Clfd

Slightly improved report plots.

Changed

- Improved report plot layout; removed frames around each subplot, they had questionable added value and did not always render well at the edges of the image.

First stable release after a complete rewrite. `clfd` is now extensively tested in a CI pipeline.

Added

- CLI app now saves both a report and an associated PNG plot for each archive processed.
- Report plots now show all the data in one figure: feature values, profile mask and time-phase spike mask (if spike removal was performed).

Fixed

- Formula for autocorrelation feature `acf` has been corrected.
- CLI app now checks whether PSRCHIVE python bindings are installed.
- CLI app now checks that there are no duplicate input archive names when `--outdir` is specified. Before, there could be name collisions between output files.
- CLI app now validates feature names.

Changed

- Minimum required Python version is now 3.9.
- By default, CLI app now uses as many parallel processes as there are CPU cores.
- CLI app won't crash if there are archives that cannot be processed. Instead, the name of the offending archive is logged with the error traceback. This choice was made because the app uses multiprocessing.
- Reports are now in a custom JSON format. They can be read using `Report.load()`.
- `matplotlib` is now a required dependency.
- Removed dependencies: `pandas`, `tables`, `scipy`.

Removed

- CLI app argument `--fmt`. It was pointless considering that the only supported data format is PSRCHIVE.
- CLI app argument `-e, --ext`. The additional extension given to clean output archives is now always `.clfd`, which used to be the default choice.

This new version fixes an issue caused by a change in PSRCHIVE Python API, and starts a much needed modernisation effort of `clfd`. **We are dropping support for Python 2.x from this point.**

Fixed

- Fixed a problem with recent PSRCHIVE versions where the Python API for loading archives has changed. Thanks to Bradley Meyers for discovering and fixing the issue [(3)](https://github.com/v-morello/clfd/issues/3)
- Fixed a warning from the `pandas` library about future changes to its HDF5 API.

Changed

- PEP 517/518 compliant packaging, i.e. no more `setup.py`.

Removed

- Support for Python 2.x
- Removed function `clfd.test()`. Tests are not shipped with the module anymore.

Added
- New features: skewness, excess kurtosis, autocorrelation. These are not sensitive to the scale of the data, and should perform better on high bit depth data (8-bit and more).
- `scipy` is now an additional dependency (for `skew` and `kurtosis` functions).

Fixed
- Fixed a bug that can occur with the new psrchive python3 bindings, where the code was passing sub-integration and channel indices as `numpy.int64` to the `archive.get_Profile()` method instead of a python `int`, and a `TypeError` was raised.

Changed
- Updated statement in `README` about psrchive python bindings, as they are now compatible with python 3 in the latest psrchive version.

Added
- The `clfd` command line application now has a `-o` / `--outdir` option to save all data products to a single output directory