Time Series Forecaster Skill for Weekly Orders in Claude Code
Compare a seasonal baseline and an ARIMA-family model with chronological holdout evaluation before forecasting future weeks.
中文Install in Claude Code
Use a synthetic series of 104 consecutive weekly order counts, Python, pandas, and statsmodels. Keep the week boundary consistent.
- Get the complete `plugins/ai-ml/time-series-forecaster/skills/forecasting-time-series-data` directory from the pinned GitHub commit.
- Copy it into the target project's `.claude/skills/`. Check `.claude/skills/forecasting-time-series-data/SKILL.md`, start Claude Code at the project root, and call `/forecasting-time-series-data`.
Example task
Build a two-year series with a 52-week pattern. Hold out the final eight weeks and compare a prior-year baseline with a seasonal model. Record MAE, then forecast four future weeks with the selected method.What to check
Deliver holdout MAE for both methods and four future rows with dates, estimates, and prediction intervals. Keep the fixture and settings reproducible.
Original SKILL.md on GitHub · Updated
Frequently asked question
Why avoid a random train and test split for these 104 weeks?
Chronological evaluation mirrors forecasting with data available at the time and keeps later observations outside the training window.