README.md

Help with querying the Land Matrix database via Python

This project aims to facilitate the retrieval of Land Matrix data through natural language queries.

This repository provides several resources:

• An end-to-end Streamlit application with optimal configuration. Explanations.
• A pipeline to reproduce our benchmark of models and methods. See below
• Educational notebooks that describe all the tasks needed for the entire pipeline. Explanations.

1. Installation

git clone https://github.com/tetis-nlp/landmatrix-graphql-python.git

• Installation of the Python environment

  conda create -n landmatrix python=3.9 pandas scikit-learn spacy streamlit
  conda activate landmatrix
  conda install -c conda-forge sentence-transformers
  pip install transformers faiss-cpu
  pip install ollama
  pip install langchain-openai
  pip install langchain-community
  pip install openpyxl

• Downloading the Spacy model

  python -m spacy download en_core_web_sm

• Installation and launch of Ollama

  curl -fsSL https://ollama.com/install.sh | sh
  ollama serve
  ollama pull llama3:8b

• Configure API keys (only compatible with chat ISDM): add your own ISDM API keys (without ")

  cp credentials.ini.default credentials.ini
  vim credentials.ini

2. Reproduce our benchmark

python src/experiments.py 

• Monitore your pipeline : tail -f logs/pipeline.log
• Stop the pipeline: Kill all the subprocess: pkill -f src/