Sequence-Based Forecast Bias Correction
Learning forecast corrections from simulated time series with a PyTorch transformer.
Why it matters
Simulated forecast errors provide a controlled setting for examining learned corrections over time. The example demonstrates sequence modeling without claiming operational forecast accuracy.
My contribution
I built a PyTorch transformer that learns corrections for simulated forecast sequences.
Result
The source code and demo page show the sequence-modeling example.
Tools
- Python
- PyTorch
- Transformers
- Time series
This simulated-data demo is separate from the space-time benchmark and is not an operational forecast model.
Project gallery
Synthetic demonstration figures; these do not show field performance. 5 images. Select an image to view it full size.
Technical details
Independent portfolio demonstration
Simulated forecast errors provide a simple way to test learned corrections.
- Create simulated forecast and reference sequences.
- Train a PyTorch transformer to learn the corrections.
- Show the workflow and outputs in a public demo.