GradientTest Class Reference

Collaboration diagram for GradientTest:

Public Types
using	Return_t = QMCTraits::RealType

Public Member Functions
	GradientTest (GradientTestInput &&input)
void	run (QMCCostFunctionBase &costFunc, const std::string &root_name)

Private Attributes
GradientTestInput	input_
std::ofstream	param_deriv_file_
bool	first_
int	param_deriv_index_

Detailed Description

Definition at line 52 of file GradientTest.h.

Member Typedef Documentation

Return_t

using Return_t = QMCTraits::RealType

Definition at line 55 of file GradientTest.h.

Constructor & Destructor Documentation

GradientTest()

GradientTest ( GradientTestInput &&  input )

inline

Definition at line 57 of file GradientTest.h.

: input_(std::move(input)), first_(true), param_deriv_index_(0) {}

Member Function Documentation

run()

void run	(	QMCCostFunctionBase &	costFunc,
		const std::string &	root_name
	)

Definition at line 19 of file GradientTest.cpp.

References qmcplusplus::abs(), qmcplusplus::app_log(), GradientTestInput::do_param_output(), qmcplusplus::Units::charge::e, GradientTest::first_, GradientTestInput::get_finite_diff_delta(), QMCCostFunctionBase::getNumParams(), QMCCostFunctionBase::getParamName(), QMCCostFunctionBase::GradCost(), GradientTest::input_, GradientTest::param_deriv_file_, GradientTest::param_deriv_index_, and QMCCostFunctionBase::Params().

{
  int num_params = costFunc.getNumParams();
  std::vector<Return_t> numeric_grad(num_params);
  std::vector<Return_t> params(num_params);
  std::vector<Return_t> analytic_grad(num_params);

  for (int i = 0; i < num_params; i++)
  {
    params[i] = std::real(costFunc.Params(i));
  }


  if (input_.do_param_output() && first_)
  {
    int namelen = root_name.length();
    // Assume that the root_name has the series suffix (".s000").
    // Remove the series suffix to get the project id
    std::string fname = root_name.substr(0, namelen - 5) + ".param.s000.scalar.dat";

    param_deriv_file_.open(fname);
    param_deriv_file_ << "# Index ";
    for (int i = 0; i < num_params; i++)
    {
      param_deriv_file_ << " " << costFunc.getParamName(i);
    }
    param_deriv_file_ << std::endl;
    first_ = false;
  }

  // Numerical gradient
  costFunc.GradCost(numeric_grad, params, input_.get_finite_diff_delta());

  // Analytic gradient
  costFunc.GradCost(analytic_grad, params);


  if (input_.do_param_output())
  {
    param_deriv_file_ << param_deriv_index_ << " ";
    param_deriv_index_++;
  }

  std::string_view param_name_header("Param_Name");
  size_t max_name_len = param_name_header.size();
  for (int k = 0; k < num_params; k++)
  {
    std::string vname = costFunc.getParamName(k);
    max_name_len      = std::max(vname.size(), max_name_len);
  }
  max_name_len += 2; // add some padding

  // clang-format off
  app_log() << std::setw(max_name_len) << std::left << param_name_header
            << std::setw(14) << std::right << " Value "
            << std::setw(20) << std::right << " Numeric "
            << std::setw(20) << std::right << " Analytic "
            << std::setw(14) << " Percent" << std::endl;
  // clang-format on

  for (int k = 0; k < num_params; k++)
  {
    std::string vname = costFunc.getParamName(k);
    std::ostringstream rel_diff_str;
    std::string over_threshold;
    if (numeric_grad[k] != 0)
    {
      double rel_diff_percent = 100 * (numeric_grad[k] - analytic_grad[k]) / numeric_grad[k];
      rel_diff_str << std::scientific << std::setprecision(2) << rel_diff_percent;
      // Highlight problematic differences
      // The thresholds are arbitrary.
      if (std::abs(rel_diff_percent) > 1e-3)
        over_threshold = " !";
      if (std::abs(rel_diff_percent) > 1e-2)
        over_threshold = " !!";
      if (std::abs(rel_diff_percent) > 1e-1)
        over_threshold = " !!!";
    }
    else
    {
      rel_diff_str << "inf";
      over_threshold = " !!!";
    }


    // clang-format off
    app_log() << std::setw(max_name_len) << std::left << vname
              << std::setprecision(6)  << std::setw(14) << std::right << params[k]
              << std::setprecision(10) << std::setw(20) << std::right << numeric_grad[k]
              << std::setw(20) << std::right << analytic_grad[k]
              << std::setw(14) << std::right << rel_diff_str.str() << over_threshold << std::endl;
    // clang-format on

    if (input_.do_param_output())
      param_deriv_file_ << std::setprecision(10) << analytic_grad[k] << " ";
  }

  // Reset precision to the default
  app_log() << std::setprecision(6);

  if (input_.do_param_output())
    param_deriv_file_ << std::endl;

  app_log() << std::endl;
}

Member Data Documentation

first_

bool first_

private

Definition at line 66 of file GradientTest.h.

Referenced by GradientTest::run().

input_

GradientTestInput input_

private

Definition at line 63 of file GradientTest.h.

Referenced by GradientTest::run().

param_deriv_file_

std::ofstream param_deriv_file_

private

Definition at line 65 of file GradientTest.h.

Referenced by GradientTest::run().

param_deriv_index_

int param_deriv_index_

private

Definition at line 67 of file GradientTest.h.

Referenced by GradientTest::run().

---

The documentation for this class was generated from the following files:

• /home/pk7/projects/qmc/for_cron_doxygen/qmcpack/src/QMCDrivers/WFOpt/GradientTest.h
• /home/pk7/projects/qmc/for_cron_doxygen/qmcpack/src/QMCDrivers/WFOpt/GradientTest.cpp