NixtlaVerse, bridging the gap between statistics and deep learning for time series
summary
NixtlaVerse delivers concrete tooling and numbers: StatsForecast (PyTorch + NumBa) runs AutoARIMA/ETS at scale — claim is 10 million series in under 43 minutes for under $20 — and classical ARIMA beats Prophet by ~70% accuracy while being ~40x faster. The neural stack NeuralForecast exposes ES‑RNN, N‑BEATS, and N‑HiTS; N‑HiTS targets long‑horizon forecasting with ~25% improvement over a top Transformer entry and roughly 50x less compute, and libraries are reproducible and benchmark‑focused.
AI-written summary · about
chapters (11)
- 0:00Introduction
- 0:45Two mountains of forecasting
- 3:31Why forecasting matters
- 7:22Classical statistical methods
- 13:00StatsForecast library
- 14:03Deep learning methods
- 17:11Bridging traditions
- 20:43N-HiTS architecture
- 24:56Nixtla ecosystem
- 25:37Hierarchical forecasting
- 27:01Q&A
related talks
28mNixtla: Deep Learning for Forecasting...
41mNeural Prophet – A powerful AI framework for Time Series Models
26mTime Series Analysis in Python: Easy Tools for Scientific Insight
40mMulti-Series Forecasting at Scale with StatsForecast
31mAdvanced machine learning time series....
19mForecasting for Everyone - Foundation Models and the Future of Time Series Prediction
23mComplex AI Forecasting Methods for Investments
42mRobust Automated Forecasting in Python and R