Skip to content

Visualization

Jump to bottom
martanto edited this page Feb 23, 2026 · 9 revisions

Visualization

All plots use colorblind-safe palettes and high-DPI output suitable for publications and presentations.

Tremor Time-Series Plots

plot_tremor()

Visualize RSAM, DSAR, and Shannon Entropy as multi-panel time-series plots.

from eruption_forecast.plots.tremor_plots import plot_tremor
import pandas as pd

df = pd.read_csv("output/VG.OJN.00.EHZ/tremor/tremor.csv", index_col=0, parse_dates=True)

# Basic plot — all columns
plot_tremor(df=df, figure_dir="output/figures", dpi=150)

# Custom interval and selected columns
plot_tremor(
    df=df,
    interval=6,
    interval_unit="hours",
    selected_columns=["rsam_f2", "rsam_f3", "dsar_f2-f3"],
    figure_dir="output/figures",
    filename="tremor_selected",
    dpi=300,
)

replot_tremor() — Batch Replot

Regenerate plots for all daily tremor CSV files (sequential or parallel).

from eruption_forecast.plots.tremor_plots import replot_tremor

results = replot_tremor(
    daily_dir="output/VG.OJN.00.EHZ/tremor/daily",
    output_dir="output/VG.OJN.00.EHZ/tremor/figures",
    n_jobs=4,
    dpi=300,
    overwrite=False,
)
print(f"Created: {results['created']}, Skipped: {results['skipped']}, Failed: {results['failed']}")

Feature Importance Plots

plot_significant_features()

Horizontal bar chart of feature importance or p-values.

from eruption_forecast.plots.feature_plots import plot_significant_features

plot_significant_features(
    features="path/to/features.csv",  # or a DataFrame
    number_of_features=50,
    top_features=20,                   # Highlighted with darker colour
    values_column="importance",        # or "p_values"
    output_dir="output/figures",
    filename="feature_importance",
    dpi=150,
)

replot_significant_features() — Batch Replot

from eruption_forecast.plots.feature_plots import replot_significant_features

results = replot_significant_features(
    all_features_dir="output/trainings/features/all_features",
    output_dir="output/trainings/features/figures/significant",
    n_jobs=4,
    number_of_features=50,
    top_features=20,
    dpi=300,
    overwrite=False,
)

Model Evaluation Plots

Single-Seed (ModelEvaluator)

plot_all() generates all 7–8 plots at once. Individual methods:

from eruption_forecast import ModelEvaluator

evaluator = ModelEvaluator.from_files(
    model_path="output/.../models/00042.pkl",
    X_test="output/.../tests/00042_X_test.csv",
    y_test="output/.../tests/00042_y_test.csv",
    model_name="xgb_seed_42",
    output_dir="output/eval",
)

evaluator.plot_all()                      # All plots
evaluator.plot_confusion_matrix()
evaluator.plot_roc_curve()
evaluator.plot_precision_recall_curve()
evaluator.plot_threshold_analysis()
evaluator.plot_feature_importance()
evaluator.plot_calibration()
evaluator.plot_prediction_distribution()
evaluator.plot_shap_summary(max_display=20)  # Requires shap>=0.46

Aggregate (MultiModelEvaluator)

Requires train_and_evaluate() output — loads test data per seed and aggregates.

from eruption_forecast import MultiModelEvaluator

base = "output/trainings/model-with-evaluation/xgb-classifier/stratified-shuffle-split"
ev = MultiModelEvaluator(
    trained_model_csv=f"{base}/trained_model_XGBClassifier-StratifiedShuffleSplit_rs-0_ts-500_top-20.csv"
)

figs = ev.plot_all(dpi=150, show_individual=True)
# Keys: roc_curve, pr_curve, calibration, prediction_distribution,
#       confusion_matrix, threshold_analysis, feature_importance,
#       shap_summary, seed_stability, frequency_band_contribution

Aggregation Strategy

Plot How seeds are combined
ROC Curve Each seed's TPR interpolated onto shared FPR grid → mean ± std band
Precision-Recall Precision interpolated onto shared recall grid → mean ± std band
Threshold Analysis Metrics computed per threshold per seed → mean ± std bands
Calibration Fraction of positives interpolated → mean ± std band
Prediction Distribution Predicted probabilities pooled across seeds → single KDE per class
Confusion Matrix Raw confusion matrices summed across all seeds
Feature Importance Importances stacked → mean bar + std error bar

Classifier Comparison Heatmap

Compare metrics across multiple classifiers. Each cell shows mean ± std; rows sorted by mean F1 descending.

from eruption_forecast import MultiModelEvaluator
from eruption_forecast.plots import plot_classifier_comparison

base = "output/trainings/model-with-evaluation"
metrics_by_clf = {}
for clf in ["xgb", "rf", "gb"]:
    ev = MultiModelEvaluator(
        metrics_dir=f"{base}/{clf}-classifier/stratified-shuffle-split/metrics"
    )
    metrics_by_clf[clf] = ev.get_metrics_list()

fig, summary_df = plot_classifier_comparison(
    metrics_by_classifier=metrics_by_clf,
    metrics_to_show=["balanced_accuracy", "f1_score", "precision", "recall", "roc_auc", "pr_auc"],
    figsize=(12, 5),
    dpi=150,
)
fig.savefig("classifier_comparison.png", bbox_inches="tight")

SHAP Explainability

Understand which features drive predictions and in which direction.

from eruption_forecast import ModelEvaluator, MultiModelEvaluator

# Single-seed beeswarm
evaluator = ModelEvaluator.from_files(
    model_path="output/.../models/00042.pkl",
    X_test="output/.../tests/00042_X_test.csv",
    y_test="output/.../tests/00042_y_test.csv",
    model_name="xgb_seed_42",
)
fig = evaluator.plot_shap_summary(max_display=20)

# Aggregate mean |SHAP| bar chart across seeds
ev = MultiModelEvaluator(trained_model_csv="output/.../trained_model_registry.csv")
fig = ev.plot_shap_summary(max_display=20)

Requires shap >= 0.46.

Seed Stability Plot

Visualize metric variability across random seeds using a violin + strip plot.

from eruption_forecast import MultiModelEvaluator
from eruption_forecast.plots import plot_seed_stability

# Single classifier
ev = MultiModelEvaluator(
    metrics_dir="output/.../xgb-classifier/stratified-shuffle-split/metrics"
)
fig = ev.plot_seed_stability(metric="f1_score")

# Compare multiple classifiers side-by-side
metrics_by_clf = {}
for clf in ["xgb", "rf", "gb", "svm"]:
    ev = MultiModelEvaluator(metrics_dir=f"output/.../{clf}/metrics")
    metrics_by_clf[clf] = ev.get_metrics_list()

fig, df = plot_seed_stability(
    metrics_by_classifier=metrics_by_clf,
    metric="balanced_accuracy",
    dpi=150,
)

Frequency Band Contribution

Shows which seismic frequency bands dominate the selected features. RSAM bands are blue; DSAR bands are orange.

from eruption_forecast import MultiModelEvaluator

# Multi-seed, mean ± std per band
ev = MultiModelEvaluator(trained_model_csv="output/.../trained_model_registry.csv")
fig = ev.plot_frequency_band_contribution()

# Single-seed standalone
from eruption_forecast.plots.feature_plots import plot_frequency_band_contribution
import pandas as pd

features = pd.read_csv("output/.../significant_features.csv", index_col=0).index.tolist()
fig, df = plot_frequency_band_contribution(feature_names=features)
# df columns: band, count

Forecast Visualization

ModelPredictor.predict_proba(plot=True) automatically generates an eruption probability time-series plot saved to figures/eruption_forecast.png.

  • Per-classifier lines (dashed)
  • Consensus line (solid black)
  • Uncertainty band (shaded ± std)
  • Eruption event markers (if labels provided)

Common Parameters

Parameter Type Default Description
dpi int 150 Resolution (use 300 for publication)
overwrite bool True Replace existing plots
n_jobs int 1 Parallel workers for batch utilities
output_dir / figure_dir str varies Directory for saved plots
verbose bool False Log plot generation
filename str auto Custom filename stem (extension added automatically)

Import Reference

from eruption_forecast.plots.tremor_plots import plot_tremor, replot_tremor
from eruption_forecast.plots.feature_plots import (
    plot_significant_features,
    replot_significant_features,
    plot_frequency_band_contribution,
)
from eruption_forecast.plots.shap_plots import plot_shap_summary, plot_aggregate_shap_summary
from eruption_forecast.plots import plot_classifier_comparison, plot_seed_stability

# Top-level shortcuts
from eruption_forecast import ModelEvaluator, MultiModelEvaluator

Clone this wiki locally