What are vectors and vector databases? Let me explain it to you by choosing a computer and writing code!
Use computer selection and programming examples to easily introduce you to the world of vectors and vector databases!
Core content:
1. Basic concepts and characteristics of vectors
2. Application scenarios of vectors in the AI era
3. Methods and technologies for implementing vectorization
Introduction to Vectors
More powerful than Excel
“Data fingerprint”
Imagine you walk into a computer mall and see that every laptop has a parameter card like this:
CPU: i7-13700H Graphics card: RTX 4060 Weight: 1.87kg Price: 8999 yuan
If we concatenate these values, we get a 4-dimensional vector:
[13700, 4060, 1.87, 8999]
This is a vector : using a string of numbers to accurately describe the characteristics of an object, just like generating a unique ID card for each object .
A vector is a quantity that has both magnitude and direction , and is often represented in computers as a digital array [v1, v2, ..., vn].
As the "memory center" of the AI era, vector databases are reshaping the boundaries of artificial intelligence applications by efficiently processing unstructured data.
Three characteristics of vectors
1. Dimensional freedom
The computer parameter vector may be 10-dimensional: [CPU, memory, hard disk, graphics card, screen size, weight, price...]
The programmer code vector may be 100-dimensional: [function length, number of loops, API calls, error types...]
2. Computability
Compare similarities through mathematical calculations:
# Calculate the similarity of two laptops (the closer the value is to 1, the more similar they are) cos_sim( [i7, 4060, 1.8kg, 8999], # Laptop A [i9, 4080, 2.1kg, 12999] # Laptop B ) = 0.76
3. Semantic magic
AI can turn text into vectors, allowing computers to understand semantics:
"laptop" → [0.23, 0.76, -0.12,...] "portable computer" → [0.25, 0.74, -0.09,...]
The similarity between these two vectors is as high as 0.98 ! Wow, no more manual comparison and selection of computers.
Industry Applications of Vectors
Scenario 1: Accurate recommendation for computer sales
When a customer says, "I want a thin and light office notebook," the system:
Convert the requirements to vectors → [ CPU weight: 0.3, graphics card weight: 0.1, weight weight: 0.6... ]
Search the stock vector library for the closest → match to HUAWEI MateBook X Pro 2024
Generate recommendation words: "This model weighs only 1.26kg and has a battery life of 18 hours, making it particularly suitable for mobile office work."
Scenario 2: Programmer code reuse
When you need to implement the "form validation" function:
Convert requirements into code feature vectors → [ validation function: 1, regular expression: 0.8, error handling: 0.7... ]
In the code vector library matching → find validateForm.ts in the 2023 project
Automatic prompt: "Refer to the TypeScript validation tool class on lines 45-78"
How to generate vectors
1. Direct conversion of numeric values
Computer price: 8999 → Directly as one-dimensional vector
Lines of code: 128 → Directly as one-dimensional vector
2. Text vectorization
Using AI models (such as BERT):
from sentence_transformers import SentenceTransformer model = SentenceTransformer('paraphrase-multilingual-MiniLM-L12-v2') # Convert text to 768-dimensional vector text = "RTX 4080 supports DLSS 3.0 technology" vector = model.encode(text) # Output is like [0.23, -0.45, 0.17,...]3. Image/speech vectorization
Image: Extracting feature vectors using the ResNet model
Speech: Convert to text using the Whisper model and then vectorize
Domain positioning of vector database
1. Technical pedigree
Database technology : belongs to the branch of unstructured database , on par with relational database
AI infrastructure : As a new storage engine in the era of big models, it supports semantic understanding capabilities
2. Technology comparison
3. Technology stack position
Database technology stack
Why do vectors improve efficiency?
1. Traditional method
Computer sales : Manually flipping through parameter tables for comparison
Programmers : Global search for code keywords
2. Vectorized method
Semantic understanding : Knowing that "thin and light notebook" ≈ "portable notebook" ≈ "ultrabook"
Fuzzy matching : even if the parameters are not exactly the same, you can find the closest option
Cross-modal search : search for images using text, search for code using error logs
Knowledge Map
Theoretical basis: Linear algebra → Vector space theory → Machine learning embedding technology
Engineering practice: Database principles → Approximate nearest neighbor algorithm → Vector database architecture
Application extension: Recommendation system/semantic search → RAG architecture → Enterprise knowledge management
The development of vector technology
? Hands-on experiment
Quickly experience vector magic with Python:
import numpy as np # Define vectors of two notebooks notebook_A = np.array([8, 16, 512, 1.87]) # [CPU cores, memory GB, hard disk GB, weight kg] notebook_B = np.array([12, 32, 1024, 2.15]) # Calculate Euclidean distance (the smaller the value, the more similar) distance = np.linalg.norm(notebook_A - notebook_B) print(f"The difference index of these two notebooks: {distance:.2f}")
