1Z0-184-25 Question 42
Select 2Your team is implementing a product recommendation feature in Oracle Autonomous Database. You store thousands of item embeddings in a dedicated VECTOR column to support semantic similarity searches. You want to optimize both performance and accuracy for approximate nearest neighbor queries. Which two statements are correct about designing and maintaining vector indexes in this scenario?
- A
- It is recommended to store embeddings in a VECTOR column specifically designed for vector data, ensuring efficient indexing and query performance.
- B
- You must completely re-run your entire data ingestion pipeline whenever you add or remove a dimension from existing embeddings, or else the vector index cannot be updated.
- C
- You can create multiple vector indexes on the same table if you want to search the same embeddings with different distance metrics (e.g., Euclidean vs. cosine similarity).
- D
- No refresh or rebuild operations are ever required after inserting a significant number of new embeddings into a table protected by a vector index.
Show answer and explanation
Correct answers: A, C
Explanation
Storing embeddings in Oracle� native VECTOR data type is a recommended best practice for semantic search and approximate nearest neighbor queries in Autonomous Database. Multiple indexes can be created when different distance metrics are required, providing flexibility in various similarity search use cases. However, vector indexes may need either automatic or scheduled reindexing depending on your data refresh patterns, ensuring optimal performance as new embeddings are introduced. Refer to Oracle Database 23c documentation for detailed guidelines on creating, maintaining, and querying vector indexes.
- A. Correct.
- Correct. Using Oracle's VECTOR data type (introduced in Oracle Database 23c and available in Autonomous Database) is a best practice because it's optimized for storing and indexing high-dimensional embeddings. This improves the performance of approximate nearest neighbor queries and reduces storage overhead.
- B. Incorrect.
- Incorrect. You do not need to re-run the entire ingestion pipeline each time the dimensionality changes. While adding or removing embedding dimensions usually requires some schema adjustment, Oracle does not force a complete data reload for every change. Proper schema design and incremental updates can accommodate evolving embeddings without re-ingesting all data.
- C. Correct.
- Correct. You can create more than one vector index on the same table if you want to enable searches using different distance metrics. For example, you might want to compare embeddings using cosine similarity in one context and Euclidean distance in another. Multiple indexes let you tailor your similarity queries efficiently.
- D. Incorrect.
- Incorrect. Although vector indexes may be maintained automatically for smaller incremental data changes, you often need to schedule maintenance or refresh operations if large volumes of embeddings are inserted or updated. Failing to refresh or rebuild can lead to suboptimal query performance.