FeatureSet¶

A named group of features for reuse across builds and audits.

`FeatureSet(name, features)` ¶

A named group of features for reuse across models.

A FeatureSet is a flat list with a name. No nesting, no inheritance.

Source code in src/timefence/core.py

def __init__(
    self,
    name: str,
    features: Sequence[Feature],
):
    self.name = name
    self.features = list(features)

Why use FeatureSet¶

When you have a stable set of features that get reused across multiple models or experiments, group them into a FeatureSet instead of managing individual lists. This gives you:

A single named reference for a group of features
Version tracking — name the set "churn_features_v2" and swap it in one place
Composability — combine FeatureSets with additional features in build() and audit()

Example¶

import timefence

# Define individual features
rolling_spend = timefence.Feature(source=transactions, columns=["amount"], embargo="1d", name="spend")
user_country = timefence.Feature(source=users, columns=["country"])
login_count = timefence.Feature(source=logins, columns=["login_count"])

# Group into a named set
base_features = timefence.FeatureSet(
    name="churn_features_v1",
    features=[rolling_spend, user_country, login_count],
)

# Use directly in build() — Timefence flattens it automatically
result = timefence.build(
    labels=labels,
    features=[base_features],
    output="train.parquet",
)

Combining FeatureSets with individual features¶

You can mix FeatureSets and individual Features in any call:

# Add an experimental feature alongside the base set
result = timefence.build(
    labels=labels,
    features=[base_features, new_experimental_feature],
    output="train.parquet",
)

# Audit with the same features
report = timefence.audit(
    data="train.parquet",
    features=[base_features],
    keys=["user_id"],
    label_time="label_time",
)

Iteration and length¶

FeatureSet supports standard Python iteration:

print(len(base_features))  # 3

for feature in base_features:
    print(feature.name)  # "spend", "country", "login_count"

Parameters¶

Parameter	Type	Description
`name`	`str`	Human-readable name for this group.
`features`	`Sequence[Feature]`	List of Feature objects.

Note

FeatureSets are flat — they contain only Feature objects, not other FeatureSets. There is no nesting.