Question about Future Feature Support Implementation

[cloudoor]

I noticed in the updates section that TimeMixer now supports using future temporal features for prediction through the use_future_temporal_feature parameter. However, I couldn't find detailed documentation or examples about:

1. The specific format required for future temporal features
2. How to properly configure the use_future_temporal_feature parameter
3. Any performance impact or benchmarks when using future features

Environment

• TimeMixer version: Latest from main branch
• Python version: 3.8
• Operating System: Ubuntu 20.04

Questions

1. Could you provide a minimal example showing how to prepare and input future temporal features?
2. Are there any limitations on the types of future features that can be used?
3. What's the recommended way to validate that future features are being correctly utilized by the model?

I'm working on a forecasting project where known future events/features could potentially improve prediction accuracy. Understanding the proper implementation of this feature would be very helpful.

Thank you!

[kevinliu2000]

Thank you for your interest in this work and for your advice! We have a number of issues to address, so it might take us some time to work through them. The issue you mentioned is quite common in time series analysis, and there are examples in many classic datasets that you can refer to. I'll work on compiling relevant documentation based on your suggestions. Thank you for your understanding and support!