fatal error: numpy/arrayobject.h: No such file or directory

[ghost]

when compiling I am getting the following error

fatal error: numpy/arrayobject.h: No such file or directory

Python 2.7 and 3 already installed. Numpy is installed. cython is installed.
using ubuntu 16.04

[lava]

According to https://packages.ubuntu.com/search?searchon=contents&keywords=numpy%2Farrayobject.h&mode=exactfilename&suite=xenial&arch=any there are three packages for xenial containing this file, if you have installed one of them it should be just a matter of verifying that the include path points to the correct location.

[joelrorseth]

I'm having the same issue on macOS. I've tried linking with Python 2.7 and 3.6 paths. Doesn't work for either. Using homebrew for my 3.6 install.

[rene-aguirre]

I'm getting the actual path using this variable in my makefile (fix the python interpreter to python2 or python3 to reflect your installation if needed), OS X Mojave:

NUMPY_INCLUDE := -I$(shell python -c 'import numpy; print(numpy.get_include())')

Also, if using homebrew on Mac OS X try using the python2-config or python3-config script to retrieve --ldflags and --cflags compilation parameters.

Anyway, I'm able to make it work with python2 only (after fixing matplotlibrc backend). I'm having run-time issues when trying to use Python 3.7.x

[asharif]

if you used pip2 to install numpy then you should need this flag to g++

-I /usr/local/lib/python2.7/site-packages/numpy/core/include/

[XiaoxingChen]

Same issue, manually find numpy/arrayobject.h at

/usr/local/lib/python3.5/dist-packages/numpy/core/include

Therefore,

set(PYTHON_INCLUDE_DIRS ${PYTHON_INCLUDE_DIRS} /usr/local/lib/python3.5/dist-packages/numpy/core/include)

could be a provisional solution.