/***************************************************************************
* TableBasedModelHF.h *
* ------------------- *
* copyright : (C) 2014 by Francisco Naveros *
* email : fnaveros@ugr.es *
***************************************************************************/
/***************************************************************************
* *
* This program is free software; you can redistribute it and/or modify *
* it under the terms of the GNU General Public License as published by *
* the Free Software Foundation; either version 3 of the License, or *
* (at your option) any later version. *
* *
***************************************************************************/
#ifndef TABLEBASEDMODELHF_H_
#define TABLEBASEDMODELHF_H_
/*!
* \file TableBasedModelHF.h
*
* \author Francisco Naveros
* \date April 2014
*
* This file is a modification of the TableBasedModel. This file declares a class which
* implements a neuron model based in look-up tables. The main difference it is that when
* a input spike arrive to this model, the neuron state variables are update, but instead of
* make a predicction in that instant, an event of type TableBasedModelHFEvent is created
* with a delay of "ToleranceTime" (normaly fixed to 1us). All the spike that arrive inside
* this microsecond are computed conjointly and only a predicction is made.
* IMPORTANT: This method is better than the traditional TableBasedModel when this one
* receives input activity in a synchronize way, because only one prediction is made. Conversely,
* when the input activity does not arrive in a synchronize way, this method create an innecesary
* the overhead due to the creation of the TableBasedModelHFEvent.
*
*/
#include "EventDrivenNeuronModel.h"
#include "../spike/EDLUTFileException.h"
class NeuronModelTable;
class Interconnection;
class TableBasedModelHFEvent;
/*!
* \class TableBasedModel
*
* \brief Spiking neuron model based in look-up tables
*
* This class implements the behavior of a neuron in a spiking neural network.
* It includes internal model functions which define the behavior of the model
* (initialization, update of the state, synapses effect, next firing prediction...).
* This behavior is calculated based in precalculated look-up tables.
*
* \author Jesus Garrido
* \date February 2010
*/
class TableBasedModelHF: public EventDrivenNeuronModel {
protected:
/*!
* \brief Number of state variables (no include time).
*/
unsigned int NumStateVar;
/*!
* \brief Number of time dependent state variables.
*/
unsigned int NumTimeDependentStateVar;
/*!
* \brief Number of synaptic variables.
*/
unsigned int NumSynapticVar;
/*!
* \brief Index of synaptic variables.
*/
unsigned int * SynapticVar;
/*!
* \brief Order of state variables.
*/
unsigned int * StateVarOrder;
/*!
* \brief Table which calculates each state variable.
*/
NeuronModelTable ** StateVarTable;
/*!
* \brief Firing time table
*/
NeuronModelTable * FiringTable;
/*!
* \brief End firing time table
*/
NeuronModelTable * EndFiringTable;
/*!
* \brief Number of tables
*/
unsigned int NumTables;
/*!
* \brief Precalculated tables
*/
NeuronModelTable * Tables;
/*!
* \brief Vector where we temporary store initial values
*/
float * InitValues;
TableBasedModelHFEvent * tableBasedModelHFEvent;
double tableBasedModelHFTime;
/*!
* \brief It loads the neuron model description.
*
* It loads the neuron type description from the file .cfg.
*
* \param ConfigFile Name of the neuron description file (*.cfg).
*
* \throw EDLUTFileException If something wrong has happened in the file load.
*/
virtual void LoadNeuronModel(string ConfigFile) throw (EDLUTFileException);
//float MaxFiringPrediction();
//float recursive(void **cpointer, int idim, int ndims);
/*!
* \brief It loads the neuron model tables.
*
* It loads the neuron model tables from his .dat associated file.
*
* \pre The neuron model must be previously initialized or loaded
*
* \param TableFile Name of the table file (*.dat).
*
* \see LoadNeuronModel()
* \throw EDLUTException If something wrong has happened in the tables loads.
*/
virtual void LoadTables(string TableFile) throw (EDLUTException);
/*!
* \brief It returns the end of the refractory period.
*
* It returns the end of the refractory period.
*
* \param index index inside the VectorNeuronState of the neuron model.
* \param VectorNeuronState of the neuron model.
*
* \return The end of the refractory period. -1 if no spike is predicted.
*/
virtual double EndRefractoryPeriod(int index, VectorNeuronState * State);
/*!
* \brief It updates the neuron state after the evolution of the time.
*
* It updates the neuron state after the evolution of the time.
*
* \param index index inside the VectorNeuronState of the neuron model.
* \param VectorNeuronState of the neuron model.
* \param CurrentTime Current simulation time.
*/
virtual void UpdateState(int index, VectorNeuronState * State, double CurrentTime);
/*!
* \brief It abstracts the effect of an input spike in the cell.
*
* It abstracts the effect of an input spike in the cell.
*
* \param index index inside the VectorNeuronState of the neuron model.
* \param InputConnection Input connection from which the input spike has got the cell.
*/
virtual void SynapsisEffect(int index, Interconnection * InputConnection);
/*!
* \brief It returns the next spike time.
*
* It returns the next spike time.
*
* \param index index inside the VectorNeuronState of the neuron model.
* \param VectorNeuronState of the neuron model.
*
* \return The next firing spike time. -1 if no spike is predicted.
*/
virtual double NextFiringPrediction(int index, VectorNeuronState * State);
public:
/*!
* \brief Default constructor with parameters.
*
* It generates a new neuron model object loading the configuration of
* the model and the look-up tables.
*
* \param NeuronTypeID Neuron model type.
* \param NeuronModelID Neuron model description file.
*/
TableBasedModelHF(string NeuronTypeID, string NeuronModelID);
/*!
* \brief Class destructor.
*
* It destroys an object of this class.
*/
~TableBasedModelHF();
/*!
* \brief It loads the neuron model description and tables (if necessary).
*
* It loads the neuron model description and tables (if necessary).
*
* \throw EDLUTFileException If something wrong has happened in the file load.
*/
virtual void LoadNeuronModel() throw (EDLUTFileException);
/*!
* \brief It creates the neuron state and initializes to defined values.
*
* It creates the neuron state and initializes to defined values.
*
* \return A new object with the neuron state.
*/
virtual VectorNeuronState * InitializeState();
/*!
* \brief It generates the first spike (if any) in a cell.
*
* It generates the first spike (if any) in a cell.
*
* \param Cell The cell to check if activity is generated.
*
* \return A new internal spike if someone is predicted. 0 if none is predicted.
*/
virtual InternalSpike * GenerateInitialActivity(Neuron * Cell);
/*!
* \brief It processes a propagated spike (input spike in the cell).
*
* It processes a propagated spike (input spike in the cell).
*
* \note This function doesn't generate the next propagated spike. It must be externally done.
*
* \param inter the interconection which propagate the spike
* \param target the neuron which receives the spike
* \param time the time of the spike.
*
* \return A new internal spike if someone is predicted. 0 if none is predicted.
*/
virtual InternalSpike * ProcessInputSpike(Interconnection * inter, Neuron * target, double time);
/*!
* \brief It predicts if the neuron would generate a internalSpike after the ToleranceTime.
*
* It predicts if the neuron would generate a internalSpike after the ToleranceTime.
*
* \param target.
*
* \return A new internal spike if someone is predicted. 0 if none is predicted.
*/
InternalSpike * ProcessActivityAndPredictSpike(Neuron * target);
/*!
* \brief It processes an internal spike (generated spike in the cell).
*
* It processes an internal spike (generated spike in the cell).
*
* \note This function doesn't generate the next propagated (output) spike. It must be externally done.
* \note Before generating next spike, you should check if this spike must be discard.
*
* \see DiscardSpike
*
* \param OutputSpike The spike happened.
*
* \return A new internal spike if someone is predicted. 0 if none is predicted.
*/
virtual InternalSpike * GenerateNextSpike(InternalSpike * OutputSpike);
/*!
* \brief Check if the spike must be discard.
*
* Check if the spike must be discard. A spike must be discard if there are discrepancies between
* the next predicted spike and the spike time.
*
* \param OutputSpike The spike happened.
*
* \return True if the spike must be discard. False in otherwise.
*/
virtual bool DiscardSpike(InternalSpike * OutputSpike);
/*!
* \brief It prints the table based model info.
*
* It prints the current table based model characteristics.
*
* \param out The stream where it prints the information.
*
* \return The stream after the printer.
*/
virtual ostream & PrintInfo(ostream & out);
/*!
* \brief It initialice VectorNeuronState.
*
* It initialice VectorNeuronState.
*
* \param N_neurons cell number inside the VectorNeuronState.
*/
virtual void InitializeStates(int N_neurons, int OpenMPQueueIndex);
/*!
* \brief It Checks if the neuron model has this connection type.
*
* It Checks if the neuron model has this connection type.
*
* \param Type input connection type.
*
* \return A a valid connection type for this neuron model.
*/
virtual int CheckSynapseTypeNumber(int Type);
};
#endif /* TABLEBASEDMODELHF_H_ */