AbstractIAFController::AbstractIAFController(const AbstractIAFControllerConf& conf)
: AbstractController("AbstractIAFController", "$Id$"), conf(conf), range(1.0) {
addParameter("leaki", conf.leakI);
addParameter("leako", conf.leakO);
addParameter("thresholdi", conf.thresholdI);
addParameter("thresholdo", conf.thresholdO);
addParameter("wiinitscale", conf.wIInitScale);
addParameter("wiinitscale", conf.wOInitScale);
addParameter("niafperinput", conf.numberIAFNeuronsPerInput);
addParameter("niafperoutput", conf.numberIAFNeuronsPerOutput);
// addInspectableValue("[I]leak", conf.leakI);
// addInspectableValue("[O]leak", conf.leakO);
// addInspectableValue("[I]t",conf.thresholdI);
// addInspectableValue("[O]t",conf.thresholdO);
addInspectableMatrix("[I]sum",&sumI);
addInspectableMatrix("[O]sum",&sumO);
// addInspectableMatrix("[I]W",&wI);
// addInspectableMatrix("[O]W",&wO);
addInspectableMatrix("[I]x",&xI);
addInspectableMatrix("[O]x",&xO);
initialised = false;
}
CuriosityLoop::CuriosityLoop(int inputSize, int motornumber, int sensornumber, int trial_length){
cout << "Making curiosity loop of input size " << inputSize << "\n";
predictorWeights.set(inputSize+1,inputSize+1); // initialized 0
predictorWeights=predictorWeights.map(random_minusone_to_one)*1.0;
// predictorMask.set(inputSize+1,inputSize+1);
//
// for(int i = 0; i < predictorMask.getM(); i++){
// for(int j = 0; j < predictorMask.getN(); j++){
// if(rand()/(RAND_MAX*1.0) < 0.01)
// predictorMask.val(i,j) = 1;
// else
// predictorMask.val(i,j) = 0;
// }
// }
prediction.set(inputSize+1, 1);
pInput.set(inputSize+1,1);
pOutput.set(inputSize+1,1);
// Standard predictors
/*
for(int i = 0; i < inputSize+1; i++){
if(rand()/(RAND_MAX*1.0) < 0.2)
pInput.val(i,0) = 1;
else
pInput.val(i,0) = 0;
if(rand()/(RAND_MAX*1.0) < 0.2)
pOutput.val(i,0) = 1;
else
pOutput.val(i,0) = 0;
};
*/
//For granger predictors the sensory input should be the last time step of the the sensory output and only sensor states serve as input and output.
for(int i = 0; i < inputSize+1; i++){
pInput.val(i,0) = 0;
pOutput.val(i,0) = 0;
if(i < sensornumber+1) {
//if( i == 1){
//Handdesigned joint angle to predict.
// pInput.val(i,0) = 1;
// pOutput.val(i,0) = 1;
if(rand()/(RAND_MAX*1.0) < 0.3){
pInput.val(i,0) = 1;
pOutput.val(i,0) = 1;
}
else{
pInput.val(i,0) = 0;
pOutput.val(i,0) = 0;
}
}
}
uPredictorWeights = predictorWeights;
//uPredictorMask = predictorMask;
uPrediction = prediction;
uPInput = pInput;
uPOutput = pOutput;
//Unrestrict upInput now to it gets input from all motors (of the actor) and the sensors of the restricted model.
for(int i = 0; i < inputSize+1; i++){
if(pInput.val(i,0) == 1)
uPInput.val(i,0) = 1; //i.e. gets input from all motors.
if(i > sensornumber + 1)
uPInput.val(i,0) = 1; //For now, the actor influences all motors.
}
actorWeights.set(motornumber,sensornumber); // initialized 0
actorWeights=actorWeights.map(random_minusone_to_one)*0.01;
offspringActorWeights.set(motornumber,sensornumber);
offspringActorWeights = actorWeights;
//Each loop has a fitness
fitness = 0;
prediction_error = 0;
prediction_error_time_average = 0;
old_prediction_error_time_average = 0;
addInspectableMatrix("PredInLoop",&prediction,false);
//addInspectableMatrix("PredError",&parent_error,false);
//addInspectableMatrix("UPredError",&offspring_error,false);
parent_error.set(trial_length,1);
offspring_error.set(trial_length,1);
};
