November 2022 Summaries

Weaviate 1.16 introduced Ref2Vec, a module that vectorizes data objects based on their cross-references to other objects. This approach allows for characterization from actions and relationships, with potential applications in recommendation systems and representing long objects. In the context of recommendations, Ref2Vec can represent users by aggregating vectors of products they have engaged with, allowing Weaviate to learn from user preferences and provide up-to-date characterizations. Additionally, Ref2Vec can help address the challenge of finding suitable representations for long objects in search technology.

Weaviate has introduced a backup feature in version 1.15 that streamlines the backup process for data, whether it be to a local file system or to a cloud storage provider. This tutorial guides users through backing up and restoring their data using this new feature. The tutorial covers creating two instances of Weaviate, populating one with a schema and data, backing up the instance, and restoring the backup data to another instance. It also discusses additional backup options such as local backup location, cloud storage systems, partial backups and restore, and asynchronous operations.

Weaviate 1.16 introduces new filter operators, distributed backups, ref2vec-centroid module, Node Status API, support for Azure-issued OIDC tokens, and patch releases ready sooner. The new filter operators allow users to filter data based on null values or array lengths. Distributed backups enable backup of objects stored in multi-shard configuration across multiple nodes. Ref2vec-centroid module calculates a mean vector of referenced objects. Node Status API checks the health of running clusters, and support for Azure-issued OIDC tokens allows authentication with Azure, Keycloak, or Dex OIDC tokens. Patch releases are now published as soon as new important fixes are available.