Search with Machine Learning is a four week class taught by Grant Ingersoll and Daniel Tunkelang and is focused on helping students quickly get up to speed on search best practices by first teaching the basics of search and then extending those basics with machine learning.
Students will learn indexing, querying, aggregations and text analysis, as well as how to use machine learning for ranking, content classification and query understanding.
The class is a hands-on project-driven course where students will work with real data and the Opensearch/Elasticsearch ecosystem along with libraries like FastText, XG Boost and OpenSearch Learning to Rank.
For our class, we have four weekly projects. Each project is a standalone Python application that interacts with an OpenSearch server (and perhaps other services).
You will find these four projects in the directories below them organized in the following way:
- Week 1:
- week1 -- The unfinished template for the week's project, annotated with instructions.
- Week 2:
- week2 -- The unfinished template for the week's project, annotated with instructions.
- Week 3 and 4: you get the picture
Our instructor annotated results for each project will be provided during the class. Please note, these represent our way of doing the assignment and may differ from your results, as there is often more than one way of doing things in search.
You will also find several supporting directories and files for Logstash, Docker and Gitpod.
- For this class, you will need a Kaggle account and a Kaggle API token.
- No prior search knowledge is required, but you should be able to code in Python or Java (all examples are in Python)
- You will need a Gitpod account.
NOTE: The Gitpod free tier comes with 50 hours of use per month. We expect our work will be done in less time than that. However, you may wish to conserve time on the platform by being sure to stop your workspace when you are done with it. Gitpod will time you out (don't worry, your work will be saved), but that may take longer to detect.
The following things must be done each time you create a new Gitpod Workspace (unfortunately, we can't automate this)
-
Fork this repository.
-
Launch a new Gitpod workspace based on this repository. This will automatically start OpenSearch and OpenSearch Dashboards.
- Note: it can take a few minutes for OpenSearch and the dashboards to launch.
-
You should now have a running Opensearch instance (port 9200) and a running Opensearch Dashboards instance (port 5601)
-
Login to the dashboards at
https://5601-<$GITPOD_URL>/
with default usernameadmin
and passwordadmin
. This should popup automatically as a new tab, unless you have blocked popups. Also note, that in the real world, you would change your password. Since these ports are blocked if you aren't logged into Gitpod, it's OK.$GITPOD_URL is a placeholder for your ephemeral Gitpod host name, e.g. silver-grasshopper-8czadqyn.ws-us25.gitpod.io
-
Run the install Kaggle API token script and follow the instructions:
./install-kaggle-token.sh
-
Accept all of the kaggle competition rules then run the download data script:
./download-data.sh
- Login to OpenSearch and point your browser at
https://5601-<$GITPOD_URL>/app/opensearch_dashboards_overview#/
- Click the "Add sample data" link
- Click the "Add Data" link for any of the 3 projects listed. In the class, we chose the "Sample flight data", but any of the three are fine for exploration.
At the command line, do the following steps to run the example.
- Activate your Python Virtual Environment. We use
pyenv
(Pyenv website)[https://github.com/pyenv/pyenv] andpyenv-virtualenv
(Pyenv Virtualenv)[https://github.com/pyenv/pyenv-virtualenv].pyenv activate search_with_ml
-- Activate the Virtualenv.
- Optional: You can run
ipython
if you like.
To run locally, you will need a few things:
- Pyenv and Pyenv-Virtualenv with Python 3.9.7 installed
- Docker
- A Git client
Note: these have only been tested on a Mac running OS 12.2.1. YMMV. Much of what you will need to do will be similar to what's in .gitpod.Dockerfile
- Install GraphViz
pyenv install 3.9.7
pip install
all of the libraries you see in.gitpod.Dockerfile
- Setup your weekly python environments per the "Weekly Project" above.
- Install Fasttext
- Run OpenSearch:
cd docker
docker-compose up
- Note: most of the scripts and projects assume the data is in
/workspace/datasets
, but have overrides to specify your own directories. You will need to download and plan accordingly. - Do your work per the Weekly Project