nnfw/src/neuralnet.cpp

/********************************************************************************
 *  Neural Network Framework.                                                   *
 *  Copyright (C) 2005-2011 Gianluca Massera <emmegian@yahoo.it>                *
 *                                                                              *
 *  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 2 of the License, or           *
 *  (at your option) any later version.                                         *
 *                                                                              *
 *  This program is distributed in the hope that it will be useful,             *
 *  but WITHOUT ANY WARRANTY; without even the implied warranty of              *
 *  MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the               *
 *  GNU General Public License for more details.                                *
 *                                                                              *
 *  You should have received a copy of the GNU General Public License           *
 *  along with this program; if not, write to the Free Software                 *
 *  Foundation, Inc., 51 Franklin St, Fifth Floor, Boston, MA  02110-1301  USA  *
 ********************************************************************************/

#include "neuralnet.h"

namespace farsa {

NeuralNet::NeuralNet() {
    dimUps = 0;
}

NeuralNet::~NeuralNet() {
}

void NeuralNet::addCluster( Cluster* c, bool isInput, bool isOutput ) {
#ifdef FARSA_DEBUG
    if ( !c ) {
        qWarning() << "Null Pointer passed to addCluster! This operation will be ignored" ;
        return;
    }
#endif
    // Check if the Cluster is already added
    if ( find( c ) ) {
#ifdef FARSA_DEBUG
        qWarning() << "Cluster already added! addCluster will be ignored" ;
#endif
        return;
    }
    clustersv.append( c );
    if ( isInput ) {
        inclusters.append( c );
    }
    if ( isOutput ) {
        outclusters.append( c );
    }
    if ( !isInput && !isOutput ) {
        hidclusters.append( c );
    }
    clsMap[c->name()] = c;
    return;
}

bool NeuralNet::removeCluster( Cluster* c ) {
#ifdef FARSA_DEBUG
    if ( !c ) {
        qWarning() << "Null Pointer passed to removeCluster! This operation will return false" ;
        return false;
    }
#endif
    if ( !find( c ) ) {
        return false;
    }
    clustersv.removeOne( c );
    inclusters.removeOne( c );
    outclusters.removeOne( c );
    hidclusters.removeOne( c );
    clsMap.remove( c->name() );
    return true;
}

void NeuralNet::markAsInput( Cluster* c ) {
#ifdef FARSA_DEBUG
    if ( !c ) {
        qWarning() << "Null Pointer passed to addCluster! This operation will be ignored" ;
        return;
    }
#endif
    // Check if the Cluster exists
    if ( !find( c ) ) {
#ifdef FARSA_DEBUG
        qWarning() << "attempt to mark a Cluster not present in this net!" ;
#endif
        return;
    }
    if ( inclusters.contains( c ) ) {
        return;
    }
    inclusters.append( c );
    hidclusters.removeOne( c );
}

void NeuralNet::markAsOutput( Cluster* c ) {
#ifdef FARSA_DEBUG
    if ( !c ) {
        qWarning() << "Null Pointer passed to addCluster! This operation will be ignored" ;
        return;
    }
#endif
    // Check if the Cluster exists
    if ( !find( c ) ) {
#ifdef FARSA_DEBUG
        qWarning() << "attempt to mark a Cluster not present in this net!" ;
#endif
        return;
    }
    if ( outclusters.contains( c ) ) {
        return;
    }
    outclusters.append( c );
    hidclusters.removeOne( c );
}

void NeuralNet::unmark( Cluster* c ) {
#ifdef FARSA_DEBUG
    if ( !c ) {
        qWarning() << "Null Pointer passed to addCluster! This operation will be ignored" ;
        return;
    }
#endif
    // Check if the Cluster exists
    if ( !find( c ) ) {
#ifdef FARSA_DEBUG
        qWarning() << "attempt to unmark a Cluster not present in this net!" ;
#endif
        return;
    }
    inclusters.removeOne( c );
    outclusters.removeOne( c );
    hidclusters.append( c );
    return;
}

void NeuralNet::unmarkAll( ) {
    inclusters.clear();
    outclusters.clear();
    hidclusters = clustersv;
    return;
}

bool NeuralNet::isIsolated( Cluster* c ) const {
#ifdef FARSA_DEBUG
    if ( !c ) {
        qWarning() << "Null Pointer passed to isIsolato! This operation will return false" ;
        return false;
    }
#endif
    return ( inLinks.count( c ) == 0 && outLinks.count( c ) == 0 );
}

ClusterList NeuralNet::clusters() const {
    return clustersv;
}

ClusterList NeuralNet::inputClusters() const {
    return inclusters;
}

ClusterList NeuralNet::outputClusters() const {
    return outclusters;
}

ClusterList NeuralNet::hiddenClusters() const {
    return hidclusters;
}

void NeuralNet::addLinker( Linker* l ) {
#ifdef FARSA_DEBUG
    if ( !l ) {
        qWarning() << "Null Pointer passed to addLinker! This operation will be ignored" ;
        return;
    }
#endif
    // Check if the Linker is already added
    if ( find( l ) ) {
#ifdef FARSA_DEBUG
        qWarning() << "Linker already added! addLinker will be ignored" ;
#endif
        return;
    }
#ifdef FARSA_DEBUG
    // --- Check: Are There in this net the Clusters that linker l connects ???
    if ( ! find( l->from() ) ) {
        qWarning() << "The linker that you want add links cluster" << l->from()->name() << "that doesn't exist in this net! This operation will be ignored" ;
        return;
    }
    if ( ! find( l->to() ) ) {
        qWarning() << "The linker that you want add links cluster" << l->to()->name() << "that doesn't exist in this net! This operation will be ignored" ;
        return;
    }
#endif
    linkersv.append( l );
    // Adding information in outLinks map
    outLinks[ l->from() ].append( l );
    // Adding information in inLinks map
    inLinks[ l->to() ].append( l );
    lksMap[l->name()] = l;
    return;
}

bool NeuralNet::removeLinker( Linker* l ) {
#ifdef FARSA_DEBUG
    if ( !l ) {
        qWarning() << "Null Pointer passed to removeLinker! This operation will return false" ;
        return false;
    }
#endif
    if ( !find(l) ) {
        return false;
    }
    linkersv.removeOne( l );
    outLinks[ l->from() ].removeOne( l );
    inLinks[ l->to() ].removeOne( l );
    lksMap.remove( l->name() );
    return true;
}

LinkerList NeuralNet::linkers() const {
    return linkersv;
}

LinkerList NeuralNet::linkers( Cluster* c, bool out ) const {
#ifdef FARSA_DEBUG
    if ( !c ) {
        qWarning() << "Null Pointer passed to linkers! This operation will return an empty LinkerList" ;
        return LinkerList();
    }
#endif
    if ( out ) {
        // Return outgoing linkers
        if ( outLinks.contains( c ) ) {
            return outLinks[c];
        }
    } else {
        // Return incoming linkers
        if ( inLinks.contains( c ) ) {
            return inLinks[c];
        }
    }
    return LinkerList();
}

void NeuralNet::setOrder( Updatable* u[], unsigned int dim ) {
    ups.clear();
    for( unsigned int i = 0; i<dim; i++ ) {
        if ( find( u[i] ) ) {
            ups.append( u[i] );
        }
#ifdef FARSA_DEBUG
        else {
            qWarning() << "In the Updatable order list passed there are some Clusters and/or Linkers not present in this NeuralNet";
        }
#endif
    }
    dimUps = ups.size();
    return;
}

void NeuralNet::setOrder( const UpdatableList& u ) {
    ups.clear();
    unsigned int dim = u.size();
    for( unsigned int i = 0; i<dim; i++ ) {
        if ( find( u[i] ) ) {
            ups.append( u[i] );
        }
#ifdef FARSA_DEBUG
        else {
            qWarning() << "In the Updatable order list passed there are some Clusters and/or Linkers not present in this NeuralNet";
        }
#endif
    }
    dimUps = ups.size();
    return;
}

void NeuralNet::randomize( double min, double max ) {
    int dim = clustersv.size();
    for( int i=0; i<dim; i++ ) {
        clustersv[i]->randomize( min, max );
    }
    dim = linkersv.size();
    for( int i=0; i<dim; i++ ) {
        linkersv[i]->randomize( min, max );
    }
}

Updatable* NeuralNet::getByName( QString name ) {
    if ( clsMap.contains( name ) ) {
        return clsMap[name];
    }
    if ( lksMap.contains( name ) ) {
        return lksMap[name];
    }
#ifdef FARSA_DEBUG
    qWarning() << "getByName: Updatable " << name << " not present in BaseNeuralNet";
#endif
    return NULL;
}

bool NeuralNet::find( const Cluster* cl ) const {
    return clustersv.contains( (Cluster*)cl );
}

bool NeuralNet::find( const Linker* l ) const {
    return linkersv.contains( (Linker*)l );
}

bool NeuralNet::find( const Updatable* u ) const {
    return (
        clustersv.contains( (Cluster*)u ) ||
        linkersv.contains( (Linker*)u )
    );
}

void NeuralNet::configure(ConfigurationParameters& params, QString prefix) {
    params.startRememberingGroupObjectAssociations();
    // the name will correspond to the group where the declaration of NeuralNet is
    QString myname = prefix;
    if ( myname.startsWith( ConfigurationParameters::GroupSeparator() ) ) {
        myname.remove( 0, 1 );
    }
    if ( myname.endsWith( ConfigurationParameters::GroupSeparator() ) ) {
        myname.chop( 1 );
    }
    setName( myname );
    
    // All Cluster has to be added before the creation of linkers, that's why the insertion is done after
    ClusterList clsToAdd;
    ClusterList clsInput;
    ClusterList clsOutput;
    LinkerList lnsToAdd;

    //--- get all subgroups, merge this list with clustersList and linkersList
    QStringList subgroups;
    foreach( QString sub, params.getGroupsList( prefix ) ) {
        // append the full path
        subgroups << (prefix+sub);
    }
    //--- clustersList and linkersList are supposed to be full path
    QString str = params.getValue(prefix + "clustersList");
    if (!str.isEmpty()) {
        subgroups << str.split(QRegExp("\\s+"), QString::SkipEmptyParts);
    }
    str = params.getValue(prefix + "linkersList");
    if (!str.isEmpty()) {
        subgroups << str.split(QRegExp("\\s+"), QString::SkipEmptyParts);
    }
    foreach( QString sub, subgroups ) {
        Updatable* up = params.getObjectFromGroup<Updatable>( sub, true );
        //--- check if is a Cluster
        Cluster* cl = dynamic_cast<Cluster*>(up);
        if ( cl && !clsToAdd.contains( cl ) ) clsToAdd << cl;
        //--- check if is a Linker
        Linker* ln = dynamic_cast<Linker*>(up);
        if ( ln && !lnsToAdd.contains( ln ) ) lnsToAdd << ln;
    }
    //--- parse the parameter inputClusters
    str = params.getValue(prefix + "inputClusters");
    if (!str.isEmpty()) {
        QStringList list = str.split(QRegExp("\\s+"), QString::SkipEmptyParts);
        foreach( QString sub, list ) {
            Cluster* cl = params.getObjectFromGroup<Cluster>( sub, true );
            if ( !clsToAdd.contains( cl ) ) clsToAdd << cl;
            clsInput << cl;
        }
    }
    //--- parse the parameter outputClusters
    str = params.getValue(prefix + "outputClusters");
    if (!str.isEmpty()) {
        QStringList list = str.split(QRegExp("\\s+"), QString::SkipEmptyParts);
        foreach( QString sub, list ) {
            Cluster* cl = params.getObjectFromGroup<Cluster>( sub, true );
            if ( !clsToAdd.contains( cl ) ) clsToAdd << cl;
            clsOutput << cl;
        }
    }
    //--- parse the parameter spreadOrder
    str = params.getValue(prefix + "spreadOrder");
    if (!str.isEmpty()) {
        QStringList list = str.split(QRegExp("\\s+"), QString::SkipEmptyParts);
        ups.clear();
        foreach( QString sub, list ) {
            Updatable* up = params.getObjectFromGroup<Updatable>( sub, true );
            // add it to the list of objects to add
            Cluster* cl = dynamic_cast<Cluster*>(up);
            if ( cl && !clsToAdd.contains( cl ) ) clsToAdd << cl;
            //--- check if is a Linker
            Linker* ln = dynamic_cast<Linker*>(up);
            if ( ln && !lnsToAdd.contains( ln ) ) lnsToAdd << ln;
            ups.append( up );
        }
        dimUps = ups.size();
    }
    foreach( Cluster* cl, clsToAdd ) {
        addCluster( cl );
    }
    foreach( Cluster* cl, clsInput ) {
        markAsInput( cl );
    }
    foreach( Cluster* cl, clsOutput ) {
        markAsOutput( cl );
    }
    foreach( Linker* ln, lnsToAdd ) {
        addLinker( ln );
    }
    params.stopRememberingGroupObjectAssociations();
}

void NeuralNet::save(ConfigurationParameters& params, QString prefix) {
    params.startObjectParameters( prefix, "NeuralNet", this );
    //--- save all Clusters and generate parameter clustersList
    QStringList list;
    list.clear();
    foreach( Cluster* cl, clustersv ) {
        params.createGroup( cl->name() );
        cl->save( params, cl->name() );
        list << cl->name();
    }
    params.createParameter( prefix, "clustersList", list.join(" ") );
    //--- save all Linkers and generate parameter linkersList
    list.clear();
    foreach( Linker* ln, linkersv ) {
        params.createGroup( ln->name() );
        ln->save( params, ln->name() );
        list << ln->name();
    }
    params.createParameter( prefix, "linkersList", list.join(" ") );
    //--- save parameter inputClusters
    list.clear();
    foreach( Cluster* cl, inclusters ) {
        list << cl->name();
    }
    params.createParameter( prefix, "inputClusters", list.join(" ") );
    //--- save parameter outputClusters
    list.clear();
    foreach( Cluster* cl, outclusters ) {
        list << cl->name();
    }
    params.createParameter( prefix, "outputClusters", list.join(" ") );
    //--- save parameter spreadOrder
    list.clear();
    foreach( Updatable* up, ups ) {
        list << up->name();
    }
    params.createParameter( prefix, "spreadOrder", list.join(" ") );
}

void NeuralNet::describe( QString type ) {
    Descriptor d = addTypeDescription( type, "Neural Network", "The neural network is a container for Clusters and Linkers. The topology and connectivity of the neural network is specified by the Linkers contained, while the order on which the Clusters and Linkers are updated is specified by the spreadOrder parameter of neural network" );
    d.describeObject( "clustersList" ).type( "Cluster" ).props( IsList ).help( "The list of all Clusters" );
    d.describeObject( "linkersList" ).type( "Linker" ).props( IsList ).help( "The list of all Linkers" );
    d.describeObject( "inputClusters" ).type( "Cluster" ).props( IsList ).help( "The list of Clusters marked as input of the neural network" );
    d.describeObject( "outputClusters" ).type( "Cluster" ).props( IsList ).help( "The list of Clusters marked as output of the neural network" );
    d.describeObject( "spreadOrder" ).type( "Updatable" ).props( IsMandatory | IsList ).help( "The order on which the Clusters and Linkers are updated", "All Clusters and Linkers specified here are also automatically added into the neural network. If this list contains all Clusters and Linkers, then the clustersList and the linkersList parameters are superfluous" );
}

}