🔧 重新设计 rm::FuncNds，完成 #213

cv-rmvl · Jan 14, 2025 · 2d93f5c · 2d93f5c
1 parent 254fbc1
commit 2d93f5c
Show file tree

Hide file tree

Showing 6 changed files with 127 additions and 121 deletions.
diff --git a/doc/tools/pyrmvl_fns.cfg b/doc/tools/pyrmvl_fns.cfg
@@ -48,8 +48,8 @@ rm                     region          func,x0[,delta]         a,b
 rm                     fminbnd         func,x1,x2[,options]    x,fval
 rm                     fminunc         func,x0[,options]       x,fval
 rm                     fmincon         func,x0,c,ceq[,options] x,fval
-rm                     lsqnonlin       funcs,x0[,options]      x
-rm                     lsqnonlinRKF    funcs,x0,rb[,options]   x
+rm                     lsqnonlin       func,x0[,options]       x
+rm                     lsqnonlinRKF    func,x0,rb[,options]    x
 rm                     dft             xt                      Xf
 rm                     idft            Xf                      xt
 rm                     Gx              x,type                  G

diff --git a/extra/combo/include/rmvl/combo/armor.h b/extra/combo/include/rmvl/combo/armor.h
@@ -178,7 +178,7 @@ const char *to_string(ArmorSizeType armor_size);
 /**
  * @brief StateType 转为装甲板大小类型
  *
- * @param[in] str 字符串
+ * @param[in] tp StateType 类型
  */
 ArmorSizeType to_armor_size_type(const StateType &tp);
 
@@ -192,9 +192,9 @@ const char *to_string(RobotType robot);
 /**
  * @brief StateType 转为机器人类型
  *
- * @param[in] str 字符串
+ * @param[in] tp StateType 类型
  */
-RobotType to_robot_type(const StateType &type);
+RobotType to_robot_type(const StateType &tp);
 
 //! @} combo_armor
 

diff --git a/modules/algorithm/include/rmvl/algorithm/numcal.hpp b/modules/algorithm/include/rmvl/algorithm/numcal.hpp
@@ -307,7 +307,7 @@ using Func1ds = std::vector<std::function<double(double)>>;
 //! 多元函数
 using FuncNd = std::function<double(const std::valarray<double> &)>;
 //! 多元函数组
-using FuncNds = std::vector<std::function<double(const std::valarray<double> &)>>;
+using FuncNds = std::function<std::valarray<double>(const std::valarray<double> &)>;
 
 //! 梯度/导数计算模式
 enum class DiffMode : uint8_t
@@ -348,7 +348,7 @@ struct RMVL_EXPORTS_W_AG OptimalOptions
  *
  * @param[in] func 一元函数
  * @param[in] x 指定位置的自变量
- * @param[in] mode 导数计算模式，默认为中心差商 `Diff_Central`
+ * @param[in] mode 导数计算模式，默认为中心差商 `DiffMode::Central`
  * @param[in] dx 坐标的微小增量，默认为 `1e-3`
  * @return 函数在指定点的导数
  */
@@ -359,7 +359,7 @@ RMVL_EXPORTS_W double derivative(Func1d func, double x, DiffMode mode = DiffMode
  *
  * @param[in] func 多元函数
  * @param[in] x 指定位置的自变量
- * @param[in] mode 梯度计算模式，默认为中心差商 `Diff_Central`
+ * @param[in] mode 梯度计算模式，默认为中心差商 `DiffMode::Central`
  * @param[in] dx 计算偏导数时，坐标的微小增量，默认为 `1e-3`
  * @return 函数在指定点的梯度向量
  */
@@ -410,13 +410,13 @@ RMVL_EXPORTS_W std::pair<std::valarray<double>, double> fmincon(FuncNd func, con
 /**
  * @brief 非线性最小二乘求解，实现与 \cite Agarwal23 类似的算法
  *
- * @param[in] funcs 最小二乘目标函数，满足 \f[F(\pmb x_k)=\frac12\|\pmb f(\pmb x_k)\|_2^2=\frac12
- *                  \left(\texttt{funcs}[0]^2+\texttt{funcs}[1]^2+\cdots+\texttt{funcs}[n]^2\right)\f]
+ * @param[in] func 最小二乘目标函数，满足 \f[F(\pmb x_k)=\frac12\|\pmb f(\pmb x_k)\|_2^2=\frac12
+ *                  \left(\texttt{func}[0]^2+\texttt{func}[1]^2+\cdots+\texttt{func}[n]^2\right)\f]
  * @param[in] x0 初始点
  * @param[in] options 优化选项，可供设置的有 `lsq_mode`、`max_iter`、`tol` 和 `dx`
  * @return 最小二乘解
  */
-RMVL_EXPORTS_W std::valarray<double> lsqnonlin(const FuncNds &funcs, const std::valarray<double> &x0, const OptimalOptions &options = {});
+RMVL_EXPORTS_W std::valarray<double> lsqnonlin(const FuncNds &func, const std::valarray<double> &x0, const OptimalOptions &options = {});
 
 //! Robust 核函数
 enum class RobustMode : uint8_t
@@ -431,13 +431,13 @@ enum class RobustMode : uint8_t
 /**
  * @brief 带 Robust 核函数的非线性最小二乘求解
  *
- * @param[in] funcs 最小二乘目标函数，参考 rm::lsqnonlin
+ * @param[in] func 最小二乘目标函数，参考 rm::lsqnonlin
  * @param[in] x0 初始点
  * @param[in] rb Robust 核函数模式，参考 rm::RobustMode ，选择 `rm::RobustMode::L2` 时退化为 `rm::lsqnonlin`
  * @param[in] options 优化选项，可供设置的有 `lsq_mode`、`max_iter`、`tol` 和 `dx`
  * @return 最小二乘解
  */
-RMVL_EXPORTS_W std::valarray<double> lsqnonlinRKF(const FuncNds &funcs, const std::valarray<double> &x0, RobustMode rb, const OptimalOptions &options = {});
+RMVL_EXPORTS_W std::valarray<double> lsqnonlinRKF(const FuncNds &func, const std::valarray<double> &x0, RobustMode rb, const OptimalOptions &options = {});
 
 //! @} algorithm_optimal
 

diff --git a/modules/algorithm/perf/perf_optimal.cpp b/modules/algorithm/perf/perf_optimal.cpp
@@ -64,14 +64,17 @@ static void cg_quadratic_cv(benchmark::State &state)
 BENCHMARK(cg_quadratic_rmvl)->Name("fminunc (conj_grad, quadratic) - by rmvl  ")->Iterations(50);
 BENCHMARK(cg_quadratic_cv)->Name("fminunc (conj_grad, quadratic) - by opencv")->Iterations(50);
 
-static inline double cle1(const std::valarray<double> &x) { return -x[0] - x[1] + 10; }
-static inline double cle2(const std::valarray<double> &x) { return 2 * x[0] + x[1] - 30; }
-static inline double cle3(const std::valarray<double> &x) { return -x[0] + x[1] - 5; }
+static inline std::valarray<double> cle(const std::valarray<double> &x)
+{
+    return {-x[0] - x[1] + 10,
+            2 * x[0] + x[1] - 30,
+            -x[0] + x[1] - 5};
+}
 
 static void com_quadratic_rmvl(benchmark::State &state)
 {
     for (auto _ : state)
-        rm::fmincon(quadraticFunc, {0, 0}, {cle1, cle2, cle3}, {});
+        rm::fmincon(quadraticFunc, {0, 0}, cle, {});
 }
 
 BENCHMARK(com_quadratic_rmvl)->Name("fmincon (conj_grad, quadratic) - by rmvl  ")->Iterations(50);
@@ -135,11 +138,18 @@ void lsqnonlin_rmvl(benchmark::State &state)
 {
     for (auto _ : state)
     {
-        rm::FuncNds lsq_sine(20);
+        std::array<rm::FuncNd, 20> lsq_sine;
         for (std::size_t i = 0; i < lsq_sine.size(); ++i)
             lsq_sine[i] = [=](const std::valarray<double> &x) { return x[0] * std::sin(x[1] * i + x[2]) + x[3] - real_f(i); };
 
-        auto x = rm::lsqnonlin(lsq_sine, {1, 0.02, 0, 1.09});
+        rm::FuncNds lsq_sine_f = [&](const std::valarray<double> &x) {
+            std::valarray<double> ret(lsq_sine.size());
+            for (std::size_t i = 0; i < lsq_sine.size(); ++i)
+                ret[i] = lsq_sine[i](x);
+            return ret;
+        };
+
+        auto x = rm::lsqnonlin(lsq_sine_f, {1, 0.02, 0, 1.09});
         benchmark::DoNotOptimize(x);
     }
 }