Ecole

Ecole (pronounced [ekɔl]) stands for Extensible Combinatorial Optimization Learning Environments and aims to expose a number of control problems arising in combinatorial optimization solvers as Markov Decision Processes (i.e., Reinforcement Learning environments). Rather than trying to predict solutions to combinatorial optimization problems directly, the philosophy behind Ecole is to work in cooperation with a state-of-the-art Mixed Integer Linear Programming solver that acts as a controllable algorithm.

The underlying solver used is SCIP, and the user facing API is meant to mimic the OpenAi Gym API (as much as possible).

import ecole

env = ecole.environment.Branching(
    reward_function=-1.5 * ecole.reward.LpIterations() ** 2,
    observation_function=ecole.observation.NodeBipartite(),
)
instances = ecole.instance.SetCoverGenerator()

for _ in range(10):
    obs, action_set, reward_offset, done, info = env.reset(next(instances))
    while not done:
       obs, action_set, reward, done, info = env.step(action_set[0])

Installation

Conda

conda install -c scipopt -c conda-forge ecole

PyScipOpt is not required but is the main SCIP interface to develop new Ecole components from Python

conda install -c scipopt -c conda-forge ecole pyscipopt

Currenlty, conda packages are only available for Linux and MacOS.

Pip

Unavailable

User Documentation

Please refer to the documentation for tutorials, examples, and reference.

---

Developer Notes

Build dependencies

Conda

All dependencies required for building Ecole can be resolved using a conda environment. Install everything in a development (named ecole) environment using

conda env create -n ecole -f dev/conda.yaml

For the following, the ecole environment always needs to be activated.

conda activate ecole
conda config --append channels conda-forge
conda config --set channel_priority flexible

Note: this environment contains tools to build ecole and scip, format code, test, generate documentation etc. These are more than the dependencies to only use Ecole.

CMake

CMake is a meta-build tool, used for configuring other build tools (e.g. Make) or IDE's. The whole build of Ecole can be done with CMake. A one-time configuration is necessary for CMake to find dependencies, detect system information, etc. Using cmake, we recommend building out of source using cmake -B build/ to configure, and cmake --build build/ to compile. CMake is made available in the ecole environment created earlier.

SCIP runtime dependency

Scip dependency is resolved from conda inside dev/conda.yaml. No other action are necessary :)

Building

In the Ecole source repository, configure using

cmake -B build/ -D ECOLE_DEVELOPER=ON

Note: this is the time to pass optional build options, such as the build type and compiler choice. For instance ``-D CMAKE_BUILD_TYPE=Debug`` can be added to compile with debug information.

The definition -D ECOLE_DEVELOPER=ON changes the default settings (such as the build type, static analysis, etc.) for added convenience and to ensure high quality contributions. Only the default settings are changed, this mode does not override any explicit setting.

Then, build Ecole with

cmake --build build/

The Python package can finally be installed from the build directory

python -m pip install -I build/python

Compiler issues

If you encounter problem with your compiler (because it is too old for instance), you can use the ones from ananconda.

conda install gxx_linux-64  # Linux

And start again the configuring of Ecole.

rm -r build/ && cmake -B build/ -D ECOLE_DEVELOPER=ON

Running the tests

C++ tests

The C++ tests are build with Catch2. It produces a single standalone executable under build/libecole/tests/test-libecole. For test options, consult

build/libecole/tests/test-libecole --help

Python tests

Python tests are build with PyTest. As any Python project, running the tests requires that Python finds the correct Ecole package. It is therefore not recommended to manually install Ecole between every tests, as it leaves room for running tests with an out of date installation. The CMake configuration creates a Python virtual environment and installs the Ecole Python package inside automatically. To run the tests (or another operation with Python Ecole), use the virtual environment

./build/venv/bin/python -m pytest python/tests/

Generating the documentation

The documentation is a CMake target that is not built by default. When using -D ECOLE DEVELOPER=ON, the documentation can be generated by providing the target to the CMake build command:

cmake --build build/ --target ecole-sphinx

The generated HTML files are located under build/doc/sphinx/html. In particular, build/doc/sphinx/html/index.html can be opened in your browser to visualize the documentation.

Coding standards

The quality and conventions of the code are enforced automatically with various tools, for instance to format the layout of the code and fix some C++ error-prone patterns.

Compilation database

Some C++ tools need access to a compilation database. This is a file called compile_commands.json that is created automatically by CMake when using -D ECOLE_DEVELOPER=ON. The file needs to be accessible at the root directory of the project, so you should symlink it like so (assuming you set CMake to configure in the directory named build as shown before).

ln -s build/compile_commands.json

Pre-commit

The tools are configured to run with pre-commit, that is they can be added to run automatically when making a commit, pushing, or on demand. To have the tools run automatically, install the pre-commit hooks using

pre-commit install

The tools are configured to run light tests only on the files that were changed during the commit, so they should not run for long. Installing the pre-commit hooks to run the tools is recommended. Similar tests will be run online and pull requests will fail if the tools have not been run.

With pre-commit hooks, commits will be rejected by git if the tests ran by the tools fail. If the tools can fix the issue for you, you will find some modifications that you can add to your commit.

Sometimes when working locally, it can be useful not to run the tools. You can tell git to ignore the pre-commit hooks by passing the --no-verify to any git command making commit, including commit, merge, rebase, push... But again, there is little interest to push commit for which the pre-commit tests fail, as they will ultimately fail online as well.

Other times you may wish to run the tools on all files unconditionally. This can be done using

pre-commit run --all-files