use eigen instead of stanmath

compile time differentiation is much faster Revert "autodiff gaussian width parameter" This reverts commit e9bced7.
shogun-toolbox · Nov 5, 2019 · a35cf20 · a35cf20
1 parent 5815a3d
commit a35cf20
Show file tree

Hide file tree

Showing 13 changed files with 9 additions and 1,200 deletions.
diff --git a/.RData b/.RData
diff --git a/.Rhistory b/.Rhistory
diff --git a/Base.h b/Base.h
diff --git a/Derived.h b/Derived.h
diff --git a/Interface.h b/Interface.h
diff --git a/src/shogun/kernel/Autodiff_benchmark.cc b/src/shogun/kernel/Autodiff_benchmark.cc
diff --git a/src/shogun/kernel/GaussianKernel.cpp b/src/shogun/kernel/GaussianKernel.cpp
@@ -6,12 +6,14 @@
  *          Tonmoy Saikia, Sergey Lisitsyn, Matt Aasted, Sanuj Sharma
  */
 
+#include <Eigen/Core>
+#include <unsupported/Eigen/AutoDiff>
 #include <shogun/lib/common.h>
 #include <shogun/kernel/GaussianKernel.h>
 #include <shogun/features/DotFeatures.h>
 #include <shogun/distance/EuclideanDistance.h>
 #include <shogun/mathematics/Math.h>
-#include <stan/math/rev/scal.hpp>
+
 
 using namespace shogun;
 
@@ -92,26 +94,25 @@ void CGaussianKernel::set_width(float64_t w)
 SGMatrix<float64_t> CGaussianKernel::get_parameter_gradient(const TParameter* param, index_t index)
 {
 	using std::exp;
-	using std::log;
 
 	require(lhs, "Left hand side features must be set!");
 	require(rhs, "Right hand side features must be set!");
 
 	if (!strcmp(param->m_name, "log_width"))
 	{
 		SGMatrix<float64_t> derivative=SGMatrix<float64_t>(num_lhs, num_rhs);
-		stan::math::var log_width = m_log_width;	
-		auto constant_part = exp(log_width * 2.0) * 2.0;
+		using EigenScalar = Eigen::Matrix<float64_t, 1, 1>;
+		Eigen::AutoDiffScalar<EigenScalar> eigen_log_width = m_log_width;
 
 		for (int k=0; k<num_rhs; k++)
 		{
 #pragma omp parallel for
 			for (int j=0; j<num_lhs; j++)
 			{
-				auto f = exp(-CShiftInvariantKernel::distance(j, k) / constant_part);
-				f.grad();
-				derivative(j, k) = log_width.adj();
-				stan::math::set_zero_all_adjoints();
+				eigen_log_width.derivatives() = EigenScalar::Unit(1,0);
+				auto el = CShiftInvariantKernel::distance(j, k);
+				Eigen::AutoDiffScalar<EigenScalar> kernel = exp(-el / (exp(eigen_log_width * 2.0) * 2.0));
+				derivative(j, k) = kernel.derivatives()(0);
 			}
 		}
 		return derivative;