MLOps Academy · 课时

使用 Feast 定义特征视图

在注册表中声明实体和特征

使用 Feast 定义特征视图 是 CoddyKit 上的免费 MLOps Academy 课时。这是第 2 节课，共 4 节。你可以在下方免费阅读本课时的完整内容 — 然后在浏览器中使用内置代码编辑器和全天候 AI 导师进行实践。这是 MLOps Academy 学习路径的一部分，你的进度在网页和 CoddyKit 应用中同步。 MLOps Academy 课程共包含 4 节课。

本课时的部分内容尚未翻译，以英文显示。

The Building Blocks

In Feast you describe features with code: an Entity, a data source, and a Feature View. Get these three right and the rest follows. 🏗️

What an Entity Is

An Entity is the thing your features describe, like a driver or a customer. It is the key you will look features up by later.

driver = Entity(name="driver", join_keys=["driver_id"])

Point at Your Data

A data source tells Feast where the raw feature values live, such as a Parquet file or a warehouse table.

src = FileSource(path="driver_stats.parquet", timestamp_field="event_timestamp")

The Feature View

A FeatureView groups related features for one entity and links them to a source. It is the heart of a Feast definition.

Declaring Fields

Inside a Feature View, each Field names a feature and its data type, so Feast knows the schema it will serve.

Field(name="conv_rate", dtype=Float32)

Putting It Together

One Feature View ties the entity, schema, and source into a single object Feast can manage. 🔗

fv = FeatureView(name="driver_stats", entities=[driver], schema=[Field(name="conv_rate", dtype=Float32)], source=src)

The TTL Window

A Feature View's ttl sets how long a value stays valid. Past that window, Feast treats the feature as missing rather than stale.

ttl=timedelta(days=1)

The Timestamp Column

Every source needs an event timestamp column. It records when each value was true, which makes time-correct lookups possible.

The Feature Repo

Your definitions live in a feature repo folder with a feature_store.yaml config. Feast scans Python files there to find your objects.

Apply the Definitions

Running feast apply registers your entities and feature views into the registry. Nothing is live until you apply.

feast apply

The Registry

The registry is Feast's catalog of every entity and feature view. Teammates query it to discover what features already exist. 📒

Quick Check

Which object does a Feature View group features around?

Recap

You defined Feast objects: an Entity, a source, and a Feature View with fields and a ttl, then ran feast apply to register them. Next, you'll serve them online.

常见问题解答

「使用 Feast 定义特征视图」课时是免费的吗？

是的 — 「使用 Feast 定义特征视图」的完整文本可在网页上免费阅读。要进行交互式练习（内置代码编辑器和全天候 AI 导师）并解锁 MLOps Academy 课程的其余内容，请升级到 CoddyKit PRO。 MLOps Academy 课程共包含 4 节课。

「使用 Feast 定义特征视图」这节课中我会学到什么？

在注册表中声明实体和特征你通过在浏览器中直接运行的动手代码来练习 MLOps Academy，全天候 AI 导师会在你学习这节课的过程中回答你的问题。

学习 MLOps Academy 需要有经验吗？

无需任何先前经验。CoddyKit 上的 MLOps Academy 课程适合初学者到高级学习者，你可以从这里开始或从头开始，按照自己的节奏学习。这是第 2 节课，共 4 节。

「使用 Feast 定义特征视图」课时需要多长时间？

大多数 CoddyKit 课程大约需要 5–10 分钟。每节课都很精短且互动，所以你能稳步进步，并在网页和应用中从离开的地方继续。

我能在这节 MLOps Academy 课中编写并运行代码吗？

能。每节 MLOps Academy 课都包含内置代码编辑器，你可以在浏览器中直接编写并运行真实代码，并获得即时 AI 反馈 — 无需本地设置。

此课程中的所有课时

← 返回 MLOps Academy