Installation¶

Install from PyPI¶

pip install deepecgkit

Or with uv:

uv add deepecgkit

Development Install¶

Clone the repository and install with development dependencies:

git clone https://github.com/stevenah/deepecg-kit.git
cd deepecg-kit
uv sync --group dev

Or, without uv:

pip install -e ".[dev]"

This installs the package in editable mode along with testing and linting tools (pytest, ruff).

Building Documentation Locally¶

To build and preview the documentation site:

uv sync --group docs
make docs-serve

This starts a local server at http://127.0.0.1:8000 with live-reloading.

Where Data Is Stored¶

Datasets are downloaded on first use into ~/.deepecgkit/datasets, one folder per dataset. deepecg list-datasets shows each dataset's download size and whether a local copy exists.

To keep them elsewhere — a scratch disk, a shared project folder on a cluster — set DEEPECGKIT_HOME:

export DEEPECGKIT_HOME=/scratch/$USER/deepecgkit

Datasets then live in $DEEPECGKIT_HOME/datasets and cached weights in $DEEPECGKIT_HOME/weights. A single dataset can also be pointed at a specific folder with --data-dir (CLI) or data_dir= (Python).

Downloads are fetched from PhysioNet's open-data mirror, fall back to physionet.org when the mirror cannot be reached, and resume from where they stopped if interrupted.

Hardware¶

DeepECG-Kit automatically detects available hardware:

Device	Detection
CUDA GPU	Used automatically if `torch.cuda.is_available()`
Apple MPS	Used automatically if `torch.backends.mps.is_available()`
CPU	Fallback when no GPU is available. On Apple Silicon, PyTorch's NNPACK backend is switched off for CPU runs, where it makes convolutions many times slower

You can override device selection with --accelerator cpu|gpu|mps (CLI) or device="cpu" (Python API).

Requirements¶

Python >= 3.10
PyTorch >= 2.1.0
See pyproject.toml for the full dependency list