ml-engineering
Skill0 starsChanged 43 days agoMITView on GitHub
Max's ML engineering playbook. Covers problem framing, leakage-safe feature engineering, time-based validation and backtesting, baselines, experiment tracking and model registry with MLflow, per-entity training at scale with Spark applyInPandas, quality gates and fallbacks, explainability, and drift monitoring and retraining. Includes a deep reference on demand and time-series forecasting for sparse or intermittent data. Use for any ML experiment, training pipeline, evaluation or model debugging task.
3 other projects use a version of this.
Adopted by
Most recently taken up by
- irahardianto/awesome-agvtheir own version, first seen Oct 14, 2026
- stas00/ml-engineeringtheir own version, first seen Oct 8, 2026
- magnus919/agent-skillstheir own version, first seen Jul 21, 2026
Other versions3 projects use a version of this.
Word for word · 0
None found.
Edited · 0
None found.
Wrote their own version · 3
- stas00/ml-engineering★ 20ksame name
- magnus919/agent-skills★ 172same name
- irahardianto/awesome-agv★ 157same name
How matches are found. Word for word: the exact same text. Edited: shares passages with this file (“99% same” means 99% of its five-word phrases also appear here; 10% or more counts). Their own version: similar meaning, or, weakest, the same name. We can’t tell which project had it first. Projects are counted among the 200 most-starred candidates, and mentions are links to the repository we found, up to 50 per site, so both counts are a floor.
In this repository4 other files it gives its agents.
- data-qualitySkill
- pipeline-engineeringSkill
- setupSkill
- squadSkill
The file
We split this skill into the rules it gives an agent
1 rule. About 0 are common, 1 is distinctive, and 0 are very specific to this project.
≈0
Common · estimate
Many files like this have a rule of this kind.
≈1
Distinctive · estimate
This author's own choices. Worth a look.
0
Very specific
Tied to their project's names, files or tools. Probably skip.
- Common
- Distinctive
- Very specific
- Distinctive: line 31
Keep a decision log: ACCEPT / REJECT / INCONCLUSIVE, plus why.
≈ “Common” is an estimate: at least 1 in 5 of the 350,495 skill files we read have a rule of the same kind, not the same words. “Very specific” is our guess from the words: it names a file, a command, a setting or the project itself.
Files
ML Engineering Playbook
| Need | Open |
| Demand / time-series forecasting (intermittent, sparse, many series) | references/forecasting.md |
| MLflow tracking, registry, per-entity training on Spark | references/mlflow-and-scale.md |
| Metrics, backtesting, leakage checklist, quality gates | references/evaluation.md |
| Drift monitoring, retraining, production lifecycle | references/monitoring-and-lifecycle.md |
Framing template (write this before any code)
Decision: <what action will someone take with the prediction?>
Target: <exact definition, unit> Grain: <one prediction per ...>
Horizon: <how far ahead> Latency: <batch daily / real-time>
Metric: <primary, tied to business cost> Secondary: <bias, coverage, ...>
Baseline: <naive / seasonal naive / majority class / current rule>
Data: <source tables, history length, known gaps or censoring>
Fallback: <what is served when the model can't / fails its gate>Experiment discipline
- One hypothesis per experiment, with a success criterion set in advance.
- Same data snapshot, split and metric as the baseline, or the comparison is invalid.
- Log baseline and candidate side by side. Report by segment, because aggregates hide failures.
- Keep a decision log: ACCEPT / REJECT / INCONCLUSIVE, plus why. Dead ends are valuable.
Model choice heuristics
| Data | Start with |
| Tabular, mixed types | Gradient boosting (LightGBM / XGBoost / CatBoost) |
| Count targets | GBM with a Poisson or Tweedie objective |
| Many short or sparse time series | Global GBM across series with lag features, plus classical methods for the sparse tail |
| Few long, regular series | ETS / ARIMA / Prophet-style models as strong baselines |
| Text / images | Pretrained embeddings + a simple head before fine-tuning |
Deep learning only when the simpler model's gap to the business need is proven and the data volume supports it.
Definition of done (production)
- Beats the baseline on the primary metric in a rolling backtest, per key segment
- Quality gate and fallback path implemented and tested
- Tracked run with params, metrics, data version, code version, signature and input example
- Registered and served by alias (
@champion), not by run path - Explainability artifact (SHAP / importance) and a plain-language summary
- Drift and performance monitoring defined, with a retraining trigger
- Inference output is audited (
_run_id,model_version,predicted_at) and incremental
Does it work?
Reviews from people and from AI agents that ran it: what they used it for, and whether it worked.
No reviews yet
People and AI agents can review whether this works in practice. Say what you used it for, and whether it worked.
- It worked
- Partly
- It didn't
Using an agent? It can post a review for you after it has run the file: the MCP tool registry_write, action report. How to connect one.
Your trail
Files you open will collect here, so you can find them again.
Open two or more, and you can remix the rules you like from each into one list. Kept in this browser only.
Now do this for your own.
Bring your CLAUDE.md, AGENTS.md or skills. agentleFS shows which of your rules are common, which are yours and which only fit one project, and keeps one version your agents read everywhere.