Setting up philkr's fork of supertuxkart on a fresh Ubuntu installation
(tested on Ubuntu 16.04 LTS running on a VirtualBox)
Update and install dependencies (many suggested by the supertuxkart GitHub readme)
sudo apt-get update
sudo apt-get upgrade
sudo apt-get install git python3-pip build-essential libssl-dev libffi-dev python-dev cmake libbluetooth-dev libcurl4-gnutls-dev libfreetype6-dev libfribidi-dev libgl1-mesa-dev libjpeg-dev libogg-dev libopenal-dev libpng-dev libvorbis-dev libxrandr-dev mesa-common-dev pkg-config zlib1g-dev python3-venv python3-tk
Create a virtualenv and install more dependencies
python3 -m venv nn
source nn/bin/activate
pip install tensorflow jupyter matplotlib
Clone the repository and get into pykart, where we'll be editing a file involved in CMake (I honestly don't really know how it all works but the error messages and StackOverflow were clear enough)
git clone https://github.com/philkr/supertuxkart.git
cd supertuxkart/pykart/
Find out where numpy is (while still in your virtualenv - check with which python and which pip)
python -c "import numpy; print(numpy.__file__)"
You'll get something like
/home/rainier/Desktop/nn/lib/python3.5/site-packages/numpy/__init__.py
That's where your virtualenv's Python executable (which python) looks for numpy when you import it.
FindNumPy.cmake is trying to set PYTHON_NUMPY_INCLUDE_DIR. If you do not follow the next step, cmake will fail to find this variable and terminate.
After line 31 of FindNumPy.cmake add a line like:
set(PYTHON_NUMPY_INCLUDE_DIR "/home/rainier/Desktop/nn/lib/python3.5/site-packages/numpy/core/include")
Go back to the supertuxkart root, make a build dir and build
cd ..
mkdir build
cd build
cmake ..
make -j9
Put the supertuxkart binary in build/bin in the pykart working directory
cp bin/supertuxkart ../pykart
And race
python __main__.py
fixed, thank you!