📘 Volume 5 — Data Science
Make it analytical. Jupyter, datasets, pipelines, ML workflows on the NAS. The volume that turns storage into insight.
Many home lab builders are also data scientists, analysts, students, or curious tinkerers with side projects. The NAS is a great platform for analytical work: lots of storage for datasets, a Linux environment for the tools, network access for cloud services. This volume covers the patterns that make the NAS a data science platform.
What this volume covers
The data scientist's NAS is a different kind of use from the family NAS or the home lab. The patterns:
- The data scientist's NAS — what changes when the workload is analysis, not files
- Jupyter — the interactive computing environment, on the NAS
- Data storage and format — Parquet, Arrow, the right file format for analysis
- Pipelines and schedules — Airflow, Prefect, cron for the home
- Machine learning — scikit-learn, PyTorch, MLflow, the model lifecycle
- Sharing and collaboration — notebooks, datasets, results, with other people
- Reproducibility — the discipline of doing the work in a way that can be repeated
Chapters
📖 Ch 01 · The Data Scientist's NAS
What changes when the workload is analysis. Datasets, environments, the data layer.
10 min read📖 Ch 02 · Jupyter
JupyterHub, JupyterLab, the interactive computing environment on the NAS.
12 min read📖 Ch 03 · Data Storage and Format
Parquet, Arrow, partitioning, the right file format for analysis. Why CSV is usually wrong.
12 min read📖 Ch 04 · Pipelines and Schedules
Airflow, Prefect, cron. The orchestration of recurring data work.
12 min read📖 Ch 05 · Machine Learning
scikit-learn, PyTorch, MLflow. The ML lifecycle on the NAS.
15 min read📖 Ch 06 · Sharing and Collaboration
Notebooks, datasets, results. How to share analytical work safely.
10 min read📖 Ch 07 · Reproducibility
Doing the work in a way that can be repeated. Git, DVC, conda, Docker. The discipline that makes analysis trustworthy.
12 min read📖 Ch 08 · Remote Development
The dev-server setup. VS Code Server, the SSH workflow, remote Jupyter. The thin client, fat server model.
12 min readThe "data" mindset
The data scientist thinks in datasets, queries, and pipelines. The patterns from earlier volumes (snapshots, scrubs, the 3-2-1 backup) all still apply, but the workload is different: lots of reads on big files, lots of writes of new files, lots of computation on the data. The NAS has to support all three well.
What you'll know by the end
- How to set up Jupyter on the NAS for interactive analysis
- How to store data in the right format for analysis (Parquet, not CSV)
- How to schedule recurring data work (cron, Airflow, Prefect)
- How to run machine learning experiments and track them
- How to share analytical work with others safely
- How to do the work in a way that can be reproduced
The data scientist's home lab is the same hardware as the home lab in Volume 4, but the workloads are different. The data scientist cares about: storage I/O for reading big datasets, CPU/RAM for in-memory analysis, and reproducibility (the discipline of doing the work in a way that can be repeated). The patterns from Volume 1 (snapshots, scrubs) and Volume 4 (containers, VMs) are the foundation. This volume is the workload layer.