SRECG 1.1

SRECG is an ECG signal super-resolution framework for improving low-resolution ECG signals used in cardiac arrhythmia classification.

This repository update is the SRECG 1.1 manual-upload version. It is intentionally limited to three root-level files:

1. README.md
2. SRECG.ipynb
3. requirements.txt

No generated result folders or auxiliary files are required for this lightweight GitHub update. Existing repository files and folders such as comparison/, LICENSE, or prior benchmark artifacts should be left unchanged unless you explicitly choose to regenerate and upload new outputs later.

What changed in 1.1

• Rewrites the end-to-end experiment notebook as a PyTorch workflow.
• Keeps the project centered on the single executable notebook: SRECG.ipynb.
• Adds Kaggle/local path discovery for the CPSC2018 dataset.
• Uses memory-mapped NumPy cache handling to reduce RAM pressure during repeated ECG loading.
• Keeps the 10-fold evaluation workflow in the notebook.
• Updates requirements.txt to a minimal runtime dependency list instead of an environment-specific full freeze.

Repository layout after manual upload

Only the following root files need to be uploaded/replaced for this 1.1 update:

SRECG/
├── README.md
├── SRECG.ipynb
├── requirements.txt
├── comparison/          # existing folder; no manual upload required for this lightweight update
├── LICENSE              # existing file; keep unchanged
└── ...                  # other existing repository files remain unchanged

Dataset expectation

The notebook expects the CPSC2018 dataset in the same structure used by the original experiment:

CPSC2018Dataset/
├── REFERENCE.csv
├── TrainingSet1/
├── TrainingSet2/
└── TrainingSet3/

The notebook automatically searches common local and Kaggle paths, including:

./CPSC2018Dataset
../input/CPSC2018Dataset
/kaggle/input/CPSC2018Dataset
/kaggle/input/cpsc2018dataset

If your Kaggle dataset path is different, edit the dataset path setting near the top of SRECG.ipynb.

Installation

Install the runtime dependencies with:

python -m pip install -r requirements.txt

Then open and run:

SRECG.ipynb

Important runtime switches

The notebook exposes key configuration variables near the top:

Variable	Default	Purpose
FAST_DEV_RUN	False	Use True only for a small smoke test.
REUSE_EXISTING_CHECKPOINTS	True	Reuse trained checkpoints if available.
REBUILD_DATA_CACHE	False	Rebuild the memory-mapped data cache when needed.
WRITE_REQUIREMENTS	False	Avoid overwriting the lightweight requirements.txt during normal runs.
SRECG_BATCH_SIZE	64	Optional environment override for batch size.
SRECG_EPOCHS	100	Optional environment override for training epochs.
SRECG_DETERMINISTIC	0	Set to 1 for stricter deterministic behavior at the cost of speed.

GitHub notebook preview note

SRECG.ipynb in this manual-upload package is saved as a GitHub-renderable notebook: code and Markdown are preserved, but saved cell outputs are cleared before upload. This avoids GitHub preview failures such as Unable to render code block caused by large embedded notebook outputs.

This does not change the experiment source code or runtime logic. Running the notebook regenerates the outputs from the same cells.

Manual GitHub upload procedure

Use GitHub's web interface to replace exactly these three files at the repository root:

README.md
SRECG.ipynb
requirements.txt

Recommended commit message:

Release SRECG v1.1 manual notebook update

Do not upload the entire ZIP file into the repository. Extract it first, then upload the three files individually.

Notes on reproducibility

This manual-upload version updates the runnable notebook and dependency declaration only. It does not upload regenerated plots, CSV summaries, NumPy result arrays, fold-index files, checkpoints, or cache files. Those artifacts should be treated as notebook-generated outputs and can be regenerated later if needed.

References

1. Chen et al., SRECG: ECG Signal Super-Resolution Framework for Portable/Wearable Devices in Cardiac Arrhythmias Classification.
2. CPSC2018 ECG dataset workflow as used by the original SRECG experiment.

License

This repository retains the original GPL-3.0 license.