Skip to content

Output Structure

Jump to bottom
Martanto edited this page Apr 26, 2026 · 12 revisions

Output Structure

All outputs are written under:

{output_dir}/{network}.{station}.{location}.{channel}/

For example, with network="VG", station="OJN", location="00", channel="EHZ":

output/VG.OJN.00.EHZ/

Full Directory Tree

output/
└── VG.OJN.00.EHZ/
    │
    ├── tremor/
    │   ├── daily/                            # Per-day CSV files (removed if cleanup_daily_dir=True)
    │   ├── figures/                          # Daily tremor plots (created if plot_daily=True)
    │   ├── tremor_*.csv                      # Final merged tremor data
    │   └── matrix/                           # Tremor matrix outputs
    │       ├── tremor_matrix_unified_*.csv   # Aligned tremor matrix (all columns)
    │       ├── per_method/                   # Per-column tremor matrices (optional)
    │       │   ├── tremor_matrix_{col}_{start}_{end}.csv
    │       │   └── ...
    │       └── tmp/                          # Per-window CSVs (save_tremor_matrix_per_id=True only)
    │
    ├── features/
    │   ├── extracted/
    │   │   ├── train/                        # Per-column tsfresh CSVs (training mode)
    │   │   └── forecast/                     # Per-column tsfresh CSVs (prediction mode)
    │   ├── all_features_*.csv                # Concatenated features (training)
    │   └── label-features_*.csv             # Labels aligned with features
    │
    ├── trainings/
    │   │
    │   ├── evaluations/            # Output of train(with_evaluation=True)
    │   │   ├── evaluations_trained_models.json   # {ClassifierName: trained_model_*.csv} registry
    │   │   ├── evaluations_config.yaml           # Pipeline config snapshot (written by save_model)
    │   │   ├── evaluations_forecast_model.pkl    # Serialised ForecastModel (written by save_model)
    │   │   ├── features/                         # Shared across all classifiers
    │   │   │   └── {cv-slug}/                    # e.g., stratified-shuffle-split
    │   │   │       ├── significant_features/     # Per-seed top-N features
    │   │   │       │   ├── 00000.csv
    │   │   │       │   └── ...
    │   │   │       ├── all_features/             # All ranked features (optional)
    │   │   │       │   ├── 00000.csv
    │   │   │       │   └── ...
    │   │   │       ├── figures/significant/      # Feature importance plots (optional)
    │   │   │       │   └── 00000.jpg
    │   │   │       ├── tests/                    # Per-seed held-out test splits
    │   │   │       │   ├── 00000_X_test.csv
    │   │   │       │   ├── 00000_y_test.csv
    │   │   │       │   └── ...
    │   │   │       ├── significant_features.csv  # Aggregated top features (all seeds)
    │   │   │       └── top_{n}_significant_features.csv
    │   │   └── classifiers/                      # Per-classifier outputs
    │   │       └── {classifier-slug}/            # e.g., random-forest-classifier
    │   │           └── {cv-slug}/                # e.g., stratified-shuffle-split
    │   │               ├── models/
    │   │               │   ├── 00000.pkl                 # Trained model — seed 0
    │   │               │   ├── 00001.pkl
    │   │               │   └── ...
    │   │               ├── metrics/
    │   │               │   ├── 00000.json                # Per-seed metrics JSON
    │   │               │   └── ...
    │   │               ├── figures/                      # Aggregate evaluation plots
    │   │               │   ├── aggregate_roc_curve.png / .csv
    │   │               │   ├── aggregate_pr_curve.png / .csv
    │   │               │   ├── aggregate_calibration.png / .csv
    │   │               │   ├── aggregate_prediction_distribution.png / .csv
    │   │               │   ├── aggregate_confusion_matrix.png / .csv
    │   │               │   ├── aggregate_threshold_analysis.png / .csv
    │   │               │   ├── aggregate_feature_importance.png / .csv
    │   │               │   ├── aggregate_shap_summary.png
    │   │               │   ├── aggregate_shap_summary.pkl   # shap.Explanation (joblib)
    │   │               │   └── aggregate_metrics.csv
    │   │               ├── trained_model_{suffix}.csv    # Registry of all trained models
    │   │               ├── merged_model_{suffix}.pkl     # SeedEnsemble (all seeds merged — optional)
    │   │               ├── all_metrics_{suffix}.csv      # All per-seed metrics
    │   │               └── metrics_summary_{suffix}.csv  # Mean ± std summary
    │   │
    │   └── predictions/                       # Output of train(with_evaluation=False)
    │       ├── predictions_trained_models.json   # {ClassifierName: trained_model_*.csv} registry
    │       ├── predictions_config.yaml           # Pipeline config snapshot (written by save_model)
    │       ├── predictions_forecast_model.pkl    # Serialised ForecastModel (written by save_model)
    │       ├── features/                         # Shared across all classifiers
    │       │   └── {cv-slug}/
    │       │       ├── significant_features/
    │       │       │   ├── 00000.csv
    │       │       │   └── ...
    │       │       ├── all_features/ (optional)
    │       │       ├── figures/significant/ (optional)
    │       │       ├── significant_features.csv
    │       │       └── top_{n}_significant_features.csv
    │       └── classifiers/                      # Per-classifier outputs
    │           └── {classifier-slug}/
    │               └── {cv-slug}/
    │                   ├── models/
    │                   │   ├── 00000.pkl
    │                   │   └── ...
    │                   ├── trained_model_{suffix}.csv    # Registry used by ModelPredictor
    │                   └── merged_model_{suffix}.pkl     # SeedEnsemble (optional — call merge_models())
    │
    ├── trainings/merged_classifiers_{suffix}.pkl   # Multi-classifier bundle (optional — call merge_classifier_models())
    │
    ├── forecast/
    │   ├── predictions.csv                   # Forecast output (predict_proba)
    │   └── figures/
    │       └── eruption_forecast.png
    │
    ├── config_forecast.yaml                  # Pipeline config snapshot (written by forecast())
    └── forecast_model.pkl                    # Serialised ForecastModel (default path for save_model())

File Name Suffixes

Training output files follow this naming pattern:

{ClassifierName}-{CVName}_rs-{random_state}_ts-{total_seed}_top-{n}

Example:

XGBClassifier-StratifiedShuffleSplit_rs-0_ts-500_top-20

So the model registry is:

trained_model_XGBClassifier-StratifiedShuffleSplit_rs-0_ts-500_top-20.csv

Classifier and CV Slugs

Classifier key Folder slug
rf random-forest-classifier
xgb xgb-classifier
gb gradient-boosting-classifier
svm svm-classifier
lr logistic-regression-classifier
nn mlp-classifier
dt decision-tree-classifier
knn knn-classifier
nb gaussian-nb-classifier
voting voting-classifier
CV strategy key Folder slug
shuffle shuffle-split
stratified stratified-k-fold
shuffle-stratified stratified-shuffle-split
timeseries time-series-split

ModelPredictor Output

Evaluation mode (predict() / predict_best())

{output_dir}/
├── metrics/
│   ├── all_metrics.csv
│   └── metrics_summary.csv
└── seed_00000/                 # Only created when plot=True
    ├── seed_00000_confusion_matrix.png
    ├── seed_00000_roc_curve.png
    ├── seed_00000_pr_curve.png
    ├── seed_00000_threshold_analysis.png
    ├── seed_00000_feature_importance.png
    ├── seed_00000_calibration.png
    └── seed_00000_prediction_distribution.png

Forecast mode (predict_proba())

{output_dir}/
├── predictions.csv             # eruption_probability, uncertainty, confidence, prediction
└── figures/
    └── eruption_forecast.png

Clone this wiki locally