Skip to content

Releases: feast-dev/feast

Feast v0.2.0

24 Nov 07:53
Compare
Choose a tag to compare

This release makes major changes to data processing in the ingestion layer of Feast. By decoupling ingestion into two phases, Feast v0.2.0 allows Feast to support decentralized serving with centralized feature management.

Change summary:

  • A centralized stream is introduced in Feast for managing state for all consuming Feast deployments.
  • Ingestion is decoupled and now produces data to the stream instead of starting jobs that directly ingest to stores.
  • Feast serving deployments consume from this central feature stream. This consumption is handled by long running jobs.

Feast v0.1.2

23 Aug 06:22
Compare
Choose a tag to compare
Feast v0.1.2 Pre-release
Pre-release

This release adds filtering capability when retrieving feature values from Feast Warehouse.

Given the following entity and features names:

entity: person
feature_names:
- age
- height
- address_city
- address_country

This release allows the retrieval of feature values for person entity from Warehouse where address_city = Bandung, as opposed to retrieving all the feature values for all address_city and performing the post-filtering at the client side.

Docker images:

gcr.io/kf-feast/feast-core:0.1.2
gcr.io/kf-feast/feast-serving:0.1.2

Python SDK:

pip install feast==0.1.2

Feast v0.1.1

29 Apr 03:35
Compare
Choose a tag to compare
Feast v0.1.1 Pre-release
Pre-release

This release contains an update to Feast Core:

  • Fix BigQuery query template to retrieve training data
    Previous query template has column names and full table id with incorrect values, resulting in Feast Core failing to retrieve training data from BigQuery.

Docker images:

gcr.io/kf-feast/feast-core:0.1.1
gcr.io/kf-feast/feast-serving:0.1.1

Feast v0.1.0

09 Apr 10:24
Compare
Choose a tag to compare
Feast v0.1.0 Pre-release
Pre-release

Feast version 0.1.0 mainly contains updates to unit and integration tests and various validations and checks to catch runtime exceptions.

In the feature spec side, it is now optional to include store information so that a feature spec only contains feature name, data types and owner, without attaching store information which may not be relevant/known to feature creators.

This release includes:

  • Feast Docker images:
    gcr.io/kf-feast/feast-core:0.1.0
    gcr.io/kf-feast/feast-serving:0.1.0
  • Feast CLI binary attached in this release
  • Feast Python SDK via pypi:
    pip install -U feast==0.1.0.*

Please refer to the documentation in the repo for more details.