diff --git a/optimization/optimizers/cma_es/CMakeLists.txt b/optimization/optimizers/cma_es/CMakeLists.txt
index 1eefff0d475009deecd255459725582ed4d99b8e..4e4b2650bb313760299289a6e3c82c59a5b0742f 100644
--- a/optimization/optimizers/cma_es/CMakeLists.txt
+++ b/optimization/optimizers/cma_es/CMakeLists.txt
@@ -3,7 +3,5 @@ add_executable(cma_es
)
target_link_libraries(cma_es optimization_lib cma-es clara)
-if (VISUALIZE)
- make_plot_target(cma_es 2d ARGS -d 2)
- make_plot_target(cma_es 10d ARGS -d 10 -n 10000)
-endif()
+make_plot_target(cma_es 2d ARGS -d 2)
+make_plot_target(cma_es 10d ARGS -d 10 -n 10000)
diff --git a/optimization/optimizers/cma_es/cma_es.h b/optimization/optimizers/cma_es/cma_es.h
index 8fb5e6fc0a4924124d84394f22d6e59caba1549c..dbf8f123e84e00ee4a203991a6543d05579cc9f9 100644
--- a/optimization/optimizers/cma_es/cma_es.h
+++ b/optimization/optimizers/cma_es/cma_es.h
@@ -1,50 +1,70 @@
#ifndef OPTIMIZATION_CMA_ES_H
#define OPTIMIZATION_CMA_ES_H
-#include "cma_es_log.h"
-#include "utils/vector.h"
#include "cmaes_interface.h"
+#include "logs/nothing.h"
+#include "utils/vector.h"
#include <iostream>
-#include <iomanip>
namespace optimization {
//--------------------------------------------------------------------------------------------------
-class CmaEs : public CmaEsLog {
+class CmaEs {
public:
- CmaEs() {}
+ CmaEs(): points_(nullptr), values_(nullptr) {}
CmaEs(uint32_t d, uint32_t ps, int32_t s, uint32_t me, uint32_t mi, Scalar ryt) :
dim_(d),
pop_size_(ps),
seed_(s),
max_evaulations_(me),
max_iterations_(mi),
- rel_y_tol_(ryt)
+ rel_y_tol_(ryt),
+ points_(nullptr),
+ values_(nullptr)
{}
-
- uint32_t n_evaluations() const { return n_evaluations_; }
- uint32_t n_iterations() const { return n_iterations_; }
+ ~CmaEs() { if (points_ != nullptr) { cmaes_exit(&cma_es_); } }
uint32_t dim() const { return dim_; }
uint32_t& dim() { return dim_; }
uint32_t pop_size() const { return pop_size_; }
uint32_t& pop_size() { return pop_size_; }
+ uint32_t n_evaluations() { return cmaes_Get(&cma_es_, "eval"); }
+ uint32_t n_iterations() const { return n_iterations_; }
+ // TODO See if I can make these functions const by modifying some CMA-ES guts.
+ inline Eigen::Map<const Eigen::MatrixXd> point();
+ Scalar value() { return cmaes_Get(&cma_es_, "fbestever"); }
+
template <typename Objective>
- VectorXs optimize(
+ Eigen::Map<const Eigen::MatrixXd> optimize(
Objective& objective,
VectorXs const& initial_point,
Scalar initial_std_dev = 0.3
);
+ template <typename Objective, typename Log>
+ Eigen::Map<const Eigen::MatrixXd> optimize(
+ Objective& objective,
+ VectorXs const& initial_point,
+ Scalar initial_std_dev,
+ Log& log
+ );
template <typename Objective>
- VectorXs optimize(
+ Eigen::Map<const Eigen::MatrixXd> optimize(
Objective& objective,
VectorXs const& initial_point,
VectorXs const& initial_std_dev
);
+ template <typename Objective, typename Log>
+ Eigen::Map<const Eigen::MatrixXd> optimize(
+ Objective& objective,
+ VectorXs const& initial_point,
+ VectorXs const& initial_std_dev,
+ Log& log
+ );
private:
+ // hyperparameters
uint32_t dim_;
uint32_t pop_size_;
int32_t seed_;
@@ -52,90 +72,119 @@ private:
uint32_t max_iterations_;
Scalar rel_y_tol_;
- uint32_t n_evaluations_;
+ // algorithm state
uint32_t n_iterations_;
+ cmaes_t cma_es_;
+ // These are used to communicate with the CMA-ES object. They also help with memory management,
+ // in that cma_es_ has memory that eventually needs to be freed if and only if points_ and
+ // values_ are nullptr.
+ double* const* points_; // points to cma_es_'s population points when we're optimizing
+ double* values_; // points to cma_es_'s population values when we're optimizing
};
+//..................................................................................................
+Eigen::Map<const Eigen::MatrixXd> CmaEs::point() {
+ double const* best_point = cmaes_GetPtr(&cma_es_, "xbestever");
+ return Eigen::Map<const Eigen::MatrixXd>(best_point, dim_, 1);
+}
+
//..................................................................................................
template <typename Objective>
-VectorXs CmaEs::optimize(
+Eigen::Map<const Eigen::MatrixXd> CmaEs::optimize(
Objective& objective,
VectorXs const& initial_point,
Scalar initial_std_dev
+) {
+ CmaEsLogNothing log;
+ return optimize(objective, initial_point, initial_std_dev, log);
+}
+
+//..................................................................................................
+template <typename Objective, typename Log>
+Eigen::Map<const Eigen::MatrixXd> CmaEs::optimize(
+ Objective& objective,
+ VectorXs const& initial_point,
+ Scalar initial_std_dev,
+ Log& log
) {
VectorXs std_dev_vector;
std_dev_vector.resize(initial_point.size());
std_dev_vector.fill(initial_std_dev);
- return optimize(objective, initial_point, std_dev_vector);
+ return optimize(objective, initial_point, std_dev_vector, log);
}
//..................................................................................................
template <typename Objective>
-VectorXs CmaEs::optimize(
+Eigen::Map<const Eigen::MatrixXd> CmaEs::optimize(
Objective& objective,
VectorXs const& initial_point,
VectorXs const& initial_std_dev
) {
- n_evaluations_ = 0;
+ CmaEsLogNothing log;
+ return optimize(objective, initial_point, initial_std_dev, log);
+}
+
+//..................................................................................................
+template <typename Objective, typename Log>
+Eigen::Map<const Eigen::MatrixXd> CmaEs::optimize(
+ Objective& objective,
+ VectorXs const& initial_point,
+ VectorXs const& initial_std_dev,
+ Log& log
+) {
n_iterations_ = 0;
+ if (points_ != nullptr) {
+ cmaes_exit(&cma_es_);
+ }
- // Decalare the CMA-ES object, and the arrays we will use to communicate with it.
- cmaes_t cma_es;
- double* const* points;
- double* values;
-
- // For some reason cmaes_init_para takes double*, not double const*, so I have to copy these.
- // TODO: Looking at Hansen's code, there's no reason they aren't const. I should submit a PR.
- VectorXs point = initial_point;
- VectorXs std_dev = initial_std_dev;
-
- cmaes_init_para(
- &cma_es,
- dim_,
- point.data(),
- std_dev.data(),
- seed_,
- pop_size_,
- "none"
- );
- cma_es.sp.stopMaxFunEvals = max_evaulations_;
- cma_es.sp.stopMaxIter = max_iterations_;
- cma_es.sp.stopTolFun = rel_y_tol_;
- values = cmaes_init_final(&cma_es);
+ // This is used for various copy operations. Shouldn't be necessary.
+ VectorXs point_vec = initial_point;
+ {
+ // For some reason cmaes_init_para takes double*, not double const*, so I have to copy these.
+ // TODO: Looking at Hansen's code, there's no reason they aren't const. I should submit a PR.
+ VectorXs std_dev = initial_std_dev;
+
+ cmaes_init_para(
+ &cma_es_,
+ dim_,
+ point_vec.data(),
+ std_dev.data(),
+ seed_,
+ pop_size_,
+ "none"
+ );
+ }
+ cma_es_.sp.stopMaxFunEvals = max_evaulations_;
+ cma_es_.sp.stopMaxIter = max_iterations_;
+ cma_es_.sp.stopTolFun = rel_y_tol_;
+ values_ = cmaes_init_final(&cma_es_);
- CmaEsLog::initialize(objective);
- std::cout << cmaes_SayHello(&cma_es) << '\n';
+ log.initialize(objective);
+ std::cout << cmaes_SayHello(&cma_es_) << '\n';
- while(!cmaes_TestForTermination(&cma_es)) {
+ while (!cmaes_TestForTermination(&cma_es_)) {
// Get a generation of points.
- points = cmaes_SamplePopulation(&cma_es);
+ points_ = cmaes_SamplePopulation(&cma_es_);
// TODO: This wants to be parallelized.
for (uint32_t i = 0; i < pop_size_; ++i) {
- auto point_map = Eigen::Map<const Eigen::MatrixXd>(points[i], dim_, 1);
+ auto point_map = Eigen::Map<const Eigen::MatrixXd>(points_[i], dim_, 1);
// TODO: make objectives handle Maps so we don't have to copy the data.
- point = point_map;
- objective.eval(point, values[i]);
+ point_vec = point_map;
+ objective.eval(point_vec, values_[i]);
}
- CmaEsLog::push_back(points, values, dim_, pop_size_);
+ log.push_back(points_, values_, dim_, pop_size_);
// Update the search distribution used for cmaes_SamplePopulation().
- cmaes_UpdateDistribution(&cma_es, values);
+ cmaes_UpdateDistribution(&cma_es_, values_);
++n_iterations_;
}
- std::cout << cmaes_TestForTermination(&cma_es) << '\n';
- //cmaes_WriteToFile(&cma_es, "all", "allcmaes.dat"); // write final results
+ std::cout << cmaes_TestForTermination(&cma_es_) << '\n';
// Get best estimator for the optimum, xmean.
- double const* final_point_data = cmaes_GetPtr(&cma_es, "xmean");
- auto point_map = Eigen::Map<const Eigen::MatrixXd>(final_point_data, dim_, 1);
- point = point_map;
- n_evaluations_ = cmaes_Get(&cma_es, "eval");
- cmaes_exit(&cma_es);
-
- return point;
+ return point();
}
}
diff --git a/optimization/optimizers/cma_es/cma_es_log.h b/optimization/optimizers/cma_es/logs/everything.h
similarity index 64%
rename from optimization/optimizers/cma_es/cma_es_log.h
rename to optimization/optimizers/cma_es/logs/everything.h
index 00aff539f390fe47689ff08675512c71095065de..de81575163a64c9d3ad2fe872c789466f4e79c1c 100644
--- a/optimization/optimizers/cma_es/cma_es_log.h
+++ b/optimization/optimizers/cma_es/logs/everything.h
@@ -1,15 +1,10 @@
-#ifndef OPTIMIZATION_CMA_ES_LOG_H
-#define OPTIMIZATION_CMA_ES_LOG_H
+#ifndef OPTIMIZATION_CMA_ES_LOGS_EVERYTHING_H
+#define OPTIMIZATION_CMA_ES_LOGS_EVERYTHING_H
-#include "utils/vector.h"
#include "utils/matrix.h"
-#include "utils/vis_only.h"
-
-#ifdef VISUALIZE
-#include <vector>
-#include "json.hpp"
+#include "utils/vector.h"
#include "utils/eigen_json.h"
-#endif
+#include <vector>
namespace optimization {
@@ -22,41 +17,24 @@ struct CmaEsState {
};
//--------------------------------------------------------------------------------------------------
-// This is used as a base class rather than a member so that the empty base class optimization can
-// be applied (the member would take up space even if it is an empty class).
-struct CmaEsLog {
- inline void reserve(uint32_t n) VIS_ONLY_METHOD;
+struct CmaEsLogEverything {
+ inline void reserve(uint32_t n) { states.reserve(n); }
template <typename Objective>
- inline void initialize(Objective const&) VIS_ONLY_METHOD;
+ inline void initialize(Objective const&) { objective_name = Objective::name; }
inline void push_back(
double const* const* points,
double const* values,
uint32_t dim,
uint32_t pop_size
- ) VIS_ONLY_METHOD;
- inline void clear() VIS_ONLY_METHOD;
+ );
+ inline void clear();
- #ifdef VISUALIZE
std::string objective_name;
std::vector<CmaEsState> states;
- #endif
};
-#ifdef VISUALIZE
-
//..................................................................................................
-void CmaEsLog::reserve(uint32_t n) {
- states.reserve(n);
-}
-
-//..................................................................................................
-template <typename Objective>
-void CmaEsLog::initialize(Objective const&) {
- objective_name = Objective::name;
-}
-
-//..................................................................................................
-void CmaEsLog::push_back(
+void CmaEsLogEverything::push_back(
double const* const* points,
double const* values,
uint32_t dim,
@@ -74,7 +52,8 @@ void CmaEsLog::push_back(
}
//..................................................................................................
-void CmaEsLog::clear() {
+void CmaEsLogEverything::clear() {
+ objective_name.clear();
states.clear();
}
@@ -87,7 +66,7 @@ inline void to_json(nlohmann::json& j, CmaEsState const& state) {
}
//--------------------------------------------------------------------------------------------------
-inline void to_json(nlohmann::json& j, CmaEsLog const& log) {
+inline void to_json(nlohmann::json& j, CmaEsLogEverything const& log) {
j = nlohmann::json{
{"algorithm", "gradient descent"},
{"objective", log.objective_name},
@@ -95,8 +74,6 @@ inline void to_json(nlohmann::json& j, CmaEsLog const& log) {
};
}
-#endif
-
}
#endif
diff --git a/optimization/optimizers/cma_es/cma_es_vis.h b/optimization/optimizers/cma_es/logs/everything_vis.h
similarity index 88%
rename from optimization/optimizers/cma_es/cma_es_vis.h
rename to optimization/optimizers/cma_es/logs/everything_vis.h
index 7dee87080796c035ce425d04386b46ac496fafbf..84fc55958d5b4f3696f6f45a3b0de0b7fa40e315 100644
--- a/optimization/optimizers/cma_es/cma_es_vis.h
+++ b/optimization/optimizers/cma_es/logs/everything_vis.h
@@ -1,15 +1,13 @@
-#ifndef OPTIMIZATION_CMA_ES_VIS_H
-#define OPTIMIZATION_CMA_ES_VIS_H
+#ifndef OPTIMIZATION_CMA_ES_LOGS_EVERYTHING_VIS_H
+#define OPTIMIZATION_CMA_ES_LOGS_EVERYTHING_VIS_H
-#include <iostream>
-#include "cma_es_log.h"
-#include "utils/eigen_json.h"
+#include "everything.h"
namespace optimization {
//--------------------------------------------------------------------------------------------------
struct CmaEsVis {
- CmaEsLog const& log;
+ CmaEsLogEverything const& log;
};
//..................................................................................................
diff --git a/optimization/optimizers/cma_es/logs/nothing.h b/optimization/optimizers/cma_es/logs/nothing.h
new file mode 100644
index 0000000000000000000000000000000000000000..cb6d1600646fd377b85bfb61963bfef7da860e61
--- /dev/null
+++ b/optimization/optimizers/cma_es/logs/nothing.h
@@ -0,0 +1,19 @@
+#ifndef OPTIMIZATION_CMA_ES_LOGS_NOTHING_H
+#define OPTIMIZATION_CMA_ES_LOGS_NOTHING_H
+
+#include <cstdint>
+
+namespace optimization {
+
+//--------------------------------------------------------------------------------------------------
+struct CmaEsLogNothing {
+ inline void reserve(uint32_t) {}
+ template <typename Objective>
+ inline void initialize(Objective const&) {}
+ inline void push_back(double const* const*, double const*, uint32_t, uint32_t) {}
+ inline void clear() {}
+};
+
+}
+
+#endif
diff --git a/optimization/optimizers/cma_es/main.cpp b/optimization/optimizers/cma_es/main.cpp
index 8148c9f4700b536a0e49a9bce23367458fc4bcaf..eb159394daf3b5f522a7db99e22409d01cdf7cd7 100644
--- a/optimization/optimizers/cma_es/main.cpp
+++ b/optimization/optimizers/cma_es/main.cpp
@@ -2,17 +2,12 @@
#include "cma_es.h"
#include "objectives/paraboloid.h"
#include "objectives/rosenbrock.h"
-#include <iostream>
-
-#ifdef VISUALIZE
-#include "cma_es_vis.h"
-#include "utils/eigen_json.h"
+#include "logs/everything_vis.h"
#include <fstream>
-
-using json = nlohmann::json;
-#endif
+#include <iostream>
using namespace optimization;
+using json = nlohmann::json;
//--------------------------------------------------------------------------------------------------
int main(int const argc, char const** argv) {
@@ -65,24 +60,29 @@ int main(int const argc, char const** argv) {
objective.dim() = dim;
CmaEs optimizer(dim, pop_size, seed, max_evaluations, max_iterations, rel_y_tol);
- VectorXs minimum = optimizer.optimize(objective, initial_point, initial_std_dev);
+ CmaEsLogEverything log;
+
+ if (log_file_path.empty() && vis_file_path.empty()) {
+ optimizer.optimize(objective, initial_point, initial_std_dev);
+ } else {
+ optimizer.optimize(objective, initial_point, initial_std_dev, log);
+ }
+
std::cout << "n evaluations: " << optimizer.n_evaluations() << '\n';
std::cout << "n generations: " << optimizer.n_iterations() << '\n';
- std::cout << "final point: " << minimum << '\n';
+ std::cout << "final point: " << optimizer.point() << '\n';
- #ifdef VISUALIZE
if (!log_file_path.empty()) {
- json data = optimizer;
+ json data = log;
std::ofstream log_file(log_file_path);
log_file << data.dump(4) << '\n';
}
if (!vis_file_path.empty()) {
- json data = CmaEsVis{optimizer};
+ json data = CmaEsVis{log};
std::ofstream vis_file(vis_file_path);
vis_file << data.dump(4) << '\n';
}
- #endif
return 0;
}