• Docs
  • Team
  • API
  • Help

Kats

One stop shop for time series analysis at Facebook

Forecasting

Detection

TSFeatures

Multivariate

Utilities

a toolKit to Analyze Time Series

Kats, a toolKit to analyze time series data, a light-weight, easy-to-use, and generalizable framework to perform Time Series analysis. Time Series analysis is an essential component of Data Science and Engineering work at Facebook, from understanding the key statistics and characteristics, detecting regressions and anomalies, to forecasting future trends. There are a few internal and external packages that can perform certain analyses, while a unified in-house framework is desired. Kats aims to provide the one-stop shop for time series analysis, like detection, forecasting, feature extraction, multivariate analysis, etc.
Learn More about Kats

Forecasting

Kats provides a set of tools to make forecasting easier and better for data scientists and developers. There are various forecasting models we support, from statistical models including ARIMA, S-ARIMA, Holt-Winters, to Machine Learning models such as AR-Net, LSTM, as well as ensemble based approaches. We also provide toolkits for back testing and param tuning for those forecasting models.

Detection

Kats provides algorithms for outlier detection, seasonality detection, anomaly detection, change point and regression detection, which covers the majority tasks in time series detection domain. We’re actively working on implementing the state-of-the-art methods in this domain

TSFeatures

In time series analysis, there is often the need for characterizing a time series using a set of meaningful features. Example features include strength of seasonality, strength of trend, spikiness, amount of level shift, presence of flat segments, to name a few. The features are usually used for identifying similar/outlying time series from a large number of samples. Kats provides APIs to calculate such features in a scalable way.

Multivariate

Multivariate time series have more than one time-dependent variable, where those time series might inter-correlated with each other. Learning such inter-dependencies can improve the performance of many time series analysis tasks including forecasting and detection. Kats currently provides multivariate models including VAR and AR-Net, we also support algorithm for multivariate anomaly detection.

Visualization

Coming soon...

Utilities

To complete the ecosystem, we also built a set of utilities including time series decomposition, dataswarm operator, tools for flexible backtesting, as well as hyper-parameter tuning methods.

The Team

Xiaodong Jiang

Research Data Scientist

Sudeep Srivastava

Research Data Scientist

Yanjun Lin

Data Scientist

Caner Komurlu

Research Data Scientist

Rakshita Nagalla

Data Scientist

Zhichao Wang

Research Data Scientist

Peng Gao

Software Engineer

Rohan Bopardikar

Data Scientist

Sourav Chatterjee

Research Data Scientist

Ahmet Koylan

Front End Engineer

Kats Project
ForecastingDetectionTSFeaturesMultivariateVisualizationUtilities
Kats Community
User ShowcaseStack OverflowProject ChatTwitter
More
BlogGitHubStar
Facebook Open Source
Copyright © 2021 Kats Project @ Facebook