Forecasting Time Series Data at scale with the TICK stack
summary
A workflow uses the TICK stack (Telegraf InfluxDB Chronograf Kapacitor) to train Prophet forecasts for ~400 GitHub repo star series, instantiate ~1,200 task templates, and compare each against two baselines (mean and Holt‑Winters) using MAPE; one run shows a Prophet outlier at ~250% error fixed by lowering the change point prior scale to ~5% error, tasks re-run in seconds, and Kapacitor alerts and dashboards automate surfacing which forecasts actually need human attention rather than hand‑tuning every model.
AI-written summary · about
chapters (15)
- 0:00Introduction
- 1:03Time series forecasting
- 3:50Forecasting at scale challenges
- 4:31Prophet algorithm workflow
- 8:03Dataset and goals
- 11:41TICK stack intro
- 13:49InfluxDB data storage
- 15:57Kapacitor forecasting tasks
- 18:26TICKscript and templates
- 21:37Model performance with MAPE
- 23:23Problematic forecasts
- 26:14Fixing misconfigured models
- 29:27Refining forecasts
- 30:32Key learnings
- 33:41Resources and closing
related talks
39mProphet at scale to tune & forecast time series at Spotify
40mMulti-Series Forecasting at Scale with StatsForecast
44mMethods and Tools for Time Series Data Science Problems
19mForecasting for Everyone - Foundation Models and the Future of Time Series Prediction
30mMethods & Tools For Time Series Data Science Problems With InfluxDB
39mProductionising Data Science at Scale
06mLightning Talk: Predictive Autoscaling in Kubernetes With KEDA and Prophet
45mForecasting at Scale