DeBEIR

A Dense Bi-Encoder for Information Retrieval library for experimenting and using neural models (with a particular emphasis on bi-encoder models) for end-to-end ranking of documents.

Requirements

• Python >= 3.10
• GPU hardware compatible with pytorch is encouraged
• Otherwise requirements for your index such as storage and CPU usage should be considered

Setup and installation

It is recommended to set up a virtual environment and install from source

```bash python3 -m venv venv source venv/bin/activate

pip install git+https://github.com/Ayuei/DeBEIR.git

Sentence Segmentation Model install

pip install https://s3-us-west-2.amazonaws.com/ai2-s2-scispacy/releases/v0.5.0/encoresci_md-0.5.0.tar.gz ```

Usage

The library has an emphasis on reproducibility and experimentation. With this in mind, settings are placed into configuration files to be used to build the pipeline.

```python3 from debeir.interfaces.pipeline import NIRPipeline

p = NIRPipeline.buildfromconfig(configfp="./tests/config.toml", engine="elasticsearch", nirconfigfp="./tests/nirconfig.toml")

The cosine offset ensures a non-negative score.

results = await p.runpipeline(cosineoffset=1.0) ```

See examples/ for more use cases and where to get started.

Documentation

API Documentation for the library with rendered HTML documentation is available at https://ayuei.github.io/DeBEIR/debeir.html which is built with the pdoc3 library and is rebuilt with every commit with gh-pages.

Statically compiled documentation (which is updated less frequently) can be found in the top level directory docs/index.html.

You can also build this documentation with the pdoc library by executing the following commands: ``` pip install -r requirements-docs.txt

pdoc -o docs/ src/debeir/ ```

Development

If you use to help with development of the library, first verify the tests cases and set up a development environment. This will take approximately 30 minutes to complete on a mid-range system.

Requires: Docker and pip installation of requirements-dev.txt packages.

```bash virtualenv venv

source virtualenv/venv/activate

pip install -r requirements-dev.txt

cd tests/

./buildtestenv.sh

pytest . ```

A helper script for removing the development environment is provided in tests/cleanup.sh

Community Guidelines

An Issue?

If you have any issue with the current library, please file an issue create an issue.

Contributing

For those wanting to contribute to the library, please see CONTRIBUTING.md and submit a pull request!

Support

If you wish to reach out to the author and maintainer of this library, please email vincent.nguyen@csiro.au