test_docker.py

"""
"""

################################################################################
# NOTE comment out to test in GitHub actions

# import sys
# from pathlib import Path

# sys.path.insert(0, "/home/wenhao/Jupyter/wenhao/workspace/torch_ecg/")
# sys.path.insert(0, "/home/wenhao/Jupyter/wenhao/workspace/bib_lookup/")
# tmp_data_dir = Path("/home/wenhao/Jupyter/wenhao/data/CinC2022/")
################################################################################

# set test flag
from cfg import set_entry_test_flag

set_entry_test_flag(True)


from copy import deepcopy  # noqa: E402

import numpy as np  # noqa: E402
import torch  # noqa: E402
from cfg import _BASE_DIR, ModelCfg, TrainCfg  # noqa: E402
from data_reader import CINC2016Reader, CINC2022Reader, EPHNOGRAMReader  # noqa: F401 E402
from dataset import CinC2022Dataset  # noqa: E402
from models import CRNN_CINC2022, SEQ_LAB_NET_CINC2022, UNET_CINC2022  # noqa: E402
from outputs import CINC2022Outputs  # noqa: E402
from torch.nn.parallel import DataParallel as DP  # noqa: E402
from torch.nn.parallel import DistributedDataParallel as DDP  # noqa: F401 E402
from torch.utils.data import DataLoader, Dataset  # noqa: F401, E402
from trainer import _MODEL_MAP, CINC2022Trainer, _set_task  # noqa: E402
from utils.scoring_metrics import compute_challenge_metrics  # noqa: E402

from torch_ecg.components.outputs import ClassificationOutput  # noqa: E402
from torch_ecg.utils.utils_nn import default_collate_fn as collate_fn  # noqa: E402

CRNN_CINC2022.__DEBUG__ = False
SEQ_LAB_NET_CINC2022.__DEBUG__ = False
UNET_CINC2022.__DEBUG__ = False
CinC2022Dataset.__DEBUG__ = False


DEVICE = torch.device("cuda" if torch.cuda.is_available() else "cpu")
if ModelCfg.torch_dtype == torch.float64:
    torch.set_default_tensor_type(torch.DoubleTensor)
    DTYPE = np.float64
else:
    DTYPE = np.float32


################################################################################
# NOTE: uncomment to test in GitHub actions

tmp_data_dir = _BASE_DIR / "tmp" / "CINC2022"
tmp_data_dir.mkdir(parents=True, exist_ok=True)
dr = CINC2022Reader(tmp_data_dir)
dr.download(compressed=True)
dr._ls_rec()
del dr
################################################################################

TASK = "classification"  # "multi_task"


def test_dataset() -> None:
    """ """
    ds_config = deepcopy(TrainCfg)
    ds_config.db_dir = tmp_data_dir

    ds_train = CinC2022Dataset(ds_config, TASK, training=True, lazy=True)
    ds_val = CinC2022Dataset(ds_config, TASK, training=False, lazy=True)

    ds_train._load_all_data()
    ds_val._load_all_data()

    print("dataset test passed")


def test_models() -> None:
    """ """
    model = CRNN_CINC2022(ModelCfg[TASK])
    model.to(DEVICE)
    ds_config = deepcopy(TrainCfg)
    ds_config.db_dir = tmp_data_dir
    ds_val = CinC2022Dataset(ds_config, TASK, training=False, lazy=True)
    ds_val._load_all_data()
    dl = DataLoader(
        dataset=ds_val,
        batch_size=16,
        shuffle=True,
        num_workers=0,
        pin_memory=True,
        drop_last=False,
        collate_fn=collate_fn,
    )
    for idx, input_tensors in enumerate(dl):
        waveforms = input_tensors.pop("waveforms").to(DEVICE)
        # input_tensors = {k: v.to(DEVICE) for k, v in input_tensors.items()}
        # out_tensors = model(waveforms, input_tensors)
        print(model.inference(waveforms))
        if idx > 10:
            break

    print("models test passed")


def test_challenge_metrics() -> None:
    """ """
    outputs = [
        CINC2022Outputs(
            murmur_output=ClassificationOutput(
                classes=["Present", "Unknown", "Absent"],
                prob=np.array([[0.75, 0.15, 0.1]]),
                pred=np.array([0]),
                bin_pred=np.array([[1, 0, 0]]),
            ),
            outcome_output=ClassificationOutput(
                classes=["Abnormal", "Normal"],
                prob=np.array([[0.6, 0.4]]),
                pred=np.array([0]),
                bin_pred=np.array([[1, 0]]),
            ),
            segmentation_output=None,
        ),
        CINC2022Outputs(
            murmur_output=ClassificationOutput(
                classes=["Present", "Unknown", "Absent"],
                prob=np.array([[0.3443752, 0.32366553, 0.33195925]]),
                pred=np.array([0]),
                bin_pred=np.array([[1, 0, 0]]),
            ),
            outcome_output=ClassificationOutput(
                classes=["Abnormal", "Normal"],
                prob=np.array([[0.5230, 0.0202]]),
                pred=np.array([0]),
                bin_pred=np.array([[1, 0]]),
            ),
            segmentation_output=None,
        ),
    ]
    labels = [
        {
            "murmur": np.array([[0.0, 0.0, 1.0]]),
            "outcome": np.array([0]),
        },
        {
            "murmur": np.array([[0.0, 1.0, 0.0]]),
            "outcome": np.array([1]),
        },
    ]

    compute_challenge_metrics(labels, outputs)

    print("challenge metrics test passed")


def test_trainer() -> None:
    """ """
    train_config = deepcopy(TrainCfg)
    train_config.db_dir = tmp_data_dir
    # train_config.model_dir = model_folder
    # train_config.final_model_filename = "final_model.pth.tar"
    train_config.debug = True

    train_config.n_epochs = 20
    train_config.batch_size = 24  # 16G (Tesla T4)
    # train_config.log_step = 20
    # # train_config.max_lr = 1.5e-3
    # train_config.early_stopping.patience = 20

    # train_config[TASK].cnn_name = "resnet_nature_comm_bottle_neck_se"
    # train_config[TASK].rnn_name = "none"  # "none", "lstm"
    # train_config[TASK].attn_name = "se"  # "none", "se", "gc", "nl"

    _set_task(TASK, train_config)

    model_config = deepcopy(ModelCfg[TASK])

    # adjust model choices if needed
    model_name = model_config.model_name = train_config[TASK].model_name
    model_config[model_name].cnn_name = train_config[TASK].cnn_name
    model_config[model_name].rnn_name = train_config[TASK].rnn_name
    model_config[model_name].attn_name = train_config[TASK].attn_name

    model_cls = _MODEL_MAP[model_config.model_name]
    model_cls.__DEBUG__ = False

    model = model_cls(config=model_config)
    if torch.cuda.device_count() > 1:
        model = DP(model)
        # model = DDP(model)
    model.to(device=DEVICE)

    trainer = CINC2022Trainer(
        model=model,
        model_config=model_config,
        train_config=train_config,
        device=DEVICE,
        lazy=False,
    )

    best_state_dict = trainer.train()

    print("trainer test passed")


from run_model import run_model  # noqa: E402
from train_model import train_challenge_model  # noqa: E402


def test_entry() -> None:
    """ """

    data_folder = str(tmp_data_dir / "training_data")
    train_challenge_model(data_folder, str(TrainCfg.model_dir), verbose=2)

    output_dir = _BASE_DIR / "tmp" / "output"
    output_dir.mkdir(parents=True, exist_ok=True)
    run_model(
        TrainCfg.model_dir,
        data_folder,
        str(output_dir),
        allow_failures=False,
        verbose=2,
    )

    print("entry test passed")


test_team_code = test_entry  # alias


if __name__ == "__main__":
    # test_dataset()  # passed
    # test_models()  # passed
    # test_trainer()  # directly run test_entry
    test_challenge_metrics()
    test_entry()
    set_entry_test_flag(False)