Tracking the tracker: Time Series Analysis in Python From First Principles
summary
Focuses on forecasting and anomaly detection using online state‑space methods implemented in Python with Kalman / state‑space models and incremental parameter updates, extended to nonlinear cases via unscented / sigma‑point filters. Applied to price prediction (Bitcoin vs S&P500) and Twitter popularity detection, the system runs in real time, adapts with forgetting factors, and converges quickly (claimed close fits with only a few sigma points, e.g. ~5). Notes practical caveats: stationarity assumptions and limited multivariate support in the current implementation.
AI-written summary · about
chapters (17)
- 0:00Introduction and speaker background
- 2:30Forecasting and anomaly detection overview
- 5:25Design goals for time series tools
- 7:12Real-time time series processing
- 10:50Stationarity in time series analysis
- 12:38Key components of time series models
- 14:30ARIMA and parameter selection
- 16:01State space models and Kalman filters
- 18:40Linear regression and Kalman filter basics
- 20:57Non-linear extensions and challenges
- 22:44Markov Chain Monte Carlo methods
- 24:33Stationarity assumptions and limitations
- 27:00Practical challenges in financial data
- 28:31Handling imbalanced datasets
- 30:02Concept drift and distribution changes
- 32:10Advanced topics and resources
- 33:25Q&A session
related talks
41mReliably forecasting time-series in real-time
26mTime Series Analysis in Python: Easy Tools for Scientific Insight
29mTime Series Forecasting Techniques in Python
24mMachine learning and financial time series analysis
39mForecasting with the Kalman Filter
42mSeanTaylorBenLetham
01h 27mBayesian Time Series
01h 43m(Serious) Time for Time Series