Skip to content

Five-minute data-free inference

This tutorial verifies the repository without downloading a research dataset. It uses one of the four frozen models and a NumPy ECG array.

1. Install

git clone https://github.com/Sanssssssssssssssss/ecg_sqi_fusion.git
cd ecg_sqi_fusion
python -m venv .venv

source .venv/bin/activate
pip install -r requirements.txt

.\.venv\Scripts\Activate.ps1
pip install -r requirements.txt

Python 3.11 is the reference environment.

2. Verify the packaged models

python -m src.ecg_sqi_inference verify-bundles

The command checks every shipped model and profile against the SHA-256 values in pretrained/inference/manifest.json.

3. Create a minimal input

import numpy as np

fs = 125
t = np.arange(10 * fs) / fs
ecg = (0.1 * np.sin(2 * np.pi * 1.2 * t)).astype("float32")
np.save("example.npy", ecg)

This signal only checks the data path; it is not a clinically meaningful test record.

4. Predict

python -m src.ecg_sqi_inference predict \
  --model singlelead-rbfsvm \
  --input example.npy \
  --fs 125 \
  --out example-output

The output directory contains:

  • example_segments.csv: one prediction per complete 10-second segment;
  • all_segments.csv: combined rows for every input record;
  • run_summary.json: model, input, output, segment count, dropped tail, and errors.

Continue with Inference and Docker for real inputs, lead-shape rules, and all four models.