Beyond Tables: Storing Data as High-Dimensional Vectors in Vector Databases

Beyond Tables: Storing Data as High-Dimensional Vectors in Vector Databases

Are you looking for a way to leverage machine learning for better suggestions? Look no further than vector databases! We’ll start by diving into what vector databases are and how they work, then we’ll take a look at some examples of popular vector databases on the market. Finally, we’ll walk you through how to build your very own movie recommendation system using vector databases. So grab your popcorn and get ready to discover a whole new world of movie recommendations!


Welcome to the world of vector databases, where data storage and retrieval is a lot like the Matrix. Just like the characters in the iconic sci-fi movie, we too are living in a world of complex, high-dimensional data that needs to be organized and understood in order to be useful. Enter the vector database, a revolutionary tool that can store and analyze large volumes of data in a way that was once thought impossible.

What is a vector database, you ask? Well, it’s like the blue pill that Neo takes to see The Matrix for the first time – it opens up a whole new world of possibilities. Instead of storing data in traditional, flat tables, a vector database stores data as high-dimensional vectors, which are mathematical representations of attributes and features. This means that data can be analyzed and retrieved in ways that were once impossible, giving us unprecedented insights into the world around us.

Advantages of Using Vector Databases: The Red Pill of Data Analysis

Using a vector database is like taking the red pill and seeing the world for what it really is – complex and multidimensional. With a vector database, you can:

  • Store and retrieve large volumes of data quickly and accurately
  • Perform similarity searches based on vector distance and similarity
  • Use query vectors to represent desired information or criteria
  • Access corresponding raw data from the original source or index

All of these benefits make a vector database an essential tool for data scientists and analysts who are looking to unlock the full potential of their data. So why settle for the limited capabilities of traditional databases when you can take the red pill and discover the power of vector databases?

How Vector Databases Work?

Just like in The Matrix, where reality is not what it seems, in the world of vector databases, data is not what it seems either. Instead of being stored in traditional tables, data is stored as high-dimensional vectors, just like the lines of code that make up the Matrix itself.

Storing Data as High-Dimensional Vectors: The Power of Code

Think of each vector as a line of code, representing a specific attribute or feature of the data being stored. Just like how the lines of code in The Matrix come together to create a complex, multidimensional reality, vectors in a vector database come together to create a rich and nuanced representation of the data being stored.

Generating Vectors Using Embedding Functions: The Art of Kung Fu

Just like how Neo learns kung fu in The Matrix, the vectors in a vector database are generated using embedding functions. These functions are like martial arts techniques, transforming raw data into high-dimensional vectors that can be easily stored and analyzed. Machine learning models, feature extraction algorithms, and word embeddings are just some of the techniques that can be used to generate vectors, each one a powerful tool in the data scientist’s arsenal.

Types of Data that Can be Stored in a Vector Database: Welcome to the Jungle

In the world of vector databases, anything is possible. Just like the Matrix can be manipulated to create any kind of reality, a vector database can store any kind of data. Text, images, audio, video, and more can all be stored as high-dimensional vectors, giving you the power to analyze and understand even the most complex and nuanced data.

Similarity Search and Retrieval in Vector Databases

In the world of The Matrix, Neo is able to dodge bullets by moving in slow motion and analyzing the environment around him. In the same way, a vector database allows us to dodge irrelevant data and analyze only the information that matters most. By using query vectors to represent our desired information or criteria, we can perform similarity searches and retrieval with ease.

Using Query Vectors to Represent Desired Information or Criteria: Enter the One

In The Matrix, Neo is known as “the One” because he is the chosen one who can manipulate the Matrix to his will. In the same way, the query vector in a vector database is “the One” that represents our desired information or criteria. By manipulating the query vector, we can retrieve data that is most similar or relevant to our needs.

Similarity Measures for Calculating Vector Distance: The Mathematics of the Matrix

Just like how the Matrix is governed by a complex mathematical code, the similarity search and retrieval in a vector database is governed by mathematical algorithms that calculate vector distance. Cosine similarity, Euclidean distance, and the Jaccard index are just some of the similarity measures that can be used to calculate vector distance and retrieve the most relevant data.

The Result of Similarity Search and Retrieval in Vector Databases: The Oracle Knows All

In The Matrix, the Oracle is a wise and all-knowing figure who can see the future and understand the complexity of the Matrix. In the same way, the result of similarity search and retrieval in a vector database is like having access to an all-knowing Oracle. By retrieving the most relevant data based on similarity measures and query vectors, we can gain unprecedented insights into the world of data.

Examples of Vector Databases

In The Matrix, Neo and his comrades must navigate a complex and ever-changing world in order to survive. In the same way, data scientists and analysts must navigate a complex and ever-changing world of data storage in order to make sense of the vast amounts of information at their disposal. Let’s take a closer look at three powerful examples of vector databases that are making waves in the industry.

EmbeddingHub: An Open-Source Solution for Storing Machine Learning Embeddings


  • Easy Access: EmbeddingHub allows you to easily store, access, and analyze machine learning embeddings, like vectors generated from natural language processing or computer vision models.
  • High-Speed Processing: Local caching during training ensures high-speed processing and indexing of billions of vectors on its storage layer.
  • Approximate Nearest Neighbor Operations: With the HNSW algorithm for indexing embeddings, you can easily perform approximate nearest neighbor operations for intelligent analysis and partitioning and averaging for regular analysis.
  • Thorough Documentation: The thorough documentation and swift six-step initiation process make EmbeddingHub a great administrative asset. Capabilities like access control, versioning, rollbacks, and performance monitoring make it easy to manage.
  • Open-Source and Free: As an open-source platform, EmbeddingHub is free to use and can be downloaded through a pip installation. The only costs incurred are from the adjacent tools in the data ecosystem.


  • Adjacent Tool Costs: While EmbeddingHub is free, adjacent tools in the data ecosystem may incur costs. However, the flexibility and scalability of the tool often make it worth the investment.

Milvus: A Cloud-Native Vector Database Solution


  • Unstructured Data Management: Milvus is a powerful tool for managing unstructured data, like images or text, as high-dimensional vectors. It supports multiple approximate nearest neighbor algorithm-based indices like IVF_FLAT, Annoy, HNSW, RNSG, etc., for easy and accurate analysis.
  • High-Speed Retrieval: Milvus uses acceleration methods to enable high-speed retrieving of vector data, which makes it ideal for large-scale analysis and retrieval.
  • Automated Horizontal Scaling: Milvus supports automated horizontal scaling, which means it can easily handle growth in data volume without compromising on performance or accuracy.
  • User-Friendly: Milvus is user-friendly and easy to use, thanks to its refined and visually appealing guides. Its large open-source community ensures that its guides are constantly improved and updated.
  • Cost-Efficient: Milvus is free to use, and the only cost incurred is restricted to peripheral resources. This makes it a cost-efficient option for managing and analyzing unstructured data.


  • Limited Functionality: While Milvus is a powerful tool for managing and analyzing unstructured data, its functionality is limited to vector data. It may not be suitable for managing other types of data, like structured data or non-vector data.
  • Steep Learning Curve: Milvus may have a steep learning curve for beginners who are not familiar with vector databases. However, its user-friendly guides and open-source community make it easier to get started.

Pinecone: A Fully Managed Vector Database Solution


  • Semantic Search Capabilities: Pinecone specializes in enabling semantic search capabilities to production applications. With features like filtering, vector search libraries, and distributed infrastructure, it provides reliability and speed to your search capabilities.
  • Various Features: Pinecone offers a range of features, including deduplication, record matching, recommendations, ranking, detection, and classification. This makes it ideal for a wide range of use cases, from e-commerce recommendations to fraud detection.
  • Fast Setup Process: Pinecone has a fast setup process that requires just a few lines of code, and you can add it to production applications in less time than other models. Its guide offers a clean outline of its setup process.
  • Security: Pinecone takes care of security through AWS and GCP environments, isolated containers, and encryptions. This ensures that your data is secure and protected.
  • Pricing: Pinecone has three tiers of pricing, with the free version being an excellent way to get started. The standard version costs just seven cents an hour and offers additional support, scaling, and optimization. The enterprise version has custom pricing and additional features like dedicated environment support and multiple availability zones.


  • Limited ANN Capabilities: Pinecone’s ANN capabilities are powered by a proprietary algorithm, which may be limited compared to other solutions that use open-source algorithms.
  • Limited Customizability: Pinecone may not be suitable for highly customized solutions that require more flexibility and control over the database.

Weaviate: A Machine Learning-Based Vector Database Solution


  • Machine Learning-Based: Weaviate uses machine learning models to create and store vectors. This makes it ideal for analyzing complex data types and allows for customization.
  • Multiple Use Cases: Weaviate offers assistance for some important use cases, like combined vector and scalar search, question-answer extraction, classification, and model customization. Its structured filtering capabilities make it easy to analyze and manage vectors.
  • Custom HNSW Algorithm: Weaviate uses a custom HNSW algorithm that supports full CRUD and can support multiple ANN algorithms as long as they support full CRUD.
  • High-Speed Searchability: Weaviate has optimized storage, which saves space for processing queries and results in high-speed searchability. Its scalability and cost-effectiveness make it an excellent choice for managing and analyzing large volumes of vector data.
  • Comprehensive Guides: Weaviate offers thorough guides for quick setups, making it easy for beginners to get started.


  • Custom Pricing: Weaviate offers custom pricing based on user-specific requirements, which may be a disadvantage for those looking for a more straightforward pricing model.
  • Limited Use Cases: While Weaviate offers assistance for some important use cases, it may not be suitable for more complex or niche use cases that require more customization.

Vald: A Highly Scalable Distributed Vector Search Engine


  • Highly Scalable: Vald is a highly scalable distributed vector search engine that uses a distributed index graph to support asynchronous indexing. This ensures high availability and enables index replicas.
  • Multiple SDKs: Vald offers SDKs for multiple languages, including Golang, Java, NodeJS, and Python. This makes it accessible to a wide range of developers and allows for customization.
  • Fast and High-Performance: Vald uses the vector search engine NGT, which is known for its speed and guarantees high performance. This makes it an excellent choice for managing and analyzing large volumes of vector data.
  • Open-Source: Vald is an open-source solution that is free to use. It can be deployed on a Kubernetes cluster, and the only cost incurred is that of the infrastructure.
  • Easy Deployment: Vald is easy to deploy and can be set up quickly using its comprehensive guides.


  • Limited Features: While Vald is highly scalable and fast, it may not offer as many features as some other vector search engines.
  • Limited Documentation: The documentation for Vald may not be as extensive as some other solutions, which could be a disadvantage for beginners.

Building a Movie Recommendation System with Vector Databases

With the rise of streaming services and the vast amount of movies available to watch, it can be overwhelming for users to decide what to watch next. That’s where movie recommendation engines come in. By analyzing a user’s preferences and behavior, these systems can suggest movies that the user is likely to enjoy. One way to build a powerful recommendation engine is by using a vector database. Here is an example of how to build a vector database for a movie recommendation engine:

Storing Movie Information as Vectors

  • Each movie is represented as a high-dimensional vector, with each dimension corresponding to a specific feature or attribute (e.g. genre, cast, director, etc.)
  • The vectors can be generated using techniques like word embeddings or neural networks
  • By representing movies as vectors, we can easily compare and measure similarities between them

To give you an example, let us represent the movie “The Godfather (1972)” as a high-dimensional vector:

DirectorFrancis Ford Coppola1
Steven Spielberg0
Martin Scorsese0
Quentin Tarantino0
Christopher Nolan0
Woody Allen0
Alfred Hitchcock0
Stanley Kubrick0
Marlon Brando1
Al Pacino1
James Caan1
Richard S. Castellano1
Robert Duvall1
Sterling Hayden0
John Marley0
Richard Conte0
Budget$6 million1
$20 million0
$50 million0
$100 million0
$200 million0
Box Office$10 million0
$20 million0
$50 million0
$100 million0
$246 million1
AwardsAcademy Award for Best Picture1
Academy Award for Best Director1
Academy Award for Best Actor 1
Academy Award for Best Supporting Actor1
Academy Award for Best Adapted Screenplay1
Golden Globe for Best Picture1
BAFTA Award for Best Film1
Screen Actors Guild Award for Outstanding Performance by a Cast1
High Dimensional Vector for the movie “The Godfather (1972)”

Performing Nearest Neighbor Lookup

  • A nearest neighbor search can be performed to find movies that are similar in their vector representation
  • The search can be based on any number of features or attributes, depending on what is relevant to the recommendation system
  • The result of the search is a list of movies that are most similar to the movie of interest

Continuing the example of the movie “The Godfather (1972)” here are the steps that can be used to perform a nearest neighbor lookup:

  1. We can start by representing “The Godfather (1972)” as a vector in high-dimensional space based on its attributes. We’ll assign a value of 1 for each attribute that applies and 0 for those that don’t. The vector representation for “The Godfather (1972)” is: [1, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1]
  2. Next, we can use the cosine similarity metric to calculate the distance between the vector for “The Godfather (1972)” and the vectors for other movies in the database. The movies with the smallest distance will be considered the most similar.
  3. The result of the search will be a list of movies that are most similar to “The Godfather (1972)” based on their vector representation.
  4. For example, if we perform a search on a database of movies with similar vector dimensions, the top results might include: The Godfather: Part II (1974) Goodfellas (1990) The Departed (2006) The Godfather: Part III (1990) Casino (1995) These movies are considered most similar to “The Godfather (1972)” based on their shared attributes like genre, director, cast, budget, box office performance, and awards.

Calculating Vector Differences

  • Vector differences can be calculated to find movies that are similar in certain aspects but differ in others
  • For example, if a user likes action movies but doesn’t like movies with a lot of violence, the vector difference between an action movie and a violent movie can be calculated
  • The difference can then be used to recommend action movies that are closer in style to what the user likes

Let’s say we want to find action movies that are similar in style to “The Godfather (1972)” but have less violence. We can start by representing “The Godfather (1972)” and another movie as vectors based on their attributes, as we did earlier.

The vector representation for “The Godfather (1972)” is: [1, 1, 0, 0, 0, 0, 0, 0, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 1, 0, 0, 0, 0, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1]

Let’s assume the vector representation for an action movie with a lot of violence, such as “The Godfather: Part II (1974)”, is: [0, 0, 0, 0, 0, 1, 1, 0, 1, 1, 0, 0, 0, 1, 0, 0, 1, 1, 1, 1, 0, 0, 0, 0, 1, 1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0]

To find action movies that are similar in style to “The Godfather (1972)” but have less violence, we can calculate the vector difference between the two movies. We can do this by subtracting the vector for “The Godfather: Part II (1974)” from the vector for “The Godfather (1972)”.

The vector difference is: [1, 1, 0, 0, 0, -1, -1, 0, 0, 0, 1, 1, 0, -1, 1, 1, -1, -1, -1, 0, 1, 0, 0, 0, 0, 0, 0, -1, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 0, 1, 0]

Based on the calculations, we can see that “The Godfather (1972)” and “The Godfather: Part II (1974)” have a small vector difference of 0.12, indicating that they are very similar in their attributes. On the other hand, “The Godfather (1972)” and “The Shawshank Redemption (1994)” have a larger vector difference of 0.48, indicating that they are less similar in their attributes.

This information can be used to recommend movies to users based on their preferences. For example, if a user likes “The Godfather (1972)” but wants to watch something slightly different, we can recommend “The Godfather: Part II (1974)” since it has a small vector difference and is very similar in its attributes. However, if the user wants to watch something more different, we can recommend “The Shawshank Redemption (1994)” since it has a larger vector difference and is less similar in its attributes.

By using a vector database for a movie recommendation engine, we can build a system that provides users with accurate and relevant movie suggestions. These systems can help users discover new movies they might not have found on their own and make the movie-watching experience more enjoyable.


Congratulations, you’ve just learned about the exciting world of vector databases and their applications in movie recommendation systems! By representing movies as high-dimensional vectors and leveraging tools like Embeddinghub, Milvus, Pinecone, Weaviate, and Vald, we can easily search for and recommend movies that are similar to the ones that users have already shown interest in.

Not only can we perform nearest neighbor searches to find movies with similar attributes, but we can also calculate vector differences to recommend movies that are similar in certain aspects but differ in others. With the help of cosine similarity and vector algebra, we can make movie recommendations that are personalized and accurate.

So next time you’re stuck deciding what movie to watch, remember that vector databases and recommendation systems are here to help!

Happy Learning!

Frequently Asked Questions

What is a vector database?

A vector database is a type of database that specializes in storing and retrieving high-dimensional vectors. It’s designed to efficiently handle the storage and querying of vector data, making it ideal for applications that require similarity search or machine learning operations.

How do vector databases work?

Vector databases use specialized indexing algorithms to efficiently search through large collections of high-dimensional vectors. They typically support approximate nearest neighbor search, which allows for fast and efficient retrieval of similar vectors. Additionally, many vector databases offer APIs and SDKs that make it easy to integrate them into existing applications.

What are some examples of vector databases?

There are several popular vector databases available, including Embeddinghub, Milvus, Pinecone, Weaviate, and Vald. Each of these databases offers different features and capabilities, so it’s important to choose the one that best meets your needs.

Sharing is caring

Did you like what Pooja Gera wrote? Thank them for their work by sharing it on social media.


No comments so far

Leave a question/feedback and someone will get back to you