การจัดเก็บเวกเตอร์ใน PostgreSQL

Welcome back! In the previous lesson, we set up pgvector. Now, let's get practical and learn how to store actual vector embeddings in your PostgreSQL database.

This lesson covers creating tables with vector columns and inserting embedding data.

Before storing vectors, let's quickly recall how tables work in PostgreSQL. You use the CREATE TABLE statement to define your table's structure, including column names and their data types.

With pgvector installed, we gain a new, powerful data type!

The core of storing embeddings with pgvector is the vector data type. When you define a column as vector(N), N represents the dimension of your embeddings.

• vector(3): Stores 3-dimensional vectors.
• vector(1536): Stores 1536-dimensional vectors (common for OpenAI embeddings).

It's crucial that N matches the dimension of the embeddings you're working with.

To store embeddings, you'll create a table just like any other, but specify one or more columns as type vector(N).

Here's an example creating a simple table to hold items and their 3-dimensional embeddings:

The number N in vector(N) is vital. It must precisely match the number of values (dimensions) in your actual embedding vectors. If your embedding model produces 768-dimensional vectors, your column must be vector(768).

Mismatched dimensions will result in errors when inserting data. Always check your embedding model's output dimension!

Once your table is set up, you can insert vector data using the standard INSERT INTO SQL command. Vector values are provided as a string representation of a list of numbers, e.g., '[1.2, 3.4, 5.6]'.

You can insert multiple vectors into your table efficiently by providing multiple sets of values in a single INSERT INTO statement.

Each set of values corresponds to a new row in your table.

In real-world applications, you'll almost always store additional information (metadata) alongside your vectors. This could be text content, a user ID, a timestamp, or any other relevant data.

Just add more columns to your table definition:

When inserting data into a table with metadata, simply provide values for all the columns you wish to populate. Make sure the order of values matches the order of columns specified in your INSERT INTO statement.

Which of the following SQL statements correctly creates a table named products with an auto-incrementing primary key id and a vector column product_embedding of 1536 dimensions?

Great job! You've learned the fundamental steps to store vector embeddings in PostgreSQL using pgvector:

• Defining vector columns with vector(N), ensuring N matches your embedding dimensions.
• Creating tables with both vector and metadata columns.
• Inserting single and multiple vector data points, along with their associated metadata.

Next, we'll explore how to query this data to find similar vectors!