Data science stack (DSS)

Data science stack (DSS) lets users create and manage ready-to-run machine-learning and data-science environments on a workstation. Its command-line interface controls containerised environment images, allowing users to work with prepared stacks instead of assembling each machine-learning configuration manually. The images run on MicroK8s and can provide environments based on PyTorch or TensorFlow, while DSS combines MicroK8s with JupyterLab and MLFlow.

DSS uses isolated, reproducible environments to reduce the configuration work involved in preparing a machine-learning workstation. When GPU resources are available, those environments are designed to use the workstation’s GPUs. The same environment model supports exploratory development as well as preparation for production-oriented workloads. It suits people starting with machine learning and engineers building more complex development or runtime environments.

Canonical publishes DSS under an open-source license for use on Ubuntu and other Snap-enabled workstations. It requires Linux with snapd and supports the x86_64 (amd64) architecture. The current version is 0.1-0fdd0b77cd50bb922f643b17899b.