KRess: Breast cancer detection through machine learning

This project is done in the context of the Master Calcul Haute Performance et Simulation at the Université de Versailles St-Quentin.

The goal of this project is to implement a tool for breast cancer detection, and optimize it accordingly.

While the first semester focused on the sequential implementation (Versions <= 0.5), the second semester sees us switching to a GPU backend for maximum performance and scalability.

Furthermore, we also this project as an opportunity to explore software architecture on a real scale.

Build: Dependencies

DNF (Yum)

sudo dnf install ocl-icd-devel \
                 opencl-headers \
                 clblast-devel \
                 boost-devel \
                 blas-devel

You will also need a blas implementation, and an OpenCL runtime (we recommend POCL for starter) Note that the Cuda toolkit already provides the necessary OpenCL files.

sudo dnf install pocl \
                 openblas-devel

Build:

git clone https://github.com/Thukisdo/projet-chps21/ kress-chps21
cd kress-chps21 && mkdir build
cd build && cmake .. -DCMAKE_BUILD_TYPE=Release -DCMAKE_CXX_COMPILER=g++
make -j

We advise you to run the tests to ensure everything went well:

cd build && make test -j

Usage

# Build the project if not done already
cd build
./bin/main <path todataset> <output path>

Raw data will be outputted to <output path>/.

Authors

Benjamin Lozes
Ugo Battiston
Mathys Jam

Thanks to

Mohammed-Salah Ibnamar, for supervising us during this project
Thomas Dufaud, for providing us access to a GPU cluster