FMCooler

FMCooler is a tool that uses the Simulated Annealing algorithm to search for valid and optimized feature configurations within feature models.

fmcooler_demo.mp4

Functionality

FMCooler explores Feature Models to find optimal or near-optimal configurations by applying heuristic search techniques. It is especially useful for working with complex software product lines and variability management.

Required Setup

This project is publicly hosted as an open-source repository on GitHub and can be cloned using the following command:

git clone http://github.com/jdanielescanez/fmcooler

Installation

FMCooler requires Python >= 3.11. Once the repository has been cloned, navigate into the project directory and install the module along with its dependencies in editable mode:

cd fmcooler
pip install -e .

This will install all necessary dependencies and make the package available for development and use. You can see the dependencies inside the pyproject.toml file.

FMCooler was also tested succesfully on the cloud-based Google Colab, so users could upload it to the session and use a notebook to install the dependencies and launch the tool.

Windows users, that want to run it locally, will require to install extra software for a succesful installation, such as Microsoft C++ Build Tools (e.g., the qubovert dependency is not precompiled so its C++ parts will require to be compiled in your computer).

Usage

A generic example of execution of fmcooler gets a feature model (e.g., tree.uvl) and a weights file (e.g., weights.csv) where we have the data about the impact of each feature for a concrete criteria (e.g., $\sum_i \text{weight}(\text{var}_i) \cdot \text{rate}_i$, defined by vars:rates:min, where the elements of vars, rates and min are separated by commas. The latter has min and max for each variable, making that when max, the corresponding weight is multiplied by -1), using 100 reads. A read, in the SA terminology, refers to one independent run of the annealing algorithm, starting from an initial state and proceeding through the annealing schedule to produce a single sample. The final result will be the best result among the independent reads. An example of multi-criteria execution can be seen below, for the minimisation of Battery with the rate of 0.4 and the maximisation of Usability with the rate of 0.6.

./fmcooler.py tree.uvl weights.csv Battery,Usability:0.4,0.6:min,max 100

Execution Examples

Mono-criteria

An example of execution of fmcooler runs 1000 independent annealing-based optimization processes on the feature model specified in examples/mobile_media.uvl, using the impact data from examples/mobile_media.csv. The goal is to minimize the weighted influence of the Battery feature with a coefficient of 1.0. Each run starts from a different initial state and follows an annealing schedule to explore possible configurations, producing samples that seek to reduce the total impact of the Battery feature as much as possible.

./fmcooler.py examples/mobile_media.uvl examples/mobile_media.csv Battery:1.0:min 1000

Multi-criteria

The last example of fmcooler execution runs 10000 processes on the same feature model and using the same impact data. The goal is to maximize the combined weighted influence of Battery and Usability features, each weighted with a coefficient of 0.5.

./fmcooler.py examples/mobile_media.uvl examples/mobile_media.csv Battery,Usability:0.5,0.5:max,max 10000

Contributing

Contributions are welcome! Feel free to fork the repository and submit pull requests. If you encounter any bugs or have feature requests, please open an issue.

License

This project is released under an open-source license. See the LICENSE file for more information.