void test()
{
Stats::vector_type a(2);
a(0) = 1;
a(1) = 2;
Stats::vector_type b(2);
b(0) = 3;
b(1) = 4;
Stats::vector_type c(2);
c(0) = 5;
c(1) = 6;
Stats::data_type data;
data.push_back(a);
data.push_back(b);
data.push_back(c);
Stats stats(data);
if (os_) *os_ << "mean: " << stats.mean() << endl;
if (os_) *os_ << "covariance: " << stats.covariance() << endl;
// mean & covariance computed using good old-fashioned reckoning
Stats::vector_type mean(2);
mean(0) = 3;
mean(1) = 4;
Stats::matrix_type covariance(2,2);
covariance(0,0) = covariance(0,1) = covariance(1,0) = covariance(1,1) = 8/3.;
// verify results
const double epsilon = 1e-12;
unit_assert_vectors_equal(stats.mean(), mean, epsilon);
unit_assert_matrices_equal(stats.covariance(), covariance, epsilon);
double rms0_good = sqrt(35./3);
double rms1_good = sqrt(56./3);
double rms0_test = sqrt(stats.meanOuterProduct()(0,0));
double rms1_test = sqrt(stats.meanOuterProduct()(1,1));
unit_assert_equal(rms0_test, rms0_good, epsilon);
unit_assert_equal(rms1_test, rms1_good, epsilon);
}
void Stats::Impl::computeSums(const Stats::data_type& data)
{
if (data.size()>0) D_ = data[0].size();
sumData_ = Stats::vector_type(D_);
sumOuterProducts_ = Stats::matrix_type(D_, D_);
sumData_.clear();
sumOuterProducts_.clear();
for (Stats::data_type::const_iterator it=data.begin(); it!=data.end(); ++it)
{
if (it->size() != D_)
{
ostringstream message;
message << "[Stats::Impl::computeSums()] " << D_ << "-dimensional data expected: " << *it;
throw runtime_error(message.str());
}
sumData_ += *it;
sumOuterProducts_ += outer_prod(*it, *it);
}
}
void processTLP(const TruncatedLorentzianParameters& tlp, int trialCount, bool verbose)
{
if (verbose)
cout << tlp << "\n\n\n";
else
cout << columnHeader_ << endl;
for (double snr=2; snr<=10; snr+=.5)
for (double gridOffset=-.5; gridOffset<=.5; gridOffset+=.1)
/*
double snr = 100;
double gridOffset = 0;
*/
{
if (verbose)
{
cout << fixed;
cout << "signal-to-noise ratio: " << snr << endl;
cout << "relative grid offset: " << gridOffset << endl;
}
Stats::data_type data;
double crBound = 0;
for (int i=0; i<trialCount; i++)
{
try
{
Stats::vector_type result = runTrial(tlp, snr, gridOffset, crBound);
data.push_back(result);
}
catch (exception& e)
{
cerr << e.what() << endl;
cerr << "Caught exception, continuing.\n";
}
}
if (verbose)
{
cout << endl <<
setw(13) << "err_f0_pred" <<
setw(14) << "err_f0" <<
setw(14) << "err_tau" <<
setw(14) << "err_magnitude" <<
setw(14) << "err_phase" <<
setw(14) << "model_error" <<
setw(14) << "z-score" <<
endl;
for (Stats::data_type::iterator it=data.begin(); it!=data.end(); ++it)
outputVector(*it);
}
Stats stats(data);
double mean_f0 = stats.mean()(F0);
double variance_f0 = stats.covariance()(F0,F0);
double expectedF0Squared = stats.meanOuterProduct()(F0,F0);
if (verbose)
{
outputStats(stats);
cout << endl;
cout << fixed << noshowpos;
cout << "<err_f0^2>: " << expectedF0Squared << endl;
cout << "Cramer-Rao bound: " << crBound << endl << endl;
cout << columnHeader_ << endl;
}
cout << noshowpos << fixed << setprecision(3)
<< setw(7) << snr << " " <<
setw(7) << gridOffset << " " <<
setprecision(10) <<
setw(14) << mean_f0 << " " <<
setw(14) << variance_f0 << " " <<
setw(14) << sqrt(variance_f0) << " " <<
setw(14) << expectedF0Squared << " " <<
setw(14) << crBound << endl;
if (verbose)
cout << "\n\n";
}
}
Stats::Impl::Impl(const Stats::data_type& data)
: D_(0),
N_(data.size())
{
computeSums(data);
}