To reduce space locally, we have utilized a consistent structure of the data folders. Below is a brief tutorial on how to replicate the data directories. First, the following directory structure must be created:
...
├── data
│ ├── README.md
│ ├── predictions
│ │ ├── GP_pred_probs_test.npz
│ │ ├── GP_pred_probs_train.npz
│ │ ├── GP_spatiotemporal_pred_probs_test.npz
│ │ ├── GP_spatiotemporal_pred_probs_train.npz
│ │ ├── GP_spatiotemporal_var_pred_probs_test.npz
│ │ ├── GP_spatiotemporal_var_pred_probs_train.npz
│ │ ├── GPvar_pred_probs_test.npz
│ │ ├── GPvar_pred_probs_train.npz
│ │ ├── GraphSAGE_pred_probs_test.npz
│ │ ├── GraphSAGE_pred_probs_train.npz
│ │ ├── baseline_pred_probs_test.npz
│ │ ├── baseline_pred_probs_train.npz
│ │ ├── diffusion_pred_probs_test.npz
│ │ ├── diffusion_pred_probs_train.npz
│ │ ├── stacking_pred_probs_test.npz
│ │ └── stacking_pred_probs_train.npz
│ ├── processed
│ │ ├── Xdata.csv
│ │ ├── Ydata.csv
│ │ ├── cols_metadata.json
│ │ ├── haversine_distances.npz
│ │ ├── location.gpkg
│ │ ├── partitions_builder.pk
│ │ ├── road_distances.npz
│ │ ├── test_pid.npz
│ │ └── train_pid.npz
│ └── raw
│ └── flint_sl_materials
│ ├── flint_sl_materials.cpg
│ ├── flint_sl_materials.dbf
│ ├── flint_sl_materials.prj
│ ├── flint_sl_materials.shp
│ └── flint_sl_materials.shx
...
All files can be replicated locally, though the distance matrices are > 5GB and thus were handled via Google Colab.
The raw data (stored as flint_sl_materials/) can be connected to the following data dictionary: