Otto-Destiny/spe-africa-dseats-datathon-2026

Physics-informed machine learning for oil discovery, combining geological anomaly correction, domain features, and rigorous validation. # SPE Africa DSEATS Hackathon 2026 ## Project Layout - `oil_presence_trainset.csv`, `oil_presence_testset.csv`: original competition data. - `notebooks/`: EDA, feature engineering, and model experiments. - `notebooks/complete_row_test_like_validation.ipynb`: executed reconstruction of the 89.3-89.6% complete-row validation experiment and its complete-only/full-data comparison. - `notebooks/model-assisted_domain_audit.ipynb`: executed in-sample HistGB disagreement audit and pattern screens; no rows are deleted. - `scripts/`: reproducible experiment runners. - `docs/`: guidelines, domain references, observations, and working notes. - `domain_features_outputs/`: generated domain-feature datasets and summaries. - `domain_model_outputs/`: model metrics, predictions, plots, and deep-tabular logs. - `domain_analysis_outputs/`: row-level error and anomaly analysis. - `domain_analysis_outputs/model_assisted_domain_audit_in_sample_disagreements.csv`: 411-row developer review queue from the in-sample model audit. - `part2_corrected_data/`: corrected working train/test CSVs with anomaly indicators, diagnostic flags, counts, and remarks. - `docs/ANOMALY_CODEBOOK.md`: definitions and counting rules for anomaly and diagnostic columns. Run scripts from any working directory, for example: ```powershell python scripts/run_suspicious_factor_grid.py ```