STELLAR porphyry copper data mining scripts

This repository contains the Python scripts and notebooks required to extract the data, train the models, and produce the figures from "Spatio-temporal copper prospectivity in the American Cordillera predicted by positive-unlabelled machine learning".

Training data can be extracted from the plate model and other input datasets using the 00a-extract_training_data.ipynb and 00b-extract_grid_data.ipynb notebooks. The first of these notebooks extracts data for the positive/negative mineral deposit observations in data/deposit_data.csv, to be used for training and testing. The second notebook extracts data for a regular grid of points, to be used to create the time-dependent mineral prospectivity maps.

Alternatively, the above process can be skipped by using pre-prepared data downloaded from the Zenodo repository (zenodo.org/record/8157691). Running the notebooks in sequence, beginning with 01-create_pu_classifier.ipynb, will automatically download this data to a directory named prepared_data.

To run the notebooks:

1. Create a conda environment using the environment.yml file: conda env create --file environment.yml
2. Run the following notebooks to download and extract training data from Zenodo (optiona):
   • 00a-extract_training_data.ipynb
   • 00b-extract_grid_data.ipynb
3. Run these notebooks to train a PU classifier and create prospectivity maps:
   • 01-create_pu_classifier.ipynb
   • 02-create_probability_maps.ipynb
   • 03-create_probability_animation.ipynb

To create the figures:

To create the figures used in the article, run the following notebooks:

• Fig-01-02-probability_snapshots.ipynb
• Fig-03-04-feature_importance.ipynb
• Fig-05-partial_dependence.ipynb
• Fig-06-performance.ipynb