HTML Reports

generate_report writes a single, self-contained HTML file with every figure embedded. It is what ecgen-eval eval produces, and it can be opened in any modern browser or shared as one file.

Contents

Section	Description
Dataset summary	Recordings, leads, length and sampling rate for each dataset
ECG gallery	Real vs synthetic recordings on ECG graph paper (overlay / sidebyside / grid)
Per-metric sections	Score table, per-lead bar chart, plus a PSD overlay for psd and a violin plot for dtw
Radar chart	All metrics and datasets at a glance, normalised so that the outer edge is best
Metrics summary table	All scores in one table, with a Download CSV button
Clinical feature summary (optional)	Mean ± std per feature, real vs synthetic
2D / 3D feature plots (optional)	Distribution plots of clinical features
References	Citation for every metric used

From the CLI

ecgen-eval eval --real real.npy --synthetic a.npy --synthetic b.npy \
  -m mmd -m fd -m psd -m dtw --feature-plots --output report.html

See Command-Line Interface for all options.

From Python

from ecgen_eval import load_ecg_npy, AVAILABLE_METRICS
from ecgen_eval.features import extract_features
from ecgen_eval.report.html_report import generate_report

real = load_ecg_npy("real.npy", label="PTB-XL")
synths = [load_ecg_npy("a.npy", label="Model A"), load_ecg_npy("b.npy", label="Model B")]

metric_results = {}
for key in ["mmd", "fd", "psd", "dtw", "pr_dist"]:
    metric = AVAILABLE_METRICS[key]()
    metric_results[key] = {
        s.label: metric.compute(real.data, s.data, lead_names=real.lead_names) for s in synths
    }

_, feat_real = extract_features(real)
feat_synth = [extract_features(s)[1] for s in synths]

generate_report(
    real=real,
    synthetics=synths,
    metric_results=metric_results,
    output_path="report.html",
    n_ecg_samples=5,
    leads=["I", "II", "V1", "V5"],
    compare_mode="sidebyside",
    plotly_cdn=True,
    feature_df_real=feat_real,
    feature_dfs_synth=feat_synth,
    feature_plots_2d=True,
    feature_plots_3d=True,
)

Parameters

Parameter	Default	Description
real	—	Real ECGDataset
synthetics	—	List of synthetic ECGDatasets
metric_results	—	{metric_key: {synth_label: MetricResult}}
output_path	—	Path of the HTML file to write
n_ecg_samples	3	Recordings shown in the gallery
leads	None	Subset of leads shown in the gallery (all if None)
compare_mode	"overlay"	overlay, sidebyside or grid
plotly_cdn	True	Load Plotly.js from a CDN (small file). False embeds it (~3–4 MB larger, works offline)
fs	500.0	Sampling rate used for PSD frequency axes
feature_df_real	None	Real summary_df from feature extraction
feature_dfs_synth	None	List of synthetic summary_dfs, same order as synthetics
feature_plots_2d	False	Add 2D feature distribution plots
feature_plots_3d	False	Add 3D feature distribution plots

Returns the Path of the written file.

Offline viewing

Reports load Plotly.js from a CDN by default. Use plotly_cdn=False (--offline on the CLI) when the report will be opened without internet access, e.g. on a secure cluster.