thuml/Time-Series-Library ? reverse-engineered prompt

Build me a Python library for deep time series analysis that I can use to train, compare, and run a bunch of modern models from one clean codebase. I want it to support the main tasks people usually need, like long term and short term forecasting, imputation, anomaly detection, and classification.

Please make it easy to add new models, easy to run experiments from scripts, and easy to reproduce results on common benchmark datasets. It should have a simple entry point for training and testing, plus clear examples and a beginner friendly tutorial so I can get started quickly. If there are any setup details or model specific dependencies, handle them cleanly and note them in the docs.

Use current best practices for the training code and make sure the project is organized so researchers can swap models without rewriting everything. If you need to check the latest docs for anything, look them up online.