A package for visualising vector embedding collections as part of the Chroma vector database.
Uses Flask, Vite, and react-three-fiber to host a live 3D view of the data in a web browser, should perform well up to 10k+ documents. Dimensional reduction is performed using PCA for colors down to 50 dimensions, followed by tSNE down to 3.
pip install chromaviz or pip install git+https://github.com/mtybadger/chromaviz/.
After installing from pip, simply call visualize_collection with a valid ChromaDB collection, and chromaviz will do the rest.
from chromaviz import visualize_collection
visualize_collection(chromadb.Collection)
It also works with Langchain+Chroma, as in:
from langchain.vectorstores import Chroma
vectordb = Chroma.from_documents(data, embeddings, ids)
from chromaviz import visualize_collection
visualize_collection(vectordb._collection)
- More dimensional reduction options and flexibility
- Refactor extremely shoddy React code
- Improve UX
