Imputation
Stage 2 of the pipeline: Reconstructing missing or invalid signal segments.
Available Methods
Ground Truth
| Method |
Description |
pupil-gt |
Human-corrected ground truth signal |
Deep Learning
| Method |
Description |
SAITS |
Self-Attention Imputation for Time Series |
CSDI |
Conditional Score-based Diffusion Imputation |
TimesNet |
TimesNet-based imputation |
Foundation Models
| Method |
Description |
MOMENT-finetune |
MOMENT model finetuned for imputation |
MOMENT-zeroshot |
MOMENT zero-shot imputation |
Traditional
| Method |
Description |
linear |
Linear interpolation |
Ensembles
| Method |
Description |
ensemble-CSDI-MOMENT-SAITS-TimesNet |
Ensemble of deep learning methods |
Configuration
# In Hydra config
imputation_method: SAITS
Evaluation Metrics
| Metric |
Description |
| MAE |
Mean Absolute Error vs ground truth |
| RMSE |
Root Mean Square Error vs ground truth |
API Reference