Deep learning library for time series forecasting, imputation, classification and anomaly detection.
Evaluate and build deep learning models for time series forecasting, imputation, and classification.
Time-Series-Library (TSLib) is an open-source codebase designed for researchers and engineers working with deep time series analysis. It covers mainstream tasks like long- and short-term forecasting, imputation, anomaly detection, and classification.
The library implements advanced models like TimeXer, TimeMixer, iTransformer, PatchTST, and TimesNet, organized around specific tasks and benchmarks. It includes specialized features for zero-shot forecasting with Large Time Series Models (LTSMs) and exogenous variable integration.
It suits data scientists and machine learning engineers developing or benchmarking deep learning models for sequential data. It is not designed for business analysts seeking no-code graphical forecasting interfaces.
Covers long- and short-term forecasting, imputation, anomaly detection, and classification.
Includes implementations of TimeXer, iTransformer, PatchTST, TimesNet, and Mamba.
Features experimental scripts for evaluating Large Time Series Models without prior training.
Implements paradigms like TimeXer to handle external variables efficiently during forecasting.
Free under its MIT licence, with no paid hosted option provided. The only cost is running the hardware required for deep learning training.
880 open remote AI-training roles, updated November 15, 2026; most pay $20–$175 an hour. Zekai lists them free; the application goes to the employer’s platform.
TSLib supports long- and short-term forecasting, imputation, anomaly detection, and classification.
No, maintainers announced that active feature additions are paused due to limited bandwidth.
See Zekai first in Google