Deleting items from vectorstore

[jwnicholas99]

After creating a Chroma vectorstore from a list of documents, I realized that I needed to delete some of the chunks that are now in the vectorstore, but I can't seem to find any function to do so in chroma.py. Is there any way to do so? Or do I have to delete the entire collection then re-create the Chroma vectorstore?

[dholendar27]

from chromadb.config import Settings
client = chromadb.Client(Settings(
chroma_db_impl="duckdb+parquet",
persist_directory= "" # Replace with your desired directory
))
collection_name = client.list_collections()[0].name
collection = client.get_collection(name=collection_name)
file_name['source'] = " " # Replace with filename
collection.delete(
where=file_name
)

The file info is stored in the metadata in the chromadb, for instance:
"metadata": "{"source": "./data/image analytics for retail store.pdf", "page": 6}"}
So I am passing the data as an dictonary
file_name['source'] = "./data/image analytics for retail store.pdf"

I implemented using RetrievalQA. It works fine.

[speedyankur]

I found it missing for Weaviate vector DB langchain wrapper as well. Does it mean they want all embeddings to be cleaned off and regenerated again

[GihanMora]

There is an explanation in this video https://www.youtube.com/watch?v=hSuCT6Z2QLk