A unified toolkit for time series work
Kats is a Python framework that consolidates the main tasks of time series analysis into one library: forecasting, anomaly detection, multivariate analysis, and feature extraction. Announced by Facebook Research, it is positioned as the first comprehensive library of its kind for generic time series work, offering both classical models and newer techniques in a single, accessible package.
Core capabilities
The library organizes its algorithms around four functional areas:
Forecasting
Kats ships with more than ten individual forecasting models, along with ensemble methods and a self-supervised meta-learning approach. Practical tooling for forecasters is included as well, such as backtesting, hyperparameter tuning, and empirical prediction intervals.
Pattern detection
For detection tasks, Kats supports identifying seasonality, outliers, change points, and slow trend shifts within a series. This covers both abrupt and gradual changes in data over time.
Feature extraction and embedding
The TSFeature extraction module produces 65 statistically defined features per series. These features can be fed directly into standard machine learning pipelines for classification or regression tasks.
Why it matters
Time series analysis underpins workflows across e-commerce, finance, capacity planning, supply chain, medicine, weather, energy, and astronomy. Previously, practitioners had to stitch together disparate tools for different tasks. By consolidating forecasting, detection, and feature engineering into one framework, Kats aims to establish a common standard and let data scientists explore the characteristics of their series, project future values, monitor for anomalies, and integrate the outputs into broader ML models without switching contexts.
