singularity pull docker://epfl/twk-lausanne:2.0 singularity exec twk-lausanne_2.0.sif twk-dashboard These containers embed all optional dependencies (CUDA, neuroimaging libraries, JupyterLab) and are . 4.4. Source Code (Git) If you prefer to develop on the bleeding edge:
dti = DTI(gpu=True) dti.fit(dataset.dwi, bvals=dataset.bval, bvecs=dataset.bvec) fa_map = dti.fa() tvis.plot_volume(fa_map, cmap='viridis') TWK Lausanne ships a Ray‑based distributed executor . Example for scaling across a Kubernetes cluster:
# ------------------------------------------------- # 4. Threshold and visualise the contrast # ------------------------------------------------- contrast = glm.contrast('2back > 0back') thresholded = tstat.threshold(contrast, p=0.05, method='fdr') tvis.plot_brain(thresholded, surface='fsaverage', cmap='cold_hot') The same pipeline can be that the web dashboard can execute without writing any code: twk lausanne download
docker pull epfl/twk-lausanne:2.0 docker run -it --rm -v $PWD:/data epfl/twk-lausanne:2.0 bash For HPC clusters that rely on Singularity:
pipeline_json = preproc.to_json() tvis.save_dashboard(pipeline_json, out="my_analysis.json") 6.1. GPU‑Accelerated Diffusion Tensor Imaging from twk.diffusion import DTI, cuda_enabled singularity pull docker://epfl/twk-lausanne:2
# Activate the environment conda activate twk-lausanne
python -m pip install "twk-lausanne[cuda]" Pre‑built images are published on Docker Hub: Example for scaling across a Kubernetes cluster: #
# Verify CUDA availability assert cuda_enabled(), "CUDA not found – install the twk-lausanne[cuda] extra."