# Feature Engineering

> The work of turning raw observations into model inputs that represent the signal needed for a prediction.

- HTML version: https://robbiepalmer.me/ideas/feature-engineering
- Source: https://developers.google.com/machine-learning/glossary#feature_engineering

Feature engineering turns raw data into the variables a model consumes. It includes selecting
signals, encoding categories, scaling values, aggregating events, defining time windows, and
combining observations into quantities that match the decision being modelled.

The useful feature is rarely the field that happened to be easiest to query. "Logins in the last
seven days" embeds a choice about identity, time, missing records, and the moment at which a
prediction is made. Those choices carry domain assumptions into the model and often matter more
than changing the learning algorithm.

Learned representations reduce some manual transformation work, but they do not remove the need to
define inputs, labels, sampling, or time boundaries. A feature pipeline must apply the same logic
during training and inference. It must also prevent leakage by calculating each historical value
using only information that would have existed at that point in time.

## Questions it prompts

* What real-world signal should this value represent?
* Was the feature available when the prediction would have been made?
* Do training and inference apply the same transformation?
* How will missing, delayed, or revised source data change the value?

## Related ideas

- [Context Engineering](https://robbiepalmer.me/ideas/context-engineering.md): The design and continual selection of the information an AI model receives so it can act well within a limited context window.
- [Feature Stores](https://robbiepalmer.me/ideas/feature-stores.md): Systems that define, retrieve, and serve machine-learning features consistently for training and inference.

## Where it appears

- Technology: [Google BigQuery](https://robbiepalmer.me/technologies/google-bigquery.md)

---

Markdown index of this site: https://robbiepalmer.me/llms.txt