matplotlib-cpp

Welcome to matplotlib-cpp, possibly the simplest C++ plotting library. It is built to resemble the plotting API used by Matlab and matplotlib.

Usage

Complete minimal example:

#include "matplotlibcpp.h"
namespace plt = matplotlibcpp;
int main() {
    plt::plot({1,2,3,4});
    plt::show();
}

// g++ minimal.cpp -std=c++11 -lpython2.7

Result:

A more comprehensive example:

#include "matplotlibcpp.h"
#include <cmath>

namespace plt = matplotlibcpp;

int main() 
{
    // Prepare data.
    int n = 5000;
    std::vector<double> x(n), y(n), z(n), w(n,2);
    for(int i=0; i<n; ++i) {
        x.at(i) = i*i;
        y.at(i) = sin(2*M_PI*i/360.0);
        z.at(i) = log(i);
    }

    // Plot line from given x and y data. Color is selected automatically.
    plt::plot(x, y);
    // Plot a red dashed line from given x and y data.
    plt::plot(x, w,"r--");
    // Plot a line whose name will show up as "log(x)" in the legend.
    plt::named_plot("log(x)", x, z);

    // Set x-axis to interval [0,1000000]
    plt::xlim(0, 1000*1000);
    // Enable legend.
    plt::legend();
    // Save the image (file format is determined by the extension)
    plt::save("./basic.png");
}

// g++ basic.cpp -lpython2.7

Result:

matplotlib-cpp doesn't require C++11, but will enable some additional syntactic sugar when available:

#include <cmath>
#include "matplotlibcpp.h"

using namespace std;
namespace plt = matplotlibcpp;

int main() 
{    
    // Prepare data.
    int n = 5000; // number of data points
    vector<double> x(n),y(n); 
    for(int i=0; i<n; ++i) {
        double t = 2*M_PI*i/n;
        x.at(i) = 16*sin(t)*sin(t)*sin(t);
        y.at(i) = 13*cos(t) - 5*cos(2*t) - 2*cos(3*t) - cos(4*t);
    }

    // plot() takes an arbitrary number of (x,y,format)-triples. 
    // x must be iterable (that is, anything providing begin(x) and end(x)),
    // y must either be callable (providing operator() const) or iterable. 
    plt::plot(x, y, "r-", x, [](double d) { return 12.5+abs(sin(d)); }, "k-");


    // show plots
    plt::show();
} 

// g++ modern.cpp -std=c++11 -lpython

Result:

Installation

matplotlib-cpp works by wrapping the popular python plotting library matplotlib. (matplotlib.org) This means you have to have a working python installation, including development headers. On Ubuntu:

sudo aptitude install python-matplotlib python2.7-dev

The C++-part of the library consists of the single header file matplotlibcpp.h which can be placed anywhere. Since a python interpreter is opened internally, it is necessary to link against libpython2.7 in order to use matplotlib-cpp. (There should be no problems using python3 instead of python2.7, if desired)

Why?

I initially started this library during my diploma thesis. The usual approach of writing data from the c++ algorithm to a file and afterwards parsing and plotting it in python using matplotlib proved insufficient: Keeping the algorithm and plotting code in sync requires a lot of effort when the C++ code frequently and substantially changes. Additionally, the python yaml parser was not able to cope with files that exceed a few hundred megabytes in size.

Therefore, I was looking for a C++ plotting library that was extremely to use and easy to add into an existing codebase, preferrably header-only. When I found none, I decided to write one myself, which is basically a C++ wrapper around matplotlib. As you can see from the above examples, plotting data and saving it to an image file can be done is as few as two lines of code.

The general approach of providing a simple C++ API for utilizing python code was later generalized and extracted into a separate, more powerful library in another project of mine, wrappy.

Todo/Issues/Wishlist

• This library is not thread safe. Protect all concurrent access with a mutex. Sadly, this is not easy to fix since it is not caused by the library itself but by the python interpreter, which is itself not thread-safe.

• It would be nice to have a more object-oriented design with a Plot class which would allow multiple independent plots per program.

• Right now, only a small subset of matplotlibs functionality is exposed. Stuff like xlabel()/ylabel() etc. should be easy to add.

• A lot of copying could be avoided if we generate numpy arrays directly instead of python lists