Features on event tables (vp.features)

Per-entity features on an event table, called on a pandas DataFrame.

Each feature is checked to read no row later than the one it is computed for.

Part of the distribution vipyr; it runs from a checkout of the repository. Planned on PyPI as pip install vipyr.

Functions

vp.features.delinquency

Days past due, worst in 12 months, ever cured after 60+ (catalog/delinquency.viv).

vp.features.ewma

Gap-aware EWMA with resets (catalog/ewma.viv).

vp.features.sessions

Gap-based sessions per user (catalog/sessions.viv).

vp.features.since_last

Seconds since the last event of a kind, and of any kind (catalog/since_last.viv).

vp.features.streaks

Consecutive active days (catalog/streaks.viv).

vp.features.velocity

Trailing-window counts, sum and distinct count (catalog/velocity.viv).

Examples in the gallery

• Customer features from a transaction log