// Analysis_Timecorr::Analyze()
Analysis::RetType Analysis_Timecorr::Analyze() {
// If 2 vectors, ensure they have the same # of frames
if (vinfo2_!=0) {
if (vinfo1_->Size() != vinfo2_->Size()) {
mprinterr("Error: # Frames in vec %s (%i) != # Frames in vec %s (%i)\n",
vinfo1_->legend(), vinfo1_->Size(),
vinfo2_->legend(), vinfo2_->Size());
return Analysis::ERR;
}
}
// Determine sizes
int frame = vinfo1_->Size();
int time = (int)(tcorr_ / tstep_) + 1;
// nsteps
int nsteps = 0;
if (time > frame)
nsteps = frame;
else
nsteps = time;
// Allocate memory to hold complex numbers for direct or FFT
if (drct_) {
data1_.Allocate( frame );
if (mode_ == CROSSCORR)
data2_.Allocate( frame );
corfdir_.Allocate( nsteps );
} else {
// Initialize FFT
pubfft_.Allocate( frame );
data1_ = pubfft_.Array();
if (mode_ == CROSSCORR)
data2_ = data1_;
}
// ----- Calculate spherical harmonics ---------
// Real + Img. for each -order <= m <= order
if (vinfo1_->CalcSphericalHarmonics(order_)) return Analysis::ERR;
if (vinfo2_ != 0) {
if (vinfo2_->CalcSphericalHarmonics(order_)) return Analysis::ERR;
}
// ----- Initialize PN output array memory -----
DataSet_double& pncf_ = static_cast<DataSet_double&>( *tc_p_ );
pncf_.Resize( nsteps );
Dimension Xdim(0.0, tstep_, nsteps, "Time");
pncf_.SetDim(Dimension::X, Xdim);
// ----- Calculate PN --------------------------
for (int midx = -order_; midx <= order_; ++midx) {
data1_.Assign( vinfo1_->SphericalHarmonics( midx ) );
if (vinfo2_ != 0)
data2_.Assign( vinfo2_->SphericalHarmonics( midx ) );
CalcCorr( frame );
for (int k = 0; k < nsteps; ++k)
pncf_[k] += data1_[2 * k];
}
// ----- Dipolar Calc. -------------------------
AvgResults Avg1, Avg2;
if (dplr_) {
DataSet_double& cf_ = static_cast<DataSet_double&>( *tc_c_ );
cf_.Resize( nsteps );
cf_.SetDim(Dimension::X, Xdim);
DataSet_double& rcf_ = static_cast<DataSet_double&>( *tc_r3r3_ );
rcf_.Resize( nsteps );
rcf_.SetDim(Dimension::X, Xdim);
// Calculate averages
std::vector<double> R3i_1 = CalculateAverages(*vinfo1_, Avg1);
std::vector<double> R3i_2;
if (vinfo2_ != 0)
R3i_2 = CalculateAverages(*vinfo2_, Avg2);
// C
for (int midx = -order_; midx <= order_; ++midx) {
data1_.Assign( vinfo1_->SphericalHarmonics( midx ) );
if (vinfo2_ != 0)
data2_.Assign( vinfo2_->SphericalHarmonics( midx ) );
for (int i = 0, i2 = 0; i < frame; ++i, i2 += 2) {
double r3i = R3i_1[ i ];
data1_[i2 ] *= r3i;
data1_[i2+1] *= r3i;
if ( vinfo2_ != 0 ) {
r3i = R3i_2[ i ];
data2_[i2 ] *= r3i;
data2_[i2+1] *= r3i;
}
}
CalcCorr( frame );
for (int k = 0; k < nsteps; ++k)
cf_[k] += data1_[2 * k];
}
// 1 / R^6
for (int i = 0, i2 = 0; i < frame; ++i, i2 += 2) {
data1_[i2 ] = R3i_1[ i ];
data1_[i2+1] = 0.0;
if ( vinfo2_ != 0 ) {
data2_[i2 ] = R3i_2[ i ];
data2_[i2+1] = 0.0;
}
}
CalcCorr( frame );
for (int k = 0; k < nsteps; ++k)
rcf_[k] = data1_[2 * k];
}
// ----- NORMALIZATION -------------------------
// 4*PI / ((2*order)+1) due to spherical harmonics addition theorem
double KN = DataSet_Vector::SphericalHarmonicsNorm( order_ );
Normalize( tc_p_, frame, KN );
if (dplr_) {
Normalize( tc_c_, frame, KN );
Normalize( tc_r3r3_, frame, 1.0 );
}
// ----- PRINT PTRAJ FORMAT --------------------
if (outfile_ != 0) {
outfile_->Printf("%ss, normal type\n",ModeString[mode_]);
if (dplr_) {
outfile_->Printf("***** Vector length *****\n");
outfile_->Printf("%10s %10s %10s %10s\n", "<r>", "<rrig>", "<1/r^3>", "<1/r^6>");
outfile_->Printf("%10.4f %10.4f %10.4f %10.4f\n",
Avg1.rave_, Avg1.avgr_, Avg1.r3iave_, Avg1.r6iave_);
if (mode_ == CROSSCORR)
outfile_->Printf("%10.4f %10.4f %10.4f %10.4f\n",
Avg2.rave_, Avg2.avgr_, Avg2.r3iave_, Avg2.r6iave_);
}
if (ptrajformat_) {
outfile_->Printf("\n***** Correlation functions *****\n");
if (dplr_) {
DataSet_double& cf_ = static_cast<DataSet_double&>( *tc_c_ );
DataSet_double& rcf_ = static_cast<DataSet_double&>( *tc_r3r3_ );
outfile_->Printf("%10s %10s %10s %10s\n", "Time", "<C>", Plegend_[order_], "<1/(r^3*r^3)>");
for (int i = 0; i < nsteps; ++i)
outfile_->Printf("%10.3f %10.4f %10.4f %10.4f\n", (double)i * tstep_,
cf_[i], pncf_[i], rcf_[i]);
} else {
outfile_->Printf("%10s %10s\n", "Time", Plegend_[order_]);
for (int i = 0; i < nsteps; ++i)
outfile_->Printf("%10.3f %10.4f\n", (double)i * tstep_, pncf_[i]);
}
}
}
return Analysis::OK;
}
int CorrAnalysis(int argc, char **argv)
{
//if (argc < 6 || !strcmp(argv[1],"-h") || argc < 8+2* (atoi(argv[4])) )
if ( argc != 7+2*(atoi(argv[5])) )
{
cerr << "\n\n\n";
cerr << "Usage: " << argv[0] << " <task> [params]\n";
cerr << "<task>: operation to be performed on data. \n";
cerr << " 0 - correlation analysis\n";
cerr << " 1 - naive prediction\n";
cerr << " 2 - perceptron prediction\n";
cerr << " 3 - neural network\n";
cerr << "<prefix>: I/O prefix\n";
cerr << "[params]: algorithm specific parameters\n";
cerr << endl;
cerr << "correlation analysis:\n";
cerr << "[datafile]: name of file containing data. \n";
cerr << "[diff]: level of differencing only zero and first level differencing is supported. \n";
cerr << "[nvar]: number of variables in datafile. \n";
cerr << "[tgtIdx]: Column of target variable (first column = 0).\n";
cerr << "[delay1]: delay before starting time-series of first variable\n";
cerr << "[nlag1]: number of lags to compose the time-series of the first variable from\n";
cerr << "[delay2]: delay before starting time-series of secon variable\n";
cerr << "[nlag2]: number of lags to compose the time-series of the second variable from\n";
cerr << "\n\n\n";
exit(-1);
}
string prefix = argv[2];
cout << "# I/O prefix: " << prefix << endl;
//
string ifile_name = argv[3];
cout << "# inputfile is: " << ifile_name << endl;
//
int diff = atoi(argv[4]);
cout << "# differencing level is: " << diff << endl;
if( diff < 0 || diff > 1)
{
cerr << "Assert: Invalid level of differencing\n\n";
exit(-1);
}
//
int nvar = atoi(argv[5]);
cout << "# nvar is: " << nvar << endl;
//
int tgtIdx = atoi(argv[6]);
cout << "# Target index is: " << tgtIdx << endl;
if( tgtIdx > nvar-1 )
{
cerr << "Assert: Invalid target index\n\n";
exit(-1);
}
//
vector<int> delay(nvar);
vector<int> nlags(nvar);
cout << "# ( delay , lag )\n";
for(int i=0; i< nvar; i++)
{
int argvIdx = 7+2*i;
delay[i] = atoi(argv[argvIdx]);
nlags[i] = atoi(argv[argvIdx+1]);
if( i==tgtIdx && delay[i] <= 0)
{
delay[i]==1;
}
cout << "# ( " << delay[i] << "," << nlags[i] << " )" << endl;
}
int nAtt = 0;
for( int i=0; i< nvar; i++)
{
nAtt += nlags[i];
}
//
cout << "# Reading Data" << endl;
vector< vector< float > > Examples;
vector< string > Timestamp;
ReadData( ifile_name, diff, nvar, tgtIdx, delay, nlags, Examples, Timestamp );
// Divide Examples into Training Set and Testing Set
int numTrainEx = int( (2.0/3.0)*Examples.size() );
vector< vector< float > > TrainExamples(numTrainEx);
vector< string > TrainTimestamp(numTrainEx);
vector< vector< float > > TestExamples(Examples.size()-numTrainEx);
vector< string > TestTimestamp(Examples.size()-numTrainEx);
for(int i=0; i<Examples.size(); i++)
{
if(i<numTrainEx)
{
TrainTimestamp[i] = Timestamp[i];
TrainExamples[i] = Examples[i];
}
else
{
TestTimestamp[i-numTrainEx] = Timestamp[i];
TestExamples[i-numTrainEx] = Examples[i];
}
}
cout << "# Total number of examples: " << Examples.size() << endl;
cout << "# Number of training examples: " << TrainExamples.size() << endl;
cout << "# Number of testing examples: " << TestExamples.size() << endl;
cout << "# Training period: ( " << TrainTimestamp[0] << " , " << TrainTimestamp[TrainTimestamp.size()-1] << " )\n";
cout << "# Testing period: ( " << TestTimestamp[0] << " , " << TestTimestamp[TestTimestamp.size()-1] << " )\n";
vector< vector< float > > Results_Train;
vector< vector< float > > Results_Test;
cout << "# Calculating Correlations" << endl;
vector< vector< float > > ExampleCorr;
CalcCorr( nvar, tgtIdx, delay, nlags, TrainExamples, ExampleCorr);
//
// write correlations to output file
string fname = prefix + "-" + "corr_analysis.dat";
ofstream ofile( fname.c_str() );
//
for( int i=0; i<ExampleCorr.size(); i++ )
{
ofile << setw(5) << i;
for( int j=0; j<nvar; j++ )
{
ofile << setw(15) << ExampleCorr[i][j];
}
ofile << endl;
}
ofile.close();
// //
// // create un-normalized training file
// vector< double > TimestampJ;
// JulianDate( TrainTimestamp, TimestampJ );
// fname = prefix + "-" + "unorm_train.dat";
// ofile.open( fname.c_str() );
// ofile << setw(15) << TrainExamples.size() << setw(15) << TrainExamples[0].size() << endl;
// for( int i=0; i<TrainExamples.size(); i++ )
// {
// cout << setw(15) << fixed << TimestampJ[i] << std::resetiosflags(std::ios::floatfield);
// for( int j=0; j<TrainExamples[0].size(); i++ )
// {
// ofile << setw(15) << TrainExamples[i][j];
// }
// ofile << endl;
// }
// ofile.close();
// //
// // create normalized training file
// vector< vector< float > > normParam;
// NormalizeExamples( 0, TrainExamples, normParam );
// fname = prefix + "-" + "norm_train.dat";
// ofile.open( fname.c_str() );
// // write # rows, # cols
// ofile << setw(15) << TrainExamples.size() << setw(15) << TrainExamples[0].size() << endl;
// // write normalization information
// for(int i=0; i<2; i++)
// {
// for(int j=0; j<normParam[i].size(); j++)
// {
// ofile << setw(15) << normParam[i][j];
// }
// ofile << endl;
// }
// // write data
// for( int i=0; i<TrainExamples.size(); i++ )
// {
// for( int j=0; j<TrainExamples[i].size(); j++ )
// {
// ofile << setw(15) << TrainExamples[i][j];
// }
// ofile << endl;
// }
// ofile.close();
// //
// // create un-normalized testing file
// JulianDate( TestTimestamp, TimestampJ );
// fname = prefix + "-" + "unorm_test.dat";
// ofile.open( fname.c_str() );
// ofile << setw(15) << TestExamples.size() << setw(15) << TestExamples[0].size() << endl;
// for( int i=0; i<TestExamples.size(); i++ )
// {
// cout << setw(15) << fixed << TimestampJ[i] << std::resetiosflags(std::ios::floatfield);
// for( int j=0; j<TestExamples[i].size(); i++ )
// {
// ofile << setw(15) << TestExamples[i][j];
// }
// ofile << endl;
// }
// ofile.close();
// //
// // create normalized testing file
// NormalizeExamples( 1, TestExamples, normParam );
// fname = prefix + "-" + "norm_test.dat";
// ofile.open( fname.c_str() );
// ofile << setw(15) << TestExamples.size() << setw(15) << TestExamples[0].size() << endl;
// for(int i=0; i<2; i++)
// {
// for(int j=0; j<normParam[i].size(); j++)
// {
// ofile << setw(15) << normParam[i][j];
// }
// ofile << endl;
// }
// for( int i=0; i<TestExamples.size(); i++ )
// {
// for( int j=0; j<TestExamples[i].size(); j++ )
// {
// ofile << setw(15) << TestExamples[i][j];
// }
// ofile << endl;
// }
// ofile.close();
}
