connection_monitor_core.cpp

Go to the documentation of this file.

/* * Copyright (c) 2016 Regents of the University of California. All rights reserved.
*
* Redistribution and use in source and binary forms, with or without
* modification, are permitted provided that the following conditions
* are met:
*
* 1. Redistributions of source code must retain the above copyright
*    notice, this list of conditions and the following disclaimer.
*
* 2. Redistributions in binary form must reproduce the above copyright
*    notice, this list of conditions and the following disclaimer in the
*    documentation and/or other materials provided with the distribution.
*
* 3. The names of its contributors may not be used to endorse or promote
*    products derived from this software without specific prior written
*    permission.
*
* THIS SOFTWARE IS PROVIDED BY THE COPYRIGHT HOLDERS AND CONTRIBUTORS
* "AS IS" AND ANY EXPRESS OR IMPLIED WARRANTIES, INCLUDING, BUT NOT
* LIMITED TO, THE IMPLIED WARRANTIES OF MERCHANTABILITY AND FITNESS FOR
* A PARTICULAR PURPOSE ARE DISCLAIMED. IN NO EVENT SHALL THE COPYRIGHT OWNER OR
* CONTRIBUTORS BE LIABLE FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL,
* EXEMPLARY, OR CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO,
* PROCUREMENT OF SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR
* PROFITS; OR BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
* LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
* NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
* SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
*
* *********************************************************************************************** *
* CARLsim
* created by: (MDR) Micah Richert, (JN) Jayram M. Nageswaran
* maintained by:
* (MA) Mike Avery <averym@uci.edu>
* (MB) Michael Beyeler <mbeyeler@uci.edu>,
* (KDC) Kristofor Carlson <kdcarlso@uci.edu>
* (TSC) Ting-Shuo Chou <tingshuc@uci.edu>
* (HK) Hirak J Kashyap <kashyaph@uci.edu>
*
* CARLsim v1.0: JM, MDR
* CARLsim v2.0/v2.1/v2.2: JM, MDR, MA, MB, KDC
* CARLsim3: MB, KDC, TSC
* CARLsim4: TSC, HK
*
* CARLsim available from http://socsci.uci.edu/~jkrichma/CARLsim/
* Ver 12/31/2016
*/
#include <connection_monitor_core.h>
 
#include <snn.h>                // CARLsim private implementation
#include <snn_definitions.h>    // KERNEL_ERROR, KERNEL_INFO, ...
 
#include <sstream>              // std::stringstream
#include <algorithm>            // std::sort
#include <iomanip>              // std::setfill, std::setw
#include <float.h>              // FLT_EPSILON
 
 
 
// we aren't using namespace std so pay attention!
ConnectionMonitorCore::ConnectionMonitorCore(SNN* snn,int monitorId,short int connId,int grpIdPre,int grpIdPost) {
    snn_ = snn;
    connId_= connId;
    grpIdPre_ = grpIdPre;
    grpIdPost_ = grpIdPost;
    monitorId_ = monitorId;
 
    wtTime_ = -1;
    wtTimeLast_ = -1;
    wtTimeWrite_ = -1;
 
    connFileId_ = NULL;
    needToWriteFileHeader_ = true;
    needToInit_ = true;
    connFileSignature_ = 202029319;
    connFileVersion_ = 0.3f;
 
    minWt_ = -1.0f;
    maxWt_ = -1.0f;
 
    connFileTimeIntervalSec_ = 1;
}
 
void ConnectionMonitorCore::init() {
    if (!needToInit_)
        return;
 
    nNeurPre_  = snn_->getGroupNumNeurons(grpIdPre_);
    nNeurPost_ = snn_->getGroupNumNeurons(grpIdPost_);
    isPlastic_ = snn_->isConnectionPlastic(connId_);
    nSynapses_ = snn_->getNumSynapticConnections(connId_);
 
    ConnectConfig connInfo = snn_->getConnectConfig(connId_);
    minWt_ = 0.0f; // for now, no non-zero min weights allowed
    maxWt_ = fabs(connInfo.maxWt);
 
    assert(nNeurPre_>0);
    assert(nNeurPost_>0);
 
    // use KERNEL_{ERROR|WARNING|etc} typesetting (const FILE*)
    fpInf_ = snn_->getLogFpInf();
    fpErr_ = snn_->getLogFpErr();
    fpDeb_ = snn_->getLogFpDeb();
    fpLog_ = snn_->getLogFpLog();
 
    // init weight matrix with right dimensions
    for (int i = 0; i < nNeurPre_; i++) {
        std::vector<float> wt;
        for (int j = 0; j < nNeurPost_; j++) {
            wt.push_back(NAN);
        }
        wtMat_.push_back(wt);
        wtMatLast_.push_back(wt);
    }
 
    // then load current weigths from SNN into weight matrix
    updateStoredWeights();
}
 
ConnectionMonitorCore::~ConnectionMonitorCore() {
    if (connFileId_!=NULL) {
        // flush: store last snapshot to file if update interval set
        if (connFileTimeIntervalSec_ > 0) {
            // make sure SNN is not already deallocated!
            assert(snn_!=NULL);
            writeConnectFileSnapshot(snn_->getSimTime(), snn_->getWeightMatrix2D(connId_));
        }
 
        // then close file and clean up
        fclose(connFileId_);
        connFileId_ = NULL;
        needToInit_ = true;
        needToWriteFileHeader_ = true;
    }
}
 
// +++++ PUBLIC METHODS: +++++++++++++++++++++++++++++++++++++++++++++++//
 
// calculate weight changes since last update (element-wise )
std::vector< std::vector<float> > ConnectionMonitorCore::calcWeightChanges() {
    updateStoredWeights();
    std::vector< std::vector<float> > wtChange(nNeurPre_, std::vector<float>(nNeurPost_));
 
    // take the naive approach for now
    for (int i=0; i<nNeurPre_; i++) {
        for (int j=0; j<nNeurPost_; j++) {
            wtChange[i][j] = wtMat_[i][j] - wtMatLast_[i][j];
        }
    }
 
    return wtChange;
}
 
 
// reset weight matrix
void ConnectionMonitorCore::clear() {
    for (int i=0; i<nNeurPre_; i++) {
        for (int j=0; j<nNeurPost_; j++) {
            wtMat_[i][j] = NAN;
            wtMatLast_[i][j] = NAN;
        }
    }
}
 
// find number of incoming synapses for a specific post neuron
int ConnectionMonitorCore::getFanIn(int neurPostId) {
    assert(neurPostId<nNeurPost_);
    int nSyn = 0;
    for (int i=0; i<nNeurPre_; i++) {
        if (!isnan(wtMat_[i][neurPostId])) {
            nSyn++;
        }
    }
    return nSyn;
}
 
// find number of outgoing synapses of a specific pre neuron
int ConnectionMonitorCore::getFanOut(int neurPreId) {
    assert(neurPreId<nNeurPre_);
    int nSyn = 0;
    for (int j=0; j<nNeurPost_; j++) {
        if (!isnan(wtMat_[neurPreId][j])) {
            nSyn++;
        }
    }
    return nSyn;
}
 
float ConnectionMonitorCore::getMaxWeight(bool getCurrent) {
    float maxVal = minWt_;
    if (getCurrent) {
        updateStoredWeights();
 
        // find currently largest weight value
        for (int i=0; i<nNeurPre_; i++) {
            for (int j=0; j<nNeurPost_; j++) {
                // skip entries in matrix where no synapse exists
                if (isnan(wtMat_[i][j]))
                    continue;
 
                if (wtMat_[i][j] > maxVal) {
                    maxVal = wtMat_[i][j];
                }
            }
        }
    } else {
        // return RangeWeight.max
        maxVal = maxWt_;
    }
 
    return maxVal;
}
 
float ConnectionMonitorCore::getMinWeight(bool getCurrent) {
    float minVal = maxWt_;
    if (getCurrent) {
        updateStoredWeights();
 
        // find currently largest weight value
        for (int i=0; i<nNeurPre_; i++) {
            for (int j=0; j<nNeurPost_; j++) {
                // skip entries in matrix where no synapse exists
                if (isnan(wtMat_[i][j]))
                    continue;
 
                if (wtMat_[i][j] < minVal) {
                    minVal = wtMat_[i][j];
                }
            }
        }
    } else {
        // return RangeWeight.min
        minVal = minWt_;
    }
 
    return minVal;
}
 
// find number of synapses whose weights changed
int ConnectionMonitorCore::getNumWeightsChanged(double minAbsChange) {
    assert(minAbsChange>=0.0);
    std::vector< std::vector<float> > wtChange = calcWeightChanges();
 
    int nChanged = 0;
    for (int i=0; i<nNeurPre_; i++) {
        for (int j=0; j<nNeurPost_; j++) {
            // skip entries in matrix where no synapse exists
            if (isnan(wtMat_[i][j]))
                continue;
 
            if (fabs(wtChange[i][j]) >= minAbsChange) {
                nChanged++;
            }
        }
    }
    return nChanged;
}
 
// finds the number of weights with values in some range
int ConnectionMonitorCore::getNumWeightsInRange(double minVal, double maxVal) {
    assert(maxVal>=minVal);
 
    updateStoredWeights();
 
    // make sure values are inside a reasonable range
    if (minVal<=getMinWeight(false) && minVal>=getMaxWeight(false)) {
        return getNumSynapses();
    }
 
    int cnt = 0;
    for (int i=0; i<nNeurPre_; i++) {
        for (int j=0; j<nNeurPost_; j++) {
            // skip entries in matrix where no synapse exists
            if (isnan(wtMat_[i][j]))
                continue;
 
            if (wtMat_[i][j]>=minVal && wtMat_[i][j]<=maxVal) {
                cnt++;
            }
        }
    }
 
    return cnt;
}
 
// finds the number of weights with some exact weight value
int ConnectionMonitorCore::getNumWeightsWithValue(double value) {
    // make sure value is inside a reasonable range
    if (value<getMinWeight(false) || value>getMaxWeight(false)) {
        return 0;
    }
 
    return getNumWeightsInRange(value-FLT_EPSILON, value+FLT_EPSILON);
}
 
// calculate total absolute amount of weight change
double ConnectionMonitorCore::getTotalAbsWeightChange() {
    std::vector< std::vector<float> > wtChange = calcWeightChanges();
    double wtTotalChange = 0.0;
    for (int i=0; i<nNeurPre_; i++) {
        for (int j=0; j<nNeurPost_; j++) {
            // skip entries in matrix where no synapse exists
            if (isnan(wtMat_[i][j]))
                continue;
            wtTotalChange += fabs(wtChange[i][j]);
        }
    }
    return wtTotalChange;
}
 
void ConnectionMonitorCore::print() {
    updateStoredWeights();
 
    KERNEL_INFO("(t=%.3fs) ConnectionMonitor ID=%d: %d(%s) => %d(%s)",
        (getTimeMsCurrentSnapshot()/1000.0f), connId_,
        grpIdPre_, snn_->getGroupName(grpIdPre_).c_str(),
        grpIdPost_, snn_->getGroupName(grpIdPost_).c_str());
 
    // generate header
    std::stringstream header, header2;
    header  << " pre\\post |";
    header2 << "----------|";
    for (int j=0; j<nNeurPost_; j++) {
        header  << std::setw(9) << std::setfill(' ') << j << " |";
        header2 << "-----------";
    }
    KERNEL_INFO("%s",header.str().c_str());
    KERNEL_INFO("%s",header2.str().c_str());
 
    for (int i=0; i<nNeurPre_; i++) {
        std::stringstream line;
        line << std::setw(9) << std::setfill(' ') << i << " |";
        for (int j=0; j<nNeurPost_; j++) {
            line << std::fixed << std::setprecision(4) << (isnan(wtMat_[i][j])?"      ":(wtMat_[i][j]>=0?"   ":"  "))
                << wtMat_[i][j]  << "  ";
        }
        KERNEL_INFO("%s",line.str().c_str());
    }
}
 
void ConnectionMonitorCore::printSparse(int neurPostId, int maxConn, int connPerLine, bool storeNewSnapshot) {
    assert(neurPostId<nNeurPost_);
    assert(maxConn>0);
    assert(connPerLine>0);
 
    // give the option of not storing the new snapshot
    std::vector< std::vector<float> > wtNew, wtOld;
    long int timeNew, timeOld;
    if (!storeNewSnapshot) {
        // make a copy of current snapshots so that we can restore them later
        wtNew = wtMat_;
        wtOld = wtMatLast_;
        timeNew = wtTime_;
        timeOld = wtTimeLast_;
    }
 
    updateStoredWeights();
    KERNEL_INFO("(t=%.3fs) ConnectionMonitor ID=%d %d(%s) => %d(%s): [preId,postId] wt (+/-wtChange in %ldms) "
        "show first %d", getTimeMsCurrentSnapshot()/1000.0f, connId_,
        grpIdPre_, snn_->getGroupName(grpIdPre_).c_str(), grpIdPost_, snn_->getGroupName(grpIdPost_).c_str(),
        getTimeMsSinceLastSnapshot(), maxConn);
 
    int postA, postZ;
    if (neurPostId==ALL) {
        postA = 0;
        postZ = nNeurPost_ - 1;
    } else {
        postA = neurPostId;
        postZ = neurPostId;
    }
 
    std::vector< std::vector<float> > wtChange;
    if (isPlastic_) {
        wtChange = calcWeightChanges();
    }
 
    std::stringstream line;
    int nConn = 0;
    int maxIntDigits = ceil(log10((double)std::max(nNeurPre_,nNeurPost_)));
    for (int i=0; i<nNeurPre_; i++) {
        for (int j = postA; j <= postZ; j++) {
            // display only so many connections
            if (nConn>=maxConn)
                break;
 
            if (!isnan(wtMat_[i][j])) {
                line << "[" << std::setw(maxIntDigits) << i << "," << std::setw(maxIntDigits) << j << "] "
                    << std::fixed << std::setprecision(4) << wtMat_[i][j];
                if (isPlastic_) {
                    line << " (" << ((wtChange[i][j]<0)?"":"+");
                    line << std::setprecision(4) << wtChange[i][j] << ")";
                }
                line << "   ";
                if (!(++nConn % connPerLine)) {
                    KERNEL_INFO("%s",line.str().c_str());
                    line.str(std::string());
                }
            }
        }
    }
    // flush
    if (nConn % connPerLine)
        KERNEL_INFO("%s",line.str().c_str());
 
    if (!storeNewSnapshot) {
        wtMat_ = wtNew;
        wtMatLast_ = wtOld;
        wtTime_ = timeNew;
        wtTimeLast_ = timeOld;
    }
 
}
 
void ConnectionMonitorCore::setConnectFileId(FILE* connFileId) {
    // \TODO consider the case where this function is called more than once
    if (connFileId_!=NULL)
        KERNEL_ERROR("ConnectionMonitorCore: setConnectFileId has already been called.");
 
    connFileId_=connFileId;
 
    if (connFileId_==NULL) {
        needToWriteFileHeader_ = false;
    }
    else {
        // for now: file pointer has changed, so we need to write header (again)
        needToWriteFileHeader_ = true;
        writeConnectFileHeader();
    }
}
 
void ConnectionMonitorCore::setUpdateTimeIntervalSec(int intervalSec) {
    assert(intervalSec==-1 || intervalSec>=1);
    connFileTimeIntervalSec_ = intervalSec;
}
 
// updates the internally stored last two snapshots (current one and last one)
void ConnectionMonitorCore::updateStoredWeights() {
    if (snn_->getSimTime() > wtTime_) {
        // time has advanced: get new weights
        wtMatLast_ = wtMat_;
        wtTimeLast_ = wtTime_;
 
        wtMat_ = snn_->getWeightMatrix2D(connId_);
        wtTime_ = snn_->getSimTime();
    }
}
 
// returns a current snapshot
std::vector< std::vector<float> > ConnectionMonitorCore::takeSnapshot() {
    updateStoredWeights();
    writeConnectFileSnapshot(wtTime_, wtMat_);
    return wtMat_;
}
 
// write the header section of the spike file
// this should be done once per file, and should be the very first entries in the file
void ConnectionMonitorCore::writeConnectFileHeader() {
    init();
 
    if (!needToWriteFileHeader_)
        return;
 
    // write file signature
    if (!fwrite(&connFileSignature_,sizeof(int),1,connFileId_))
        KERNEL_ERROR("ConnectionMonitorCore: writeConnectFileHeader has fwrite error");
 
    // write version number
    if (!fwrite(&connFileVersion_,sizeof(float),1,connFileId_))
        KERNEL_ERROR("ConnectionMonitorCore: writeConnectFileHeader has fwrite error");
 
    // write connection id
    if (!fwrite(&connId_,sizeof(short int),1,connFileId_))
        KERNEL_ERROR("ConnectionMonitor: writeConnectFileHeader has fwrite error");
 
    // write pre group info: group id and Grid3D dimensions
    Grid3D gridPre = snn_->getGroupGrid3D(grpIdPre_);
    if (!fwrite(&grpIdPre_,sizeof(int),1,connFileId_))
        KERNEL_ERROR("ConnectionMonitor: writeConnectFileHeader has fwrite error");
    if (!fwrite(&(gridPre.numX),sizeof(int),1,connFileId_))
        KERNEL_ERROR("ConnectionMonitor: writeConnectFileHeader has fwrite error");
    if (!fwrite(&(gridPre.numY),sizeof(int),1,connFileId_))
        KERNEL_ERROR("ConnectionMonitor: writeConnectFileHeader has fwrite error");
    if (!fwrite(&(gridPre.numZ),sizeof(int),1,connFileId_))
        KERNEL_ERROR("ConnectionMonitor: writeConnectFileHeader has fwrite error");
 
    // write post group info: group id and # neurons
    Grid3D gridPost = snn_->getGroupGrid3D(grpIdPost_);
    if (!fwrite(&grpIdPost_,sizeof(int),1,connFileId_))
        KERNEL_ERROR("ConnectionMonitor: writeConnectFileHeader has fwrite error");
    if (!fwrite(&(gridPost.numX),sizeof(int),1,connFileId_))
        KERNEL_ERROR("ConnectionMonitor: writeConnectFileHeader has fwrite error");
    if (!fwrite(&(gridPost.numY),sizeof(int),1,connFileId_))
        KERNEL_ERROR("ConnectionMonitor: writeConnectFileHeader has fwrite error");
    if (!fwrite(&(gridPost.numZ),sizeof(int),1,connFileId_))
        KERNEL_ERROR("ConnectionMonitor: writeConnectFileHeader has fwrite error");
 
    // write number of synapses
    if (!fwrite(&nSynapses_,sizeof(int),1,connFileId_))
        KERNEL_ERROR("ConnectionMonitor: writeConnectFileHeader has fwrite error");
 
    // write synapse type (fixed=false, plastic=true)
    if (!fwrite(&isPlastic_,sizeof(bool),1,connFileId_))
        KERNEL_ERROR("ConnectionMonitor: writeConnectFileHeader has fwrite error");
 
    // write minWt and maxWt
    if (!fwrite(&minWt_,sizeof(float),1,connFileId_))
        KERNEL_ERROR("ConnectionMonitorCore: writeConnectFileHeader has fwrite error");
    if (!fwrite(&maxWt_,sizeof(float),1,connFileId_))
        KERNEL_ERROR("ConnectionMonitorCore: writeConnectFileHeader has fwrite error");
 
 
    // \TODO: write delays
 
    needToWriteFileHeader_ = false;
}
 
void ConnectionMonitorCore::writeConnectFileSnapshot(int simTimeMs, std::vector< std::vector<float> > wts) {
    // don't write if we have already written this timestamp to file (or file doesn't exist)
    if ((long long)simTimeMs <= wtTimeWrite_ || connFileId_==NULL) {
        return;
    }
 
    wtTimeWrite_ = (long long)simTimeMs;
 
    // write time stamp
    if (!fwrite(&wtTimeWrite_,sizeof(long long),1,connFileId_))
        KERNEL_ERROR("ConnectionMonitor: writeConnectFileSnapshot has fwrite error");
 
    // write all weights
    for (int i=0; i<nNeurPre_; i++) {
        for (int j=0; j<nNeurPost_; j++) {
            if (!fwrite(&wts[i][j],sizeof(float),1,connFileId_)) {
                KERNEL_ERROR("ConnectionMonitor: writeConnectFileSnapshot has fwrite error");
            }
        }
    }
}