morphometrics

Morphometric modules of the EuroFab project

To run:

1. Clone this repository.

2. Run pixi install, then pixi run build and optionally pixi run tests . These commands setup the enviroment and all the required packages.

   • Alternatively you can manually install the conda-forge dependencies, but you have to still run the pixi build and tests commands: pixi add momepy umap-learn fast_hdbscan jupyterlab pyarrow matplotlib lonboard folium mapclassify datashader dask pip sidecar glasbey scikit-image colorcet pandas holoviews bokeh=3.1 esda pytest hdbscan

3. To run jupyter use either pixi run jupyter lab or pass extra arguments like pixi run jupyter lab --port 8888.

4. To run the analysis on the whole dataset - first, make sure you have the correct folder structure and the target label data in place. Then, run:

   • notebooks/1_download_data.ipynb to download all the Microsoft building footprints, split them into regions, and then download the overture streets.
   • notebooks/2_process_regions.ipynb to run the entire processing pipeline from building, street preprocessing, element generation and characters calculations.
   • notebooks/3_assign_targets.ipynb to assign target labels to each generated element.
   • notebooks/4_assign_buildings_to_h3.ipynb to assign buildings to h3 hexagons, in preparation for train/test splitting.
   • notebooks/5_train_test_split.ipynb to split the data into 5 training and test sets, based on the country of origin.
   • notebooks/6_single_model_training.ipynb to train and test a specific model. Alternatively, you can run the generate_predictions.py file.
   • notebooks/7_figures_and_tables.ipynb to generate paper/report figures and tables.
   • notebooks/interactive_exploration.ipynb to interactively explore the morphometric characters.