L-GTA Extension: Adaptive Augmentation Controller (AAC)
L-GTA re-implementation with a CVAE–GRU based Adaptive Augmentation Controller for feature-aware time-series augmentation.
Technologies / Concepts Used
- Python
- TensorFlow
- NumPy
- SciPy
- scikit-learn
- statsmodels
- Matplotlib
- Jupyter Notebook
- Time Series Analysis
Key Features
- CVAE-based L-GTA with GRU encoder–decoder for time-series augmentation
- Feature-aware Adaptive Augmentation Controller (AAC) with dynamic policy selection
- Latent-space perturbations with adaptive intensity tuning per series
- Automatic statistical feature extraction (trend, variance, autocorrelation, shape)
- Robust support for non-stationary, seasonal, and noisy datasets
- Comprehensive visual evaluation including reconstructions, PCA, ACF, and distribution shifts
- Validated on real datasets such as Tourism, M5 (Walmart), and Police crime data