Explore setting up Elasticsearch document store

[ugm2]

Since Elasticsearch needs a special setup for it to work (you need to spin up an Elasticsearch instance), this issue aims at exploring a way of doing that for Huggingface Spaces demo.

Using this Document Store will enable the use of BM25 keyword search algorithm, which outperforms TFIDF (currently the one we are using).

This HF space demo seems to have been able to set it up, it could be used as an example. It seems it's using Environment Variables that are internal in Huggingface. The caveat of this is that when running this demo locally, the user needs to have an Elasticsearch instance running and the env variables must be set.

Definition of done:

• Set up elasticsearch on HF Spaces
• Use it as Document Store
• Use BM25 Retriever
• Document in readme how to run the demo locally using Elasticsearch
• Think about a workaround if Elasticsearch is not running by defaulting to TFIDF and InMemoryStore with a warning message

[anakin87]

Hey @ugm2,
some months ago I contributed to Haystack with:

• feat: add index parameter to TfidfRetriever deepset-ai/haystack#3666
• feat: add support for BM25Retriever in InMemoryDocumentStore deepset-ai/haystack#3561

They could be useful for your project...