Chroma Vector Store Example with LangChain and Ollama

This example demonstrates how to use the Chroma vector store with LangChain and Ollama to perform similarity searches on a collection of city data. The program showcases various querying techniques, including basic similarity search, filtering, and score thresholding.

What This Example Does

1. Setup:

   • Initializes an Ollama language model (LLM) with the "llama2" model.
   • Creates an embedder using the Ollama LLM.
   • Sets up a Chroma vector store with custom configurations.

2. Data Loading:

   • Adds a collection of city documents to the vector store. Each document contains the city name, population, and area.

3. Similarity Searches: The example performs three different similarity searches:

   a. "Up to 5 Cities in Japan":

   • Searches for Japanese cities with a score threshold of 0.8.
   • Limits the results to a maximum of 5 cities.

   b. "A City in South America":

   • Looks for a South American city with a score threshold of 0.8.
   • Returns only one result.

   c. "Large Cities in South America":

   • Searches for South American cities with specific filters:
     • Area greater than or equal to 1000
     • Population greater than or equal to 13 million

4. Results Display:

   • Prints the results of each search query, showing the matching city names.

Key Features

• Demonstrates the use of Chroma vector store for similarity searches.
• Shows how to use Ollama for embeddings and as an LLM.
• Illustrates different querying techniques:
  • Basic similarity search
  • Score thresholding
  • Filtering based on metadata

This example is perfect for developers looking to understand how to implement and use vector stores for semantic search applications, especially when working with geographical data.