CLI Usage

DeepECG-Kit provides the deepecg command for training, evaluation, and inference.

Global Options

Flag	Description
--verbose, -v	Enable verbose output
--quiet, -q	Suppress non-essential output
--config, -c	Path to configuration file (YAML or JSON)

Commands

deepecg train

Train a model on a dataset.

deepecg train -m kanres -d af-classification --epochs 50

Flag	Default	Description
--model, -m	required	Model architecture (see deepecg list-models)
--dataset, -d	required	Dataset to train on (see deepecg list-datasets)
--data-dir	auto-detect	Directory containing the dataset
--output-dir, -o	runs/{timestamp}-{model}-{dataset}	Output directory
--weights, -w	None	Pretrained weight name or file path
--epochs, -e	50	Number of training epochs
--batch-size, -b	32	Batch size
--learning-rate, --lr	0.001	Learning rate
--val-split	0.2	Validation split ratio
--test-split	0.1	Test split ratio
--num-workers	4	Data loading workers
--accelerator	auto	Device: auto, cpu, gpu, mps
--devices	1	Number of devices
--force-download	false	Force re-download the dataset
--early-stopping-patience	10	Epochs without improvement before stopping
--seed	42	Random seed
--multi-label	false	Use multi-label classification with BCE loss
--sampling-rate	None	Target sampling rate in Hz
--binary-classification	false	Binary AF vs Non-AF classification
--normalization	None	Normalization method: zscore, minmax, none

deepecg evaluate

Evaluate a trained model on a dataset split.

deepecg evaluate --checkpoint model.pt -m kanres -d af-classification

Flag	Default	Description
--checkpoint	required	Path to model checkpoint
--model, -m	required	Model architecture
--dataset, -d	required	Dataset to evaluate on
--data-dir	None	Directory containing the dataset
--batch-size, -b	32	Batch size
--num-workers	4	Data loading workers
--accelerator	auto	Device selection
--devices	1	Number of devices
--force-download	false	Force re-download the dataset
--split	test	Dataset split: test or val

deepecg predict

Run inference on an ECG file.

deepecg predict --checkpoint model.pt -m kanres --input ecg.npy

Flag	Default	Description
--checkpoint	required	Path to model checkpoint
--model, -m	required	Model architecture
--input, -i	required	Input file (.npy, .csv, or WFDB .dat/.hea)
--output, -o	None	Save predictions to JSON file
--batch-size, -b	1	Batch size
--accelerator	auto	Device selection

deepecg resume

Resume training from a checkpoint.

deepecg resume --checkpoint model.pt -m kanres --epochs 100

Flag	Default	Description
--checkpoint	required	Path to checkpoint to resume from
--model, -m	required	Model architecture
--epochs, -e	None	Additional epochs (default: continue original)
--output-dir, -o	None	Directory to save new checkpoints
--accelerator	auto	Device selection
--devices	1	Number of devices
--early-stopping-patience	10	Early stopping patience

deepecg info

Show model information.

deepecg info -m kanres

deepecg list-models

List all available model architectures.

deepecg list-models

deepecg list-datasets

List all available datasets.

deepecg list-datasets

Configuration Files

Instead of passing all flags on the command line, you can use a YAML or JSON config file:

train:
  learning_rate: 0.001
  epochs: 100
  batch_size: 64
  val_split: 0.2
  early_stopping_patience: 15

deepecg train -m kanres -d af-classification --config train_config.yaml

CLI flags override values from the config file.