Torch-ecg

----------------------------------------------------------------------------------------

Fixed
~~~~~

- Import errors for early versions of pytorch.
- Cached table of PhysioNet databases is added as ``package_data`` in
``setup.py`` to avoid the error of missing the table file when
installing the package.

Security
~~~~~~~~

----------------------------------------------------------------------------------------

Added
~~~~~

- Add methods ``__len__`` and ``__getitem__`` for the base class
``torch_ecg.databases.base._DataBase``.

Changed
~~~~~~~

- The base class of ``CPSC2021`` is changed from ``CPSCDataBase`` to
``PhysioNetDataBase``.
- Function ``compute_output_shape`` is enhanced to support different
paddings in two ends of the input signal.
- ``README`` is updated.
- Docstrings of many classes and functions are updated.
- ``black`` is used for code formatting.

----------------------------------------------------------------------------------------

Added
~~~~~

- Add cached table of PhysioNet databases as a data file
stored in the package.
- Add ``requests`` as a dependency in the ``requirements.txt`` file.

Changed
~~~~~~~

- An optional argument ``btype`` is added to the function
``butter_bandpass_filter`` to specify the type of the filter:
``"lohi"``, ``"hilo"``.
- A ``compressed`` argument is added to the ``download`` method of the
``PhysioNetDataBase`` class to specify whether to download the
compressed version of the database.

Fixed
~~~~~

- Fix bugs in the function ``preprocess_multi_lead_signal``.

----------------------------------------------------------------------------------------

Added
~~~~~

- Add ``ReprMixin`` class for better representation of the classes
(e.g., models, preprocessors, database readers, etc.).
- Added model_dir to default config.
- Add ``Dataset`` classes for generating input data for the models:
- ``CINC2020``
- ``CINC2021``
- ``CPSC2019``
- ``CPSC2021``
- ``LUDB``
- Add ``sample-data`` directory for storing sample data for testing.
- Add ``url`` property to the database classes.
- Add utility functions for the computation of metrics.
- Add ``BeatAnn`` class for better annotation of ECG beats.
- Add download utility functions.
- Add ``Output`` classes for the models. The output classes are used to
store the output of the models and provide methods for post-processing.

Changed
~~~~~~~

- Manipulation of custom preprocessor classes is enhanced.
- ``SizeMixin`` class is improved for better computation of the sizes of the models.
- Replace ``os`` with ``pathlib``, which is more flexible for path operations.
- Several database reader classes are updated: mitdb, ltafdb.
- Improve ``PhysioNetDataBase`` by using wfdb built-in methods of
getting database version string and downloading the database.
- Update the ``README`` file.

Removed
~~~~~~~

- Unnecessary imports are removed.

Fixed
~~~~~

- Fix bugs in the ``flush`` method of the ``TxtLogger``.

-----------------------------

This release was yanked.

----------------------------------------------------------------------------------------

Added
~~~~~

- Add ``Preprocessor`` classes for ECG data preprocessing (ref.
``torch_ecg.preprocessors``).
- Add ``Augmenter`` classes for ECG data augmentation (ref.
``torch_ecg.augmenters``).
- Add database reader classes for reading ECG data from different
sources (ref. ``torch_ecg.databases``).
- Add model classes for ECG signal analysis, including classification,
segmentation (R-peak detection, wave delineation, etc., ref.
``torch_ecg.models``).
- Add several benchmark studies for ECG signal analysis tasks:
- CinC2020, multi-label classification.
- CinC2021, multi-label classification.
- CPSC2019, QRS detection.
- CPSC2020, single-label classification.
- CPSC2021, single-label classification.
- LUDB, wave delineation.
ref. the ``benchmarks`` directory of the project.
- Add documentation for the project (ref. ``docs`` directory).
- Add CodeQL action for security analysis (ref. ``.github/workflows``).
- Add unit tests for the project (ref. ``test`` directory).