1Z0-184-25 Question 62
Single answerYou are implementing an e-commerce product similarity feature on Oracle Database 23c running in OCI. Your table includes a 512-dimensional vector column to store product embeddings, and you have created an IVF (Inverted File) index to enable approximate nearest neighbor (ANN) queries. You aim to speed up product searches while maintaining a reasonable level of accuracy. Which configuration best ensures a practical balance between lower latency and acceptable recall in your IVF-based ANN queries?
- A
Increase the number of centroids in the IVF index to the maximum supported value while keeping the search parameter (nprobe) at its default setting.
- B
Use a moderate number of centroids in the IVF index and configure a moderate nprobe value when issuing queries, adjusting both parameters according to performance needs.
- C
Create multiple IVF indexes with different centroid configurations and always query them all simultaneously for the same vectors.
- D
Keep all default IVF settings and rely on the database to dynamically adjust indexing and querying parameters automatically.
Show answer and explanation
Correct answer: B
Explanation
When using IVF vector indexing in Oracle Database 23c for approximate nearest neighbor searches, tuning both the number of centroids (defined during index creation) and the nprobe (specified during queries) is crucial to optimize performance. Using too many centroids or a very high nprobe can degrade query speed, while too few centroids or a low nprobe can reduce accuracy. For more details, refer to Oracle Database 23c Vector Search documentation, which provides guidance on configuring IVF indexes and query parameters to achieve a balanced approach for large-scale vector similarity searches.
- A. Incorrect.
Option 1: Increasing the number of centroids excessively might improve recall but often leads to higher memory usage and slower indexing/querying. Leaving nprobe at its default may not provide the best trade-off if your dataset is very large.
- B. Correct.
Option 2: Correct. A moderate number of centroids combined with adjusting nprobe at query time can strike a balance between speed and accuracy. You can fine-tune these parameters based on your workload and desired recall.
- C. Incorrect.
Option 3: Creating multiple IVF indexes for the same data with different centroid configurations can complicate query logic and maintenance. Typically, one well-tuned IVF index suffices for most ANN use cases.
- D. Incorrect.
Option 4: Relying exclusively on defaults does not guarantee an optimal trade-off. You generally need to fine-tune both the index and query parameters to match your specific performance and accuracy requirements.