We've noticed that our users of the Chorus Electonics website are truly terrible at spelling, for example this query: Toner catrdge from leximark brand is returning laptop cases:
Notice that the first toner cartridge is in the third row, and it's an Epson brand!
This is a great use case for vectors, aka neural search. They can really improve the matching, without the complexity of more traditional spellcheck type approaches, as you can see below:
To run this Kata, we're trying something new, we're going to use a Jupyter Notebook as there are a lot of commands to run to set up hybrid search in OpenSearch.
WARNING This Kata requires Opensearch 2.18 with Multi-Search, which means we disabled UBI to make it work with a nightly build.
To get started you need to have a recent Python version.
-
Open a terminal and change to the katas directory:
cd ./katas -
We're going to use a "Virtual Environment" to organize everything:
python3 -m venv .venv -
Now start up the env:
source .venv/bin/activate -
Lastly, install all the required libraries:
pip install -r requirements.txt
Or:
cd ./katas
python3 -m venv .venv
source .venv/bin/activate
pip install -r requirements.txt
jupyter notebook 005_2_run_a_hybrid_search.ipynb
There are two ways to play with this.
First, you can use notebook visualization tool Mercury to play with keyword and neural weightings and try out different queries.
Run mercury run and browse to http://127.0.0.1:8000/ to see the Hybrid notebook. It will take 2 minutes to load as
reindexing the data with vector embeddings takes a while!
If you want to see all the commands for setting Hybrid search, then use the Jupyter notebook directory.
Run: jupyter notebook 005_2_run_a_hybrid_search.ipynb