woher

A useless reverse geocoder powered by AI

The goal of this project is to generate city names based on a location (lat/lon). The model is trained on geonames.org data called cities500 which contain city names and locations for cities with a population > 500.

Gen 1 - LSTM

Pretty good at generating realistic city names, but coordinates not taken into account properly.

Development Environment

Install Anaconda then run:

conda create -n woher python=3 tensorflow keras pandas matplotlib jupyter

Activate the environemnt:

conda activate woher

Run Jupyter Notebook for Model Training

With the conda environment setup and activated run:

jupyter notebook

Gen 2 - Fine Tuned Llama 2

Notebooks need to run with a decent GPU and sizable RAM. E.g. Colab with extra RAM.