Machine Learning Plugin¶
A deep technical reference for practical machine learning: framing the problem and recognising when not to use ML, data work and the leakage that follows from getting splits wrong, the classical toolkit, deep learning fundamentals, modern architectures, the PyTorch/JAX ecosystem, training at scale, fine-tuning and parameter-efficient adaptation, evaluation done honestly, debugging training runs, inference and serving, MLOps and reproducibility, hardware and cost, and the failure modes that make models look better than they are.
One reference, split into 4 skills along its section groups so a task loads only the part it needs. Section numbers (§N) are shared across the set and cross-references into a sibling skill are written as §N → skill. Reference, not tutorial: sections are independent, every claim is tagged by how durable it is (stable fundamentals vs. versioned specifics vs. genuinely contested questions), and a currency snapshot (verified August 2026) flags what goes stale first.
Skills¶
- ml-framing-data-and-classical — Framing, Data, and the Classical Toolkit (§0–§3): Routing; Framing; Data; The Classical Toolkit.
- ml-deep-learning-and-training — Deep Learning, Architectures, the Ecosystem, and Training at Scale (§4–§8): Deep Learning Fundamentals; Architectures; The Ecosystem; Training at Scale; Fine-Tuning and Adaptation.
- ml-evaluation-serving-mlops-and-safety — Evaluation, Debugging, Inference and Serving, MLOps, and Responsible ML (§9–§14): Evaluation; Debugging Training; Inference and Serving; MLOps and Reproducibility; Hardware and Cost; Interpretability, Fairness, Safety.
- ml-reference — Anti-Patterns, Contested Questions, Currency, and Canon (§15–§20): Anti-Patterns; Contested Questions; Currency Snapshot; The Canon; Quick Reference; Sources and Method.
Skills in this plugin¶
- ml-deep-learning-and-training
- ml-evaluation-serving-mlops-and-safety
- ml-framing-data-and-classical
- ml-reference