Tutorial

A short introduction to the main building blocks. Complete, runnable versions of these patterns are provided in the examples.

1. Read a database

Every supported database has a reader class in torch_ecg.databases that handles local caching, record listing and signal/annotation loading:

from torch_ecg.databases import LUDB

reader = LUDB("~/data/ludb")      # local storage path
reader.download()                  # from the physionet-open S3 bucket
data = reader.load_data(0)         # (n_leads, n_samples) array
ann = reader.load_ann(0)           # waveform annotations

2. Wrap it in a dataset

Datasets pair readers with a training configuration (leads, classes, segment length, …) and yield tensors:

from torch_ecg.databases.datasets import LUDBDataset, LUDBTrainCfg

from copy import deepcopy

train_cfg = deepcopy(LUDBTrainCfg)
train_cfg.db_dir = "~/data/ludb"
train_cfg.n_epochs = 10
ds_train = LUDBDataset(train_cfg, training=True, lazy=False)

3. Pick a model

Models take a config from torch_ecg.model_configs (or a custom CFG), together with the class list and the number of leads:

from torch_ecg.models import ECG_CRNN
from torch_ecg.model_configs import ECG_CRNN_CONFIG

from copy import deepcopy

config = deepcopy(ECG_CRNN_CONFIG)
config.cnn.name = "resnet_nature_comm_bottle_neck_se"
model = ECG_CRNN(classes=classes, n_leads=12, config=config)

4. Train and reload

Subclass BaseTrainer (or reuse one of the benchmark trainers) and call train(). Checkpoints written by the trainer contain the full model config, so a trained model can be reloaded with a single call:

model, train_config = ECG_CRNN.from_checkpoint("path/to/checkpoint_dir")