How omnimahui/thetadata-spx works
This repository contains SPX options feature engineering, a Temporal Fusion Transformer training workflow, an options-analysis notebook, and a reusable trend-scanning label function. The visible feature script reads a prebuilt one-minute parquet dataset and writes engineered features. The training excerpt shows preprocessing, time-series dataset construction, model training or weight loading, and checkpoint copying. The notebook’s contents and the training data-loading section are not available, so their wiring is left unspecified.
Main entry points: train.py TFT model · thetadata-spx-features.py SPX feature job · trend_scanning_labels.py Trend-scanning labels
Components 12
Feature engineering
- SPX feature job
thetadata-spx-features.py - Engineered features
Model training
- Training data
- Prepare time series
train.py - Time-series datasets
train.py - TFT model
train.py - Model checkpoint
Research tools
- Options analysis
thetadata-spx-options.ipynb - Trend-scanning labels
trend_scanning_labels.py - Trend labels
Other components
- Minute options data
- Price series
Connections 9
- Minute options data → SPX feature job: reads
- SPX feature job → Engineered features: writes
- Training data → Prepare time series: reads
- Prepare time series → Time-series datasets: builds
- Time-series datasets → TFT model: feeds
- TFT model → Model checkpoint: saves
- Model checkpoint → TFT model: loads weights
- Price series → Trend-scanning labels: labels
- Trend-scanning labels → Trend labels: returns