteLine("Set output: " + folder);
evolve.setOutputFolder(folder);
if (filename != null)
{
if (seed_population)
{
evolve.initializeEvolutionFromPopFile(filename);
}
else
{
evolve.initializeEvolution(populationSize, experiment.genome);
}
}
else
{
evolve.initializeEvolution(populationSize);
}
Console.WriteLine("About to evolve");
evolve.evolve(generations);
/*for (int i = 0; i < generations; i++)
* {
* evolve.oneGeneration(i);
* }*/
}
else
{
experimentName = "EXP_FourTasks-avg-1M.xml";
for (int j = 0; j < args.Length; j++)
{
if (j <= args.Length - 2)
{
switch (args[j])
{
case "-experiment":
experimentName = args[++j];
break;
case "-genome":
filename = args[++j];
break;
}
}
}
SimulatorVisualizer vis = new SimulatorVisualizer(experimentName, filename);
Application.Run(vis);
}
}
