Examples

1. 1.

   For example, you might determine that temperature might be a useful feature. Then, you might experiment with bucketing to optimize what the model can learn from different temperature ranges.

2. 2.

   Feature engineering is sometimes called feature extraction or featurization.

3. 3.

   Click the icon for additional notes about TensorFlow.

Real-world uses

Created for this library

1. 1.

   A pricing analytics team invests months in feature engineering for its demand model because well-designed features deliver more lift than swapping algorithms.

2. 2.

   A risk modeling team's feature engineering produces a stable set of ratios like debt-to-income that survive multiple model generations.

3. 3.

   A retail forecasting team's feature engineering includes lag features, rolling averages, and holiday indicators that improve accuracy more than the choice of model.