Data & drivers

Feature engineering

Feature engineering turns raw timestamps and series into the lags, rolling stats, and calendar flags a model can use. In time series, the unforgivable sin is building those features with the future mixed in.

A good feature pipeline is boring and scheduled. The same code must run at 6am inference and at training. Notebooks that 'just compute the month total' will leak.

Start with lags, a seasonal lag, a short rolling mean, and a calendar. Add covariates you can actually know. Stop when the backtest stops moving.