AI-Aquatica

Real-dataset workflow

AI-Aquatica includes a single executable workflow for the real dataset used in the SoftwareX illustrative example:

python examples/real_dataset_workflow.py --output outputs/real_dataset

The workflow:

1. detects CP1250 encoding, semicolon separation and comma decimals;
2. loads 148 observations from four stations;
3. derives bicarbonate from alkalinity and calculates charge-balance diagnostics;
4. marks incomplete or invalid ion sets as indeterminate;
5. performs campaign-grouped station classification with month as the group;
6. performs a second classification after filtering to hydrochemically acceptable rows;
7. performs campaign-grouped chlorophyll-a regression;
8. compares regression performance with a mean-prediction baseline;
9. exports bootstrap 95% confidence intervals, fold metrics, out-of-sample predictions, figures and an HTML report.

Primary outputs:

outputs/real_dataset/
  ai_aquatica_real_dataset_report.html
  processed_water_quality.csv
  validation_summary.csv
  artifacts/
    ion_balance_diagnostics.csv
    station_classification_grouped_metrics.json
    station_classification_grouped_confusion_matrix.csv
    station_classification_grouped_confusion_matrix.png
    station_classification_grouped_predictions.csv
    station_classification_grouped_fold_metrics.csv
    station_classification_qc_filtered_*.csv/json/png
    chlorophyll_regression_grouped_metrics.json
    chlorophyll_regression_grouped_predictions.csv
    chlorophyll_regression_grouped_observed_vs_predicted.png
    chlorophyll_regression_grouped_fold_metrics.csv

The script is the source of truth for numerical results reported in the revised manuscript. Results should not be copied from earlier notebooks or releases.

This site is open source. Improve this page.