// Analysis_TI::Setup()
Analysis::RetType Analysis_TI::Setup(ArgList& analyzeArgs, AnalysisSetup& setup, int debugIn)
{
debug_ = debugIn;
int nq = analyzeArgs.getKeyInt("nq", 0);
ArgList nskipArg(analyzeArgs.GetStringKey("nskip"), ","); // Comma-separated
avg_increment_ = analyzeArgs.getKeyInt("avgincrement", -1);
avg_max_ = analyzeArgs.getKeyInt("avgmax", -1);
avg_skip_ = analyzeArgs.getKeyInt("avgskip", 0);
n_bootstrap_pts_ = analyzeArgs.getKeyInt("bs_pts", -1);
n_bootstrap_samples_ = analyzeArgs.getKeyInt("bs_samples", 0);
bootstrap_seed_ = analyzeArgs.getKeyInt("bs_seed", -1);
bootstrap_fac_ = analyzeArgs.getKeyDouble("bs_fac", 0.75);
if (!nskipArg.empty()) {
avgType_ = SKIP;
// Specified numbers of points to skip
nskip_.clear();
for (int i = 0; i != nskipArg.Nargs(); i++) {
nskip_.push_back( nskipArg.getNextInteger(0) );
if (nskip_.back() < 0) nskip_.back() = 0;
}
} else if (avg_increment_ > 0)
avgType_ = INCREMENT;
else if (n_bootstrap_samples_ > 0)
avgType_ = BOOTSTRAP;
else
avgType_ = AVG;
masterDSL_ = setup.DslPtr();
// Get lambda values
ArgList xArgs(analyzeArgs.GetStringKey("xvals"), ","); // Also comma-separated
if (!xArgs.empty()) {
xval_.clear();
for (int i = 0; i != xArgs.Nargs(); i++)
xval_.push_back( xArgs.getNextDouble(0.0) );
}
std::string setname = analyzeArgs.GetStringKey("name");
DataFile* outfile = setup.DFL().AddDataFile(analyzeArgs.GetStringKey("out"), analyzeArgs);
curveout_ = setup.DFL().AddDataFile(analyzeArgs.GetStringKey("curveout"), analyzeArgs);
// Select datasets from remaining args
if (input_dsets_.AddSetsFromArgs( analyzeArgs.RemainingArgs(), setup.DSL() )) {
mprinterr("Error: Could not add data sets.\n");
return Analysis::ERR;
}
if (input_dsets_.empty()) {
mprinterr("Error: No input data sets.\n");
return Analysis::ERR;
}
if (SetQuadAndWeights(nq)) return Analysis::ERR;
// Determine integration mode
if (nq > 0)
mode_ = GAUSSIAN_QUAD;
else
mode_ = TRAPEZOID;
// Check that # abscissas matches # data sets
if (xval_.size() != input_dsets_.size()) {
mprinterr("Error: Expected %zu data sets for integration, got %zu\n",
input_dsets_.size(), xval_.size());
return Analysis::ERR;
}
// Set up output data sets
DataSet::DataType dtype = DataSet::DOUBLE;
if (avgType_ == SKIP || avgType_ == INCREMENT)
dtype = DataSet::XYMESH;
dAout_ = setup.DSL().AddSet(dtype, setname, "TI");
if (dAout_ == 0) return Analysis::ERR;
if (outfile != 0) outfile->AddDataSet( dAout_ );
MetaData md(dAout_->Meta().Name(), "TIcurve");
if (avgType_ == AVG) {
// Single curve
curve_.push_back( setup.DSL().AddSet(DataSet::XYMESH, md) );
if (curve_.back() == 0) return Analysis::ERR;
curve_.back()->ModifyDim(Dimension::X).SetLabel("Lambda");
if (curveout_ != 0) curveout_->AddDataSet( curve_.back() );
if (outfile != 0) outfile->ProcessArgs("noxcol");
} else if (avgType_ == SKIP) {
// As many curves as skip values
for (Iarray::const_iterator it = nskip_.begin(); it != nskip_.end(); ++it) {
md.SetIdx( *it );
DataSet* ds = setup.DSL().AddSet(DataSet::XYMESH, md);
if (ds == 0) return Analysis::ERR;
ds->ModifyDim(Dimension::X).SetLabel("Lambda");
ds->SetLegend( md.Name() + "_Skip" + integerToString(*it) );
if (curveout_ != 0) curveout_->AddDataSet( ds );
curve_.push_back( ds );
}
} else if (avgType_ == BOOTSTRAP) {
// As many curves as resamples
for (int nsample = 0; nsample != n_bootstrap_samples_; nsample++) {
md.SetIdx(nsample);
DataSet* ds = setup.DSL().AddSet(DataSet::XYMESH, md);
if (ds == 0) return Analysis::ERR;
ds->ModifyDim(Dimension::X).SetLabel("Lambda");
ds->SetLegend( md.Name() + "_Sample" + integerToString(nsample) );
if (curveout_ != 0) curveout_->AddDataSet( ds );
curve_.push_back( ds );
}
// Standard devation of avg free energy over samples
dA_SD_ = setup.DSL().AddSet(DataSet::DOUBLE, MetaData(md.Name(), "SD"));
if (dA_SD_ == 0) return Analysis::ERR;
if (outfile != 0) {
outfile->AddDataSet( dA_SD_ );
outfile->ProcessArgs("noxcol");
}
}
// NOTE: INCREMENT is set up once data set size is known
mprintf(" TI: Calculating TI");
if (mode_ == GAUSSIAN_QUAD) {
mprintf(" using Gaussian quadrature with %zu points.\n", xval_.size());
mprintf("\t%6s %8s %8s %s\n", "Point", "Abscissa", "Weight", "SetName");
for (unsigned int i = 0; i != xval_.size(); i++)
mprintf("\t%6i %8.5f %8.5f %s\n", i, xval_[i], wgt_[i], input_dsets_[i]->legend());
} else {
mprintf(" using the trapezoid rule.\n");
mprintf("\t%6s %8s %s\n", "Point", "Abscissa", "SetName");
for (unsigned int i = 0; i != xval_.size(); i++)
mprintf("\t%6i %8.5f %s\n", i, xval_[i], input_dsets_[i]->legend());
}
mprintf("\tResult(s) of integration(s) saved in set '%s'\n", dAout_->legend());
if (avgType_ == AVG)
mprintf("\tUsing all data points in <DV/DL> calc.\n");
else if (avgType_ == SKIP) {
mprintf("\tSkipping first");
for (Iarray::const_iterator it = nskip_.begin(); it != nskip_.end(); ++it)
mprintf(" %i", *it);
mprintf(" data points for <DV/DL> calc.\n");
} else if (avgType_ == INCREMENT) {
mprintf("\tCalculating <DV/DL> starting from point %i, increment by %i.",
avg_skip_, avg_increment_);
if (avg_max_ != -1)
mprintf(" Max %i points.", avg_max_);
mprintf("\n");
} else if (avgType_ == BOOTSTRAP) {
mprintf("\tStandard devation of result stored in set '%s'\n", dA_SD_->legend());
mprintf("\tCalculating <DV/DL> from %i bootstrap resamples.\n", n_bootstrap_samples_);
if (n_bootstrap_pts_ > 0)
mprintf("\tBootstrap resample size is %i data points.\n", n_bootstrap_pts_);
else
mprintf("\tWill use bootstrap resample size of %g%% of total points.\n",
bootstrap_fac_*100.0);
if (bootstrap_seed_ != -1)
mprintf("\tBoostrap base seed is %i\n", bootstrap_seed_);
}
mprintf("\tTI curve(s) saved in set(s)");
if (avgType_ != INCREMENT)
for (DSarray::const_iterator ds = curve_.begin(); ds != curve_.end(); ++ds)
mprintf(" '%s'", (*ds)->legend());
else
mprintf(" named '%s'", md.PrintName().c_str());
mprintf("\n");
if (outfile != 0) mprintf("\tResults written to '%s'\n", outfile->DataFilename().full());
if (curveout_!= 0) mprintf("\tTI curve(s) written to '%s'\n", curveout_->DataFilename().full());
return Analysis::OK;
}
// Analysis_TI::Calc_Increment()
int Analysis_TI::Calc_Increment() {
// Determine max points if not given.
int maxpts = avg_max_;
if (maxpts == -1) {
for (unsigned int idx = 0; idx != input_dsets_.size(); idx++) {
DataSet_1D const& ds = static_cast<DataSet_1D const&>( *(input_dsets_[idx]) );
if (maxpts == -1)
maxpts = (int)ds.Size();
else if (maxpts != (int)ds.Size()) {
mprintf("Warning: # points in '%s' (%zu) is different than %i.\n",
ds.legend(), ds.Size(), maxpts);
maxpts = std::min( maxpts, (int)ds.Size() );
mprintf("Warning: Will only use %i points.\n", maxpts);
}
}
}
if (maxpts < 1) {
mprinterr("Error: Max points to use is < 1.\n");
return 1;
}
if (avg_skip_ >= maxpts) {
mprinterr("Error: 'avgskip' (%i) > max (%i).\n", avg_skip_, maxpts);
return 1;
}
// sum: Hold the results of integration for each curve (increment)
Darray sum;
// points: Hold point values at which each avg is being calculated
Iarray points;
// Loop over input data sets.
for (unsigned int idx = 0; idx != input_dsets_.size(); idx++) {
DataSet_1D const& ds = static_cast<DataSet_1D const&>( *(input_dsets_[idx]) );
if (CheckSet(ds)) return 1;
// Calculate averages for each increment
Darray avg;
Iarray increments;
int count = 0;
int endpt = maxpts -1;
double currentSum = 0.0;
if (debug_ > 0) mprintf("DEBUG: Lambda %g\n", xval_[idx]);
for (int pt = avg_skip_; pt != maxpts; pt++)
{
currentSum += ds.Dval(pt);
count++;
if (count == avg_increment_ || pt == endpt) {
avg.push_back( currentSum / ((double)(pt - avg_skip_ + 1)) );
increments.push_back(pt+1);
if (debug_ > 0)
mprintf("DEBUG:\t\tAvg from %i to %i: %g\n", avg_skip_+1, pt+1, avg.back());
count = 0;
}
}
if (sum.empty()) {
sum.resize(avg.size());
points = increments;
} else if (sum.size() != avg.size()) {
mprinterr("Error: Different # of increments for set '%s'; got %zu, expected %zu.\n",
ds.legend(), avg.size(), sum.size());
return 1;
}
// Create increment curve data sets
if (curve_.empty()) {
MetaData md(dAout_->Meta().Name(), "TIcurve");
for (unsigned int j = 0; j != avg.size(); j++) {
md.SetIdx( increments[j] );
DataSet* ds = masterDSL_->AddSet(DataSet::XYMESH, md);
if (ds == 0) return Analysis::ERR;
ds->ModifyDim(Dimension::X).SetLabel("Lambda");
ds->SetLegend( md.Name() + "_Skip" + integerToString(increments[j]) );
if (curveout_ != 0) curveout_->AddDataSet( ds );
curve_.push_back( ds );
}
}
for (unsigned int j = 0; j != avg.size(); j++) {
DataSet_Mesh& CR = static_cast<DataSet_Mesh&>( *(curve_[j]) );
CR.AddXY(xval_[idx], avg[j]);
if (mode_ == GAUSSIAN_QUAD)
sum[j] += (wgt_[idx] * avg[j]);
}
} // END loop over data sets
if (mode_ == TRAPEZOID) Integrate_Trapezoid(sum);
// Store final integration values
DataSet_Mesh& DA = static_cast<DataSet_Mesh&>( *dAout_ );
DA.ModifyDim(Dimension::X).SetLabel("Point");
for (unsigned int j = 0; j != points.size(); j++)
DA.AddXY(points[j], sum[j]);
return 0;
}
