cadop / seg1d

Automated segmentation of one-dimensional (1D) data

Hi there 👋

seg1d's Issues

[Joss Review] Automatic dependency installation

At the moment all dependencies need to be installed manually, because the package itself is imported in the setup.py file.

The easiest fix would be to remove this import and maintain the package version at two different places. If you want to avoid this, a set of other solutions (e.g. bumpversion) exist.

I would argue that the convenience of installing the package with a single command is very important.

[Joss Review] Tests

To my understanding, the package is tested using doctest only. I could consider that a good starting point, but the currently included test can be considered happy-path regression/integration tests. Considering that some of the implemented functions have several inputs and possible execution branches, a more elaborate suite of unittests would be nice.

I would not consider this an exclusion criteria for the review, but I personally would not feel comfortable using the current package in any project of relevance given the current state of testing.

[Joss review] Building the docs

There are a couple of issues with the docs:

Dependencie doumentation:

• numpydoc is required, but not documented in the README
• matplotlib is required, but not documented (btw. matplotlib is also required for the examples, but is not documented there as well)

After trying a couple of things, I was still not able to build the docs or run the doctests. In a clean conda env (Py 3.8) the autodoc-process-signature extension threw a number of errors:

...
WARNING: error while formatting signature for seg1d.algorithm.get_peaks: Handler <function process_numba_docstring at 0x7febb2fbcca0> for event 'autodoc-process-signature' threw an exception
WARNING: error while formatting signature for seg1d.algorithm.resample: Handler <function process_numba_docstring at 0x7febb2fbcca0> for event 'autodoc-process-signature' threw an exception
WARNING: error while formatting signature for seg1d.algorithm.rolling_corr: Handler <function process_numba_docstring at 0x7febb2fbcca0> for event 'autodoc-process-signature' threw an exception
WARNING: error while formatting signature for seg1d.algorithm.uniques: Handler <function process_numba_docstring at 0x7febb2fbcca0> for event 'autodoc-process-signature' threw an exception
WARNING: error while formatting signature for seg1d.optimized_funcs.rcor: Handler <function process_numba_docstring at 0x7febb2fbcca0> for event 'autodoc-process-signature' threw an exception
WARNING: error while formatting signature for seg1d.optimized_funcs.vcor: Handler <function process_numba_docstring at 0x7febb2fbcca0> for event 'autodoc-process-signature' threw an exception

Extension error:
Handler <function process_numba_docstring at 0x7febb2fbcca0> for event 'autodoc-process-signature' threw an exception

[Joss Review] Comparison to tslearn/seglearn

Even though I havn't checked out all functionality of the seg1d, a certain overlap in features with rather popular libraries like tslearn adn seglearn is undeniable. tslearn, seglearn and potentially other packages should be mentioned as related work in the paper.

Further it would great if the standout features of seg1d in comparison with other packages could be highlighted. When and why should I use seg1d over tslearn, sklearn, or seglearn directly?

For example, tslearn implements cross correlation based clustering and various method for time series segmentation.

[Joss Review] Update gitignore

The repository contains a couple of file that usually should not be tracked by git.

This includes:

• ./dist folder
• *.egg file
• ./docs/build

These files should be removed and added to the gitignore file.

More information on how to do this can be found here: https://www.git-tower.com/learn/git/faq/ignore-tracked-files-in-git

Resolve Documentation Location

Currently the website uses one of githubs default documentation locations, which is on master/docs. It is possible to put on its own branch, but this was more difficult to develop with. Need to look into it again.

[Joss Review] Author List

The paper contains multiple authors, but according to the git history only a single author contributed to the software project. I would be nice if the role and work of the other authors could be highlighted in the paper