[⭐] Add to /experiments a predict.py and forecast.py

[jacoboggleon-bbva]

Description of the new feature / enhancement | 您所需的新功能或能力

The goal is to add scripts (predict.py and/or forecast.py) to the /experiments directory to enable inference capabilities for trained models in BasicTS. These scripts would allow users to:

1. Perform one-step predictions using .predict.
2. Perform multi-step forecasting over a window using .forecast.

Scenario when this would be used? | 使用场景

1. Production Inference:
   Users can generate predictions for real-time applications, such as forecasting weather conditions or other time series data.
2. Testing New Data:
   Users can test their trained models on unseen sequences, enabling a practical evaluation of model performance in various scenarios.
3. Pipeline Integration:
   The scripts would integrate seamlessly with other BasicTS components, such as training and evaluation, for end-to-end workflows.

Supporting information | 附加信息

• Functionalities:

  • predict.py: For one-step predictions, taking an input sequence and returning the next predicted value.
  • forecast.py: For multi-step forecasting, taking an input sequence and generating predictions over a user-defined window.

• Inputs and Outputs:

  • Inputs: Processed data sequences (e.g., .npz or .csv files).
  • Outputs: Predictions saved as .npz files or printed to the console for user inspection.

• Proposed API:
  # One-step prediction
python experiments/predict.py --cfg baselines.PatchTST.Weather --input /path/to/input.csv --output /path/to/output.csv

  # Multi-step forecasting
python experiments/forecast.py --cfg baselines.PatchTST.Weather --input /path/to/input.csv --forecast-steps 168 --output /path/to/output.csv

• References

  • Training script: /experiments/train.py.
  • Evaluation script: /experiments/evaluate.py.

Adding these scripts would enhance BasicTS’s functionality by providing a complete workflow for time series prediction and forecasting.

[zezhishao]

Thank you very much for your detailed proposals. This is very meaningful! We will add this feature as soon as possible.