1Z0-184-25 Question 36
Single answerYou are helping a data science team optimize similarity searches on a large set of 768-dimensional embeddings stored in an Autonomous Database on OCI. They currently store the embeddings in a VECTOR column but notice that queries are running slowly. They ask for your advice on how to properly index the data to significantly improve approximate nearest neighbor search performance. Which recommendation should you provide?
- A
A. Use the default B-tree index on the VECTOR column and rely on Oracle� Automatic Indexing to accelerate similarity searches.
- B
B. Create an Approximate Nearest Neighbor vector index specialized for high-dimensional data, specifying appropriate search parameters and distance metric.
- C
C. Convert the VECTOR column to a CLOB-based JSON structure and create a JSON index for faster retrieval of embeddings.
- D
D. Rely exclusively on database partitioning for the VECTOR column, assuming partition pruning will optimize the similarity search.
Show answer and explanation
Correct answer: B
Explanation
To efficiently run nearest neighbor queries on high-dimensional embeddings, Oracle Database offers specialized vector index functionality. A dedicated approximate nearest neighbor (ANN) index can handle the complexity of similarity matching, which B-tree or JSON indexes cannot address effectively. For best practices, refer to the official Oracle Database documentation on vector indexes, including how to choose an appropriate distance metric (e.g., cosine or Euclidean) and tune indexing parameters for your data distribution.
- A. Incorrect.
Option A: Incorrect. A traditional B-tree index is not designed for high-dimensional vector similarity searches; it is best suited for exact matches and range-based queries.
- B. Correct.
Option B: Correct. Oracle Database offers specialized vector indexing capabilities that enable approximate nearest neighbor search on high-dimensional data. Configuring an ANN index (e.g., with the appropriate distance metric for embeddings) can dramatically improve performance.
- C. Incorrect.
Option C: Incorrect. Storing embeddings as CLOB JSON and using a JSON index is not an effective approach for vector similarity searches. JSON indexes handle structured or semi-structured text data, not specialized vector math.
- D. Incorrect.
Option D: Incorrect. While partitioning can help manage large data sets, it does not in itself optimize vector similarity queries. You still need a vector index for nearest neighbor functionality.