AI Snips
Chapters
Transcript
Episode notes
Boosted Embeddings Remove Seasonality
- Boosted embeddings remove seasonal and categorical effects yielding a purified time series for modeling.
- Freezing embedding weights while successively removing categorical impacts fits naturally into a boosting framework.
Use DeepGB As Preprocessor
- Use DeepGB to compute robust embeddings removing nonlinear seasonal components and covarying categorical effects.
- After embedding, fit any forecasting model like CatBoost or ARIMA modularly on the adjusted time series.
Hard Seasonal Variation Anecdote
- Sankeerth describes struggles forecasting promotional events with shifting dates using ARIMA.
- DeepGB effectively handles such non-stationary, aberrant seasonality by iterative embedding removal of covariates.