ModuleNotFoundError: No module named 'operon'

[heal-research]

ModuleNotFoundError: No module named 'operon'

[foolnotion]

Thanks for the report. That's because the module you see in the build folder is an internal module (that's why it's called _operon) that binds the methods from the C++ library almost 1:1. The internal module is wrapped by a pure python module operon which additionally provides the sklearn estimator.

Normally, you would provide an install prefix with -DCMAKE_INSTALL_PREFIX=/path/to/module and then run make install when the build is done. /path/to/module should be in the PYTHONPATH so python can pick it up.

I can update the install script but I would need to know where you want it installed.

EDIT: also, if you don't provide an install prefix the CMake script will try to figure out your python library path and install the module there (in this case it might be necessary to sudo make install). The install script will copy the dll and the pure python module from here into your desired install folder.

[lacava]

Operon should end up in $CONDA_PREFIX/lib/python3.9/site-packages. I just tried providing this as an argument to -DCMAKE_INSTALL_PREFIX with no luck. I also tried creating a test_dir in the operon directory and specifying that, but no files were copied there. I do see the line "-- Python bindings will be installed in /path/to/operon/test_dir" during Cmake build, so maybe it is failing silently.

Could be nice to have the python module install as a python package using a minimal setup.py script and a call to python setup.py install.

[foolnotion]

This install script works on my ubuntu machine. It seems that CMake is able to detect the correct library path, so I'd recommend omitting CMAKE_INSTALL_PREFIX option so that we don't hardcode the python version.

The module import failure was caused by a name conflict, the file regression-benchmark/experiment/methods/operon.py should be renamed to something else like OperonRegressor.py.

# Installing operon
export CC=gcc-9
export CXX=gcc-9
# Dependencies 
# - TBB - installed via conda
# - Eigen- installed via conda
# - Ceres- installed via conda
# - {fmt}- installed via conda

git clone https://github.com/heal-research/operon
cd operon
git checkout 79617d76c97838d91293cff5d4e4fda7230b904b
# run cmake
mkdir build; cd build; 
cmake .. -DCMAKE_BUILD_TYPE=Release -DBUILD_PYBIND=ON -DUSE_OPENLIBM=ON -DUSE_SINGLE_PRECISION=ON -DCERES_TINY_SOLVER=ON

# build
make VERBOSE=1 -j pyoperon

# install
make install

After configuring and installing and renaming the regressor, I was actually able to run a small experiment:

time python evaluate_model.py -ml OperonRegressor Poly-10.csv -seed 1234 -results_path ./results/

[lacava]

ah ok. Adding make VERBOSE=1 -j pyoperon; make install worked like a charm. and i've fixed the name conflict. Would be good to get this in the documentation. Thank you!

[foolnotion]

Thanks! Just one last thing: it's maybe better to not fix the revision number in the install script, (at least not right now) since there are still some minor issues in the sklearn wrapper that need to be ironed out.