This repo contains the code to work with the Hierarchical Point-Of-Interest Kernels.

Requirements

• Install a Python Virtual Environment using python>=3.3. Then install the required packages from the provided requirements.txt

pip install -r requirements.txt

• To run the experiments you will need HPK. We implemented HPK as an additional kernel in Kelp, a java machine learning framework focusing on kernel machines for structured data. HPK can be installed with the provided kelp-classify.jar file.

You will also need `GeoL, The Geospatial Library developed for running generic data cleaning and pre-processing tasks when using geographic data with Machine Learning.

Project Structure

The project is organised as follows:

• notebooks = This folder contains the notebooks used to prepare and run the experiments. There are three notebooks: two for pre-processing respectively the Urban Atlas and Foursquare datasets, and a third one to run the experiments which should be run last.

• scripts = Simple Python executables to perform generic operations (e.g. prepare the folder structure etc...)

• data = Test and Train datasets available for all the analysed cities.

A note on the datasets:

Due to legal constrains, we could not provide the data as they are. Both Urban Atlas and Foursquare require in fact a registration and forbid a straightforward re-use.

• Urban Atlas can be freely downloaded through its portal.
• Foursquare data can be retrieved using GeoL's crawler. It requires an appropriate API key.