Method newControlWires

src/WireFitQLearn.cpp:252–272

}

std::vector<Wire> WireFitQLearn::newControlWires(const Wire &correctWire, std::vector<Wire> controlWires) {
	double error = 0;
	int iterations = 0;

	do {
		for(unsigned int a = 0; a < controlWires.size(); a++) {
			double deltaReward = -2 * (-interpolator->getReward(controlWires, correctWire.action) + correctWire.reward)*interpolator->rewardDerivative(correctWire.action, controlWires[a], controlWires);
			controlWires[a].reward = controlWires[a].reward - controlPointsGDLearningRate*deltaReward;
			for(unsigned int b = 0; b < controlWires[a].action.size(); b++) {
				double deltaActionTerm = -2 * (-interpolator->getReward(controlWires, correctWire.action) + correctWire.reward)*interpolator->actionTermDerivative(correctWire.action[b], controlWires[a].action[b], correctWire.action, controlWires[a], controlWires);
				controlWires[a].action[b] = controlWires[a].action[b] - controlPointsGDLearningRate*deltaActionTerm;
			}
		}

		error = pow(correctWire.reward - interpolator->getReward(controlWires, correctWire.action), 2);
		iterations++;
	} while(error > controlPointsGDErrorTarget && iterations < controlPointsGDMaxIterations);

	return controlWires;

}

std::vector<Wire> WireFitQLearn::newControlWires(const std::vector<Wire> &correctWires, std::vector<Wire> controlWires) {
	double error = 0;