TimeMixer is an official ICLR 2024 implementation for time series forecasting, offering a fully MLP-based architecture designed to achieve state-of-the-art performance in both long-term and short-term forecasting. It targets researchers and practitioners in time series analysis seeking efficient and accurate predictive models.

How It Works

TimeMixer utilizes a novel Past-Decomposable-Mixing (PDM) block to separately process seasonal and trend components across multiple scales. This approach leverages the observation that different time series scales exhibit distinct characteristics, allowing for more effective extraction of past information. The Future-Multipredictor-Mixing (FMM) block then integrates forecasts from these mixed multiscale series, combining complementary predictive capabilities for improved accuracy.

Quick Start & Requirements

• Install via pip install -r requirements.txt. Note: For Python 3.8, change sktime version to 0.29.1 in requirements.txt.
• Datasets are available via Google Driver, Baidu Driver, or Kaggle Datasets.
• Training can be initiated using provided scripts, e.g., ./scripts/long_term_forecast/ETT_script/TimeMixer_ETTm1.sh.
• Official paper: Paper Page
• Video explanation: ICLR Video

Highlighted Details

• Achieves consistent state-of-the-art performance across 18 real-world benchmarks for both long-term and short-term forecasting.
• Demonstrates favorable efficiency in terms of GPU memory and running time compared to recent SOTA models.
• Supports using future temporal features for prediction (parameter use_future_temporal_feature).
• Includes a time-series decomposition method based on DFT and downsampling via 1D convolution.

Maintenance & Community

• The project has been integrated into PyPOTS and NeuralForecast.
• An upgraded version, TimeMixer++, has been released, supporting 8 diverse analytical tasks.
• Contact information for current maintainers is provided.

Licensing & Compatibility

• The repository does not explicitly state a license in the README.
• Compatibility for commercial use or closed-source linking is not specified.

Limitations & Caveats

The README does not specify any limitations or known bugs. The project is presented as a mature implementation with extensive experimental validation.