From the course: Vector Databases in Practice: Deep Dive

A high-level view of vector databases

From the course: Vector Databases in Practice: Deep Dive

A high-level view of vector databases

“

- [Lecturer] So what is a vector database and how are they different from other types of databases? At a very high level, a vector database is a type of database that is capable of organizing data by their meaning. This allows a vector database to perform searches, to find database entries that are most similar to the search query. For instance, let's say you search a vector database containing words for the word cat. Using a vector search, the results would not only include the word cat and derivative words like cats, but also words with similar meaning like kitten, lion, jaguar, and so on. And if you searched a vector database containing paragraphs for the phrase history of computing, the top results would include those most relevant to this idea or concept of history of computing. You'll notice that these results don't necessarily include the same words that were used for search, and yet our vector database was able to identify those results as relevant. This is what vectors allow. Vectors capture or represent meaning as a series of numbers, and that goes for a word like cat, sentences, paragraphs, or even other media. These representations work a little bit like how different systems like RGB or CMYK systems represent colors. An RGB representation of a color is a set of three numbers, like 255, 0, 0 for red, or 65, 105, 225 for royal blue. By changing each of these three numbers like a dial, we can represent any color that we'd like to. And if we plot colors in three dimensions according to these numbers, similar colors magically end up next to each other. Vector databases extend this idea, but instead of three dials, we have hundreds of dials to represent more nuanced meaning. And these vectors are used to find object with similar meaning, just like how similar colors can be found by similar RGB values. This ability to deal with data by their meaning provides vector databases with tremendous power and flexibility, as you'll see throughout the course. And of course, vector databases can make use of traditional searching tools as well. Vector databases can, for example, perform keyword filtering and keyword searches for use cases where the exact matches matter. These can also be combined with vector search capabilities as you'll see later on. So at a high level, vector databases catalog and retrieve information based on the concept of vectors. And vectors represent an object's meaning as a series of numbers, which is the idea at the heart of this technology. - Now, these capabilities are very exciting, but might also sound a little bit abstract at the moment. But don't worry, in the upcoming videos, you'll see all of these capabilities and more in action. So let's get into it.

Contents