Unverified Commit 23581f9a authored by Mccaskey, Alex's avatar Mccaskey, Alex Committed by GitHub
Browse files

Merge pull request #233 from tnguyen-ornl/tnguyen/add-option-to-change-default-autograd-method 

Added compiler switch to select kernel's autograd method
parents 61b96424 bb6513e0

examples/simple/gradients_optimization.cpp

+12 −3
Original line number Diff line number Diff line 
// Compile with:

// qcor gradients_optimization.cpp

// By default, kernel::autograd will use `central` gradient strategy.

// This can be changed by using the `-autograd` qcor compiler flag:

// e.g.,

// qcor -autograd forward gradients_optimization.cpp

// Options: forward, backward, central

// Note: users may need to adjust the optimizer step size to guarantee

// convergence.

__qpu__ void ansatz(qreg q, double theta) {

  X(q[0]);

  Ry(q[1], theta);
@@ -33,7 +42,7 @@ int main(int argc, char **argv) { 
  // Simple case 1: variational ansatz takes a single double

  {

    // Create the Optimizer (gradient-based)

    auto optimizer = createOptimizer("nlopt", {{"nlopt-optimizer", "l-bfgs"}});

    auto optimizer = createOptimizer("mlpack", {{"step-size", 1e-2}});

    ObjectiveFunction opt_function(

        [&](const std::vector<double> &x, std::vector<double> &dx) {

          auto q = qalloc(2);
@@ -54,7 +63,7 @@ int main(int argc, char **argv) { 
  // Simple case 2: variational ansatz takes a vector<double>

  {

    // Create the Optimizer (gradient-based)

    auto optimizer = createOptimizer("nlopt", {{"nlopt-optimizer", "l-bfgs"}});

    auto optimizer = createOptimizer("mlpack", {{"step-size", 1e-2}});

    ObjectiveFunction opt_function(

        [&](const std::vector<double> &x, std::vector<double> &dx) {

          auto q = qalloc(2);
@@ -73,7 +82,7 @@ int main(int argc, char **argv) { 
  }

  {

    // Create the Optimizer (gradient-based)

    auto optimizer = createOptimizer("nlopt", {{"nlopt-optimizer", "l-bfgs"}});

    auto optimizer = createOptimizer("mlpack", {{"step-size", 1e-2}});

    ObjectiveFunction opt_function(

        [&](const std::vector<double> &x, std::vector<double> &dx) {

          // Using kernel auto-gradient helper

handlers/token_collector/token_collector_util.cpp

+1 −1
Original line number Diff line number Diff line
@@ -296,7 +296,7 @@ std::string run_token_collector( 
          mutable_parameters.begin(),

          "__" + kernel_name + "_" + counter_str + "__compute_functor");



      auto src_code = __internal__::qcor::construct_kernel_subtype(

      auto src_code = ::__internal__::qcor::construct_kernel_subtype(

          tmpss.str(), internal_kernel_function_name, mutable_arg_types,

          mutable_parameters, bufferNames);

runtime/objectives/gradient_function.cpp

+0 −2
Original line number Diff line number Diff line
@@ -15,8 +15,6 @@ 
#include "xacc_internal_compiler.hpp"

namespace qcor {

namespace __internal__ {

std::string DEFAULT_GRADIENT_METHOD = "central";



std::shared_ptr<GradientFunction>

get_gradient_method(const std::string &type,

                    std::shared_ptr<ObjectiveFunction> obj_func,

runtime/objectives/gradient_function.hpp

+0 −1
Original line number Diff line number Diff line
@@ -39,7 +39,6 @@ public: 
};



namespace __internal__ {

extern std::string DEFAULT_GRADIENT_METHOD;

std::shared_ptr<GradientFunction>

get_gradient_method(const std::string &type,

                    std::shared_ptr<ObjectiveFunction> obj_func,

runtime/objectives/gradients/FiniteDifference/FiniteDifferenceGradients.cpp

+3 −3
Original line number Diff line number Diff line
@@ -87,7 +87,7 @@ class KernelForwardDifferenceGradient : public KernelGradientService { 
    m_kernel_eval = _kernel_eval;

    gradient_func = [&](const std::vector<double> &x,

                        double cost_val) -> std::vector<double> {

      return run_gradient_strategy(x, cost_val, "forward", m_step, obs,

      return run_gradient_strategy(x, cost_val, "forward", -m_step, obs,

                                   m_kernel_eval);

    };

  }
@@ -101,7 +101,7 @@ class KernelForwardDifferenceGradient : public KernelGradientService { 
    gradient_func = [&](const std::vector<double> &x,

                        double cost_val) -> std::vector<double> {

      auto obs = m_objFunc->get_observable();

      return run_gradient_strategy(x, cost_val, "forward", m_step, obs,

      return run_gradient_strategy(x, cost_val, "forward", -m_step, obs,

                                   m_objFunc->get_kernel_evaluator());

    };

  }
@@ -129,7 +129,7 @@ class KernelBackwardDifferenceGradient : public KernelGradientService { 
    m_kernel_eval = kernel_eval;

    gradient_func = [&](const std::vector<double> &x,

                        double cost_val) -> std::vector<double> {

      return run_gradient_strategy(x, cost_val, "forward", m_step, obs,

      return run_gradient_strategy(x, cost_val, "backward", m_step, obs,

                                   m_kernel_eval);

    };

  }