simple_similarity
Documentation for Vector Search Functions¶
This document provides an overview of several methods for generating PostgreSQL functions that facilitate vector search operations in a database. Each function is tailored for specific needs within the context of vector searches.
1. generate_simple_vector_search_function¶
Functionality
Generates a PostgreSQL function that performs a simple vector search. The generated SQL function retrieves vectors from a database using a specified distance metric. It is designed for efficient retrieval when only basic sorting and vector inclusion are needed.
Parameters
- model_id: A string representing the embedding model ID. It is used to generate table names and a unique function identifier.
- metric_type: Optional metric type from MetricType (COSINE, EUCLID, or DOT). Defaults to MetricType.COSINE.
Returns
A SQL string that creates a PostgreSQL function for vector search.
Usage
Use this function when a fast, simple vector search is required without advanced filtering or user-specific customizations.
Example
sql = generate_simple_vector_search_function("abc", MetricType.COSINE)
execute_sql(sql)
2. generate_simple_vector_search_no_vectors_function¶
Functionality
Generates a SQL function for simple vector search optimized for cases where vector values are excluded from the result set. This reduces data transfer and memory usage while preserving core search features.
Parameters
- model_id: ID of the embedding model, used to generate table names.
- metric_type: Optional. Metric for vector distance (COSINE, EUCLID, DOT). Defaults to MetricType.COSINE.
Usage
Create a PostgreSQL function to perform a vector search that omits the actual vectors in the output.
Example
sql = generate_simple_vector_search_no_vectors_function('example_model', MetricType.COSINE)
3. generate_simple_vector_search_similarity_ordered_function¶
Functionality
Generates SQL code for creating a PostgreSQL function that performs a simple vector search with vectors in the results. This function is ideal when no advanced filtering is required.
Parameters
- model_id: A string identifier for the embedding model, used to name tables and functions.
- metric_type: An optional metric type (COSINE, EUCLID, DOT) that determines the distance operator. Defaults to COSINE.
Returns
A SQL string that creates a PostgreSQL function for performing the search.
Raises
- ValueError: If an unsupported metric type is provided.
Usage
Provide a model ID and optionally a metric type to generate the SQL for a simple vector search function. The output SQL can be used to create a PostgreSQL function.
Example
sql_str = generate_simple_vector_search_similarity_ordered_function("example_model")
print(sql_str)
4. generate_simple_vector_search_similarity_ordered_no_vectors_function¶
Functionality
Creates a PostgreSQL function that performs a similarity-ordered search without including vector values. It returns similarity scores and object metadata, making it ideal for pure ranking and high-performance queries where vector data is not needed.
Parameters
- model_id: Embedding model identifier used for table name generation.
- metric_type: Vector distance metric type. Options are COSINE, EUCLID, and DOT.
Usage
Generate the SQL function for similarity-based search when only ranking and metadata are needed. This function creates a PostgreSQL function that can be executed for efficient similarity queries.
Example
sql = generate_simple_vector_search_similarity_ordered_no_vectors_function(
model_id="my_model", metric_type=MetricType.COSINE)
print(sql)