#ifndef _HPP_LEARN
#define _HPP_LEARN

#include "simelement.hpp"
#include "connection.hpp"
#include "libcsim.hpp"
#include "layer.hpp"


class learning : public SimElement
{
protected:
     layer  *TargetLayer,  *SourceLayer;
     int ns, nt;  // target and source layer dimensions
     int maxN_pre;
     int M;  // number of postsynaptic connections per presynaptic neuron
     connection *con;
     float maxWeight, minWeight;
     int Dmax;
     int maximumDelay;
     short **delays_length;
     short ***delays;
     int **post;
     int **I_pre, **D_pre;
     int *N_pre;
     int **m_pre;
     pfloat **s_pre, **sd_pre;
public:
     learning(connection*, float _maxWeight=1);
     ~learning();
     virtual int proceede(int);
     virtual int prepare(int =0);          
//     virtual void learning::SetMinWeight(float value);  
     virtual void SetMinWeight(float value);  // removed extra qualification learning::
     virtual void SetMaxWeight(float value);
     float **sd, **s;
};

class levylearning: public learning
{
protected:
     float **LTP1, **LTP2;
     float TauA, TauB, decA, decB;  // decay and raise time constant for presynaptic potential
     float LearnRate;
//     float DeLearnFac;
     float LtpInc;  
public:
     levylearning(connection*, float =1, float =0.1, float =150, float =1.785, float _LtpInc=1);
     ~levylearning();
     virtual int proceede(int);
     virtual int prepare(int =0);     
};

class izhlearning: public learning
{
protected:
     float **LTP;
     float *LTD;     
     float dec_pre, dec_post;   // decay of presynaptic and postsynaptic learning potential  
public:
     izhlearning(connection* con, float _maxWeight=1);
     ~izhlearning();
     virtual int proceede(int =0);
     virtual int prepare(int =0);     
};


class LearnRossum1: public learning
{
protected:
     float **SourceLearnPot;
     float *TargetLearnPot;     
     float dec_pre, dec_post;   // decay of presynaptic and postsynaptic learning potential  
     float Cp, Cd, SynNoiseSigma;
public:
     LearnRossum1(connection* con, float _Cp=0.1, float _Cd=0.003, float _SynNoiseSigma=0.015, float _maxWeight=1);
     ~LearnRossum1();
     virtual int proceede(int =0);
     virtual int prepare(int =0);     
};

class LearnAntiRossum1: public LearnRossum1
{
protected:
public:
     LearnAntiRossum1(connection* con, float _Cp=0.1, float _Cd=0.003, float _SynNoiseSigma=0.015, float _maxWeight=1);
     virtual int proceede(int =0);
};


class LearnHebbLP2: public learning
{
protected: 
     float **LTP;
     float LtpDecFac;
     float LtpInc;
     bool Accumulate;
     float LearnSpeed;
     float BaseLine;
public:
     LearnHebbLP2(connection* con, float _MaxWeight=1, float TauDec=20, float BaseLine=0.1, float _LearnSpeed=0.001, bool _Accumulate=true);
     ~LearnHebbLP2();
     virtual int proceede(int =0);
     virtual int prepare(int =0);     
     virtual int WriteSimInfo(fstream &fw);
};

class LearnHebbLP2_norm: public learning
{
protected: 
     float **LTP;
     float LtpDecFac;
     float LtpInc;
     float LearnSpeed;
     float BaseLine;
     bool Normalize;
     float WeightSum;
public:
     LearnHebbLP2_norm(connection* con, float _MaxWeight=1, float TauDec=20, float BaseLine=0.1, float _LearnSpeed=0.001, float _WeightSum=0);
     ~LearnHebbLP2_norm();
     virtual int proceede(int =0);
     virtual int prepare(int =0);     
     int NormalizeWeights();
};

class LearnPrePost: public learning
{
protected:
     float **LPpre; //presynaptic potentials
     float *LPpost;     // postsynaptic potentials
     float dec_pre, dec_post;   // decay of presynaptic and postsynaptic learning potential  
public:
     LearnPrePost(connection* con, float _maxWeight=1, float TauLearnPre=30, float TauLearnPost=30);
     ~LearnPrePost();
     virtual int proceede(int =0);
     virtual int prepare(int =0);     
};

class LearnFBInh: public LearnPrePost
{
protected:
     float Cpre, Cdep;
public:
     LearnFBInh(connection* con, float _maxWeight=1, float TauLearnPre=30, float TauLearnPost=30, float _Cpre=1, float _Cdep=0.03);
     virtual int proceede(int =0);
};


class LearnFWInh: public LearnPrePost
{
protected:
     float Cpre, Cdep;
public:
     LearnFWInh(connection* con, float _maxWeight=1, float TauLearnPre=30, float TauLearnPost=30, float _Cpre=1, float _Cdep=0.03);
     virtual int proceede(int =0);
};



// wie LearnHebbLP2, aber mit st�rkerer Gewichts�nderung bei hoher 
// postsynaptischer Rate (d.h. Multiplikation der lernrate mit dem Postsynaptischen Spike Trace)
// daher auch von LearnPrePost abgeleitet
class LearnHebbLP3: public LearnPrePost
{
protected:
     float LearnRate, BaseLine;
     float LtpInc;
     bool Accumulate;
public:
     LearnHebbLP3(connection* con, float _maxWeight=1, float TauLearnPre=30, float TauLearnPost=30, float _BaseLine=0.2, float _LearnRate=0.0001, bool _Accumulate=false);
     virtual int proceede(int =0);
};


class LearnAntiHebb: public LearnPrePost
{
protected:
     float Cpre, Cdep;
public:
     LearnAntiHebb(connection* con, float _maxWeight=1, float TauLearnPre=30, float TauLearnPost=30, float _Cpre=1, float _Cdep=0.03);
     virtual int proceede(int =0);
};


// Learning rule according to Bi&Poo 1998, implemented as in Shon, Rao, Sejnowski 2004
// using alpha functions for positive and negative part of learning window
// calculate alpha function in constructor and save in lookup table
// save spike times for every neuron
// every pre and postsynaptic spike is a learning event
class LearnBiPoo: public learning
{
protected:
     int** LastSpikesPre; // arrays for time of last spike at presynaptic site
     int* LastSpikesPost;  // array for time of last spike of postsynaptic neurons
     float* PreLearnWindowLut; // lookup table for learning window
     float* PostLearnWindowLut; // lookup table for learning window
     float LearnRate, TauPeak1, TauPeak2, Amp1, Amp2;
     int PreLutN, PostLutN;
public:
     LearnBiPoo(connection* con, float _maxWeight=1, float LearnRate=0.001, float TauPeak1=10, float TauPeak2=10, float Amp1=-1.25, float Amp2=1);     
     ~LearnBiPoo();
     virtual int proceede(int TotalTime);
     virtual int prepare(int step);
};


// Learning rule according to Froemke & Dan 2002
// similar to Bi&Poo but additionally for each spike a spike efficacy is calculated
// 0<=Eff<=1
// after each spike Eff is set to zero and then falls back to 1 exponentially
// TauPre=28ms
// TauPost=88ms
// ERROR: not correctly implemented
// because only neares neighbor interactions are used
// but Froemke/Dan did not limit the interactions to neares neighbors
// should be implemented using decaying potentials
class LearnFroemkeDan: public learning
{
protected:
     int** LastSpikesPre; // arrays for time of last spike at presynaptic site
     int* LastSpikesPost;  // array for time of last spike of postsynaptic neurons
     int** SecondLastSpikesPre; // time of second last spike at presynaptic site
     int* SecondLastSpikesPost;  // time of second  last spike of postsynaptic neurons
     float* PreLearnWindowLut; // lookup table for learning window
     float* PostLearnWindowLut; // lookup table for learning window
     int PreLutN, PostLutN;
     float LearnRate, TauPeak1, TauPeak2, Amp1, Amp2;

     float* PreEfficacyLut; // lookup table for spike efficacy
     float* PostEfficacyLut; // lookup table for spike efficacy
     float TauPreEff, TauPostEff;
     int PreEffLutN, PostEffLutN;
public:
     LearnFroemkeDan(connection* con, float _maxWeight=1, float LearnRate=0.001, 
		     float PreEffTau=28, float PostEffTau=88, 
		     float TauPeak1=35, float TauPeak2=15, float Amp1=-.5, float Amp2=1);     
     ~LearnFroemkeDan();
     virtual int proceede(int TotalTime);
     virtual int prepare(int step);
};




// Learning rule inspired by Sj�str�m, Turrigiano, Nelson 2001
// only nearest neighbor interactions as in LearnFroemkeDan and LearnBiPoo
// frequency dependence (??)
// LTP "winns" over LTD
// Dependence of LTP on initial synaptic strength (stronger synapse, smaller relative change 
// --> additive weight change as in LearnRossum1?? see Rossum, Bi, turrigano 2000)

class LearnSjoestroem: public learning
{
protected:
     int** LastSpikesPre; // arrays for time of last spike at presynaptic site
     int** LastSpikesPost;  // array for time of last spike of postsynaptic neurons

     float* LTPotLearnWindowLut; // lookup table for learning window
     float* LTDepLearnWindowLut; // lookup table for learning window
     int LTPotLutN, LTDepLutN;
     float LearnRate, TauLTDep, TauLTPot, Amp1, Amp2;

public:
     LearnSjoestroem(connection* con, float _maxWeight=1, float LearnRate=0.001, 
		     float TauLTDep=35, float TauLTPot=15, float AmpLTDep=-.5, float AmpLTPot=0.2);     
     ~LearnSjoestroem();
     virtual int proceede(int TotalTime);
     virtual int prepare(int step);
};




#endif /*_HPP_LEARN */