Make the library created by Eigen in C ++ available from Python with Boost.Numpy.

This article

Make a note of what you did at the Sapporo C ++ Study Group Online Mokumokukai # 36.

Installation method

Please hit another place. Do your best.

things to do

As a simple example, the sum of two-dimensional arrays prepared by Python numpy It calculates using the C ++ Eigen library and returns it to Python again.

C ++ source code

cpplib.cpp

#define EIGEN_DEFAULT_TO_ROW_MAJOR
#include<Eigen/Core>
#include<boost/numpy.hpp>
namespace py = boost::python;
namespace np = boost::numpy;

np::ndarray add_double(const np::ndarray lhs, const np::ndarray rhs) {
    using Stride = Eigen::Stride<Eigen::Dynamic, Eigen::Dynamic>;
    const auto
        lrows = lhs.shape(0),
        lcols = lhs.shape(1),
        rrows = rhs.shape(0),
        rcols = rhs.shape(1);

    const Stride
        lstride(lhs.strides(0)/sizeof(double), lhs.strides(1)/sizeof(double)),
        rstride(rhs.strides(0)/sizeof(double), rhs.strides(1)/sizeof(double));

    const Eigen::Map<const Eigen::MatrixXd, Eigen::Unaligned, Stride>
        lmat(reinterpret_cast<double*>(lhs.get_data()), lrows, lcols, lstride),
        rmat(reinterpret_cast<double*>(rhs.get_data()), rrows, rcols, rstride);

    np::ndarray ret = np::empty(
        py::make_tuple(lmat.rows(),lmat.cols()), 
        np::dtype::get_builtin<double>());

    Eigen::Map<Eigen::MatrixXd>
        ret_mat(reinterpret_cast<double*>(ret.get_data()), lmat.rows(), lmat.cols());

    ret_mat = lmat + rmat;
    return ret;
}

BOOST_PYTHON_MODULE(cpplib) {
    Py_Initialize();
    np::initialize();
    py::def("add_double",add_double); 
}

It may be easier to write if you just want to add matrices, but I've written it so that it can be applied. It is unavoidable that the type is limited to double. The only way to do it is to check dtype and branch, but there is no choice but to judge at runtime.

How to compile

I will do it with SCons. For example, like this. I'm doing it in a mingw environment.

SConstruct

# vim: filetype=python
# SConstruct

env = Environment(
    SHLIBPREFIX="",
    SHLIBSUFFIX=".pyd",
    CXX="/mingw64/bin/g++",
    CXXFLAGS=["-std=c++14"],
    LINKFLAGS=["-std=c++14"],
    CPPPATH=[
        "/mingw64/include/eigen3",
        "/mingw64/include/python3.5m", ],
    LIBPATH=["/mingw64/lib"])

env.SharedLibrary("cpplib", ["cpplib.cpp"],
    LIBS=["boost_numpy", "python3.5.dll", "boost_python3-mt"])

Python code

usecpp.py

import numpy as np
from cpplib import add_double

a = np.array(
    [[1, 2, 3, 4],
     [5, 6, 7, 8],
     [9, 1, 2, 3],
     [4, 5, 6, 7], ], dtype=np.float64)
b = np.array(
    [[5, 6],
     [4, 5], ], dtype=np.float64)

result = add_double(a[1::2, 1::2], b[::1, ::1])
print(result)

Result is Since [[6,8], [5,7]] and [[5,6], [4,5]] are added

$ python3 usecpp.py
[[ 11.  14.]
 [  9.  12.]]

It will be. The cloudy meeting time has expired here as well.

Postscript

https://github.com/ignisan/boost_numpy_project/blob/master/to_eigen_map.hpp