neat.h

#ifndef _NERO_NEAT_H_
#define _NERO_NEAT_H_

#include <cstdlib>

namespace NEAT {

	extern int time_alive_minimum; // Minimum time alive to be considered for selection or death in real-time evolution 
	const int num_trait_params = 8;

	extern double trait_param_mut_prob;
	extern double trait_mutation_power; // Power of mutation on a signle trait param 
	extern double linktrait_mut_sig;  // Amount that mutation_num changes for a trait change inside a link
	extern double nodetrait_mut_sig; // Amount a mutation_num changes on a link connecting a node that changed its trait 
	extern double weight_mut_power;  // The power of a linkweight mutation 
	extern double recur_prob;        // Prob. that a link mutation which doesn't have to be recurrent will be made recurrent 

	// These 3 global coefficients are used to determine the formula for
	// computating the compatibility between 2 genomes.  The formula is:
	// disjoint_coeff*pdg+excess_coeff*peg+mutdiff_coeff*mdmg.
	// See the compatibility method in the Genome class for more info
	// They can be thought of as the importance of disjoint Genes,
	// excess Genes, and parametric difference between Genes of the
	// same function, respectively. 
	extern double disjoint_coeff;
	extern double excess_coeff;
	extern double mutdiff_coeff;

	// This global tells compatibility threshold under which two Genomes are considered the same species 
	extern double compat_threshold;

	// Globals involved in the epoch cycle - mating, reproduction, etc.. 
	extern double age_significance;          // How much does age matter? 
	extern double survival_thresh;           // Percent of ave fitness for survival 
	extern double mutate_only_prob;          // Prob. of a non-mating reproduction 
	extern double mutate_random_trait_prob;
	extern double mutate_link_trait_prob;
	extern double mutate_node_trait_prob;
	extern double mutate_link_weights_prob;
	extern double mutate_toggle_enable_prob;
	extern double mutate_gene_reenable_prob;
	extern double mutate_add_node_prob;
	extern double mutate_add_link_prob;
	extern double interspecies_mate_rate;    // Prob. of a mate being outside species 
	extern double mate_multipoint_prob;     
	extern double mate_multipoint_avg_prob;
	extern double mate_singlepoint_prob;
	extern double mate_only_prob;            // Prob. of mating without mutation 
	extern double recur_only_prob;  // Probability of forcing selection of ONLY links that are naturally recurrent 
	extern int pop_size;  // Size of population 
	extern int dropoff_age;  // Age where Species starts to be penalized 
	extern int newlink_tries;  // Number of tries mutate_add_link will attempt to find an open link 
	extern int print_every; // Tells to print population to file every n generations 
	extern int babies_stolen; // The number of babies to siphon off to the champions 

	extern int num_runs; //number of times to run experiment

	//extern MRandomR250 NEATRandGen; // Random number generator; can pass seed value as argument

	//const char *getUnit(const char *string, int index, const char *set);
	//const char *getUnits(const char *string, int startIndex, int endIndex, const char *set);
	int getUnitCount(const char *string, const char *set);

	// Inline Random Functions 
	extern inline int randposneg() {
        if (rand()%2) 
            return 1; 
        else 
            return -1;
    }
    
	extern inline int randint(int x,int y) {
        return rand()%(y-x+1)+x;
    }

    extern inline double randfloat() {
        return rand() / (double) RAND_MAX;        
    }


	// SIGMOID FUNCTION ********************************
	// This is a signmoidal activation function, which is an S-shaped squashing function
	// It smoothly limits the amplitude of the output of a neuron to between 0 and 1
	// It is a helper to the neural-activation function get_active_out
	// It is made inline so it can execute quickly since it is at every non-sensor 
	// node in a network.
	// NOTE:  In order to make node insertion in the middle of a link possible,
	// the signmoid can be shifted to the right and more steeply sloped:
	// slope=4.924273
	// constant= 2.4621365
	// These parameters optimize mean squared error between the old output,
	// and an output of a node inserted in the middle of a link between
	// the old output and some other node. 
	// When not right-shifted, the steepened slope is closest to a linear
	// ascent as possible between -0.5 and 0.5
	extern double fsigmoid(double,double,double);

	// Hebbian Adaptation Function
	// Based on equations in Floreano & Urzelai 2000
	// Takes the current weight, the maximum weight in the containing network,
	// the activation coming in and out of the synapse,
	// and three learning rates for hebbian, presynaptic, and postsynaptic
	// modification
	// Returns the new modified weight
	// NOTE: For an inhibatory connection, it makes sense to
	//      emphasize decorrelation on hebbian learning!
	extern double oldhebbian(double weight, double maxweight, double active_in, double active_out, double hebb_rate, double pre_rate, double post_rate);

	// Hebbian Adaptation Function
	// Based on equations in Floreano & Urzelai 2000
	// Takes the current weight, the maximum weight in the containing network,
	// the activation coming in and out of the synapse,
	// and three learning rates for hebbian, presynaptic, and postsynaptic
	// modification
	// Returns the new modified weight
	// NOTE: For an inhibatory connection, it makes sense to
	//      emphasize decorrelation on hebbian learning!	
	extern double hebbian(double weight, double maxweight, double active_in, double active_out, double hebb_rate, double pre_rate, double post_rate);

	// Returns a normally distributed deviate with 0 mean and unit variance
	// Algorithm is from Numerical Recipes in C, Second Edition
	extern double gaussrand();

	//This is an incorrect gassian distribution...but it is faster than gaussrand (maybe it's good enough?)
	//inline double gaussrand_wrong() {return (randposneg())*(sqrt(-log((rand()*1.0)/RAND_MAX)));}   

	bool load_neat_params(const char *filename, bool output = false);

} // namespace NEAT

#endif