MAGRE - A Multimodal and Multitask Approach for Adaptive Geospatial Region Embeddings

MAGRE builds adaptive spatial region embeddings that can be used in multiple downstream tasks.

Prerequisites

• Python >= 3.8
• pytorch

Setup

1. Clone the repository

git clone https://github.com/Rajjat/MAGRE
cd MAGRE

2. Install the Python Requirements

pip install -r requirements.txt

Usage

python model.py # to train the model   
python model -tr embed # to generate the embeddings        
python regression_tasks.py # to run crime prediction and check-in prediction    
python3 landusage_classfication.py # to run the land use classification task

Environment

The experiments were conducted using Ubuntu 22.04.2 LTS, utilizing a system equipped with 1TB RAM and 96 cores