Disstans

This is a minor release including bugfixes in various places as well as changes to address deprecations in upstream packages.

**Full Changelog**: https://github.com/tobiscode/disstans/compare/v2.0...v2.0.1

This is the final release for the **massively-overhauled DISSTANS Version 2.**

The major change introduced is the **increased numerical stability for L0 solutions**. As a downside, **all `penalty`, `eps`, and `scale` parameters of `lasso_regression` and `ReweightingFunction` will have to be updated**. This is because of a change in the cost function that is optimized. A good initial starting point is to increase `penalty` by a factor of 10, decrease `eps` by one order of magnitude, and compensate `scale` by increasing it by one order of magnitude. All tutorials and examples have been updated to reflect this change and can be seen as a helping hand in the transition. **Specifying reweighting functions has become mandatory now**, and the spatial/local keyword names have been changed for increased clarity.

Other major changes (continued from the 1.2 Beta) include:

1. Python 3.11-style type annotations directly in the source code.
2. A Frequently Asked Questions section in the documentation.
3. `Network.gui()` can now take a latitude/longitude bounding box as input.
4. Multiple timescales for individual `Exponential` & `Logarithmic` models, including an option for (all of) the solvers to enforce sign constraints.
5. A `DecayingSplineSet` that is one-sided (becoming a child class of the new parent `BaseSplineSet` class).
6. Network availability plots for regular `Network` objects.
7. Added support for UNR's highrate timeseries file format `.kenv`.
8. The model functions creating the mapping columns, i.e., the model coefficients, are now explicitly documented in the `Model.get_mapping_single` methods.

Minimum dependencies have also been updated. As a result and the major changes above, **a complete reinstall of the environment and the package is recommended**. Compared to the `v2.0rc0` release candidate, only minor bugfixes and documentation improvements were added.

**Full Changelog**: https://github.com/tobiscode/disstans/compare/v1.1.2...v2.0

This is the first release candidate for the **massively-overhauled DISSTANS Version 2.**

The major change introduced is the **increased numerical stability for L0 solutions**. As a downside, **all `penalty`, `eps`, and `scale` parameters of `lasso_regression` and `ReweightingFunction` will have to be updated**. This is because of a change in the cost function that is optimized. A good initial starting point is to increase `penalty` by a factor of 10, decrease `eps` by one order of magnitude, and compensate `scale` by increasing it by one order of magnitude. All tutorials and examples have been updated to reflect this change and can be seen as a helping hand in the transition. **Specifying reweighting functions has become mandatory now**, and the spatial/local keyword names have been changed for increased clarity.

Other major changes (continued from the 1.2 Beta) include:

1. Python 3.11-style type annotations directly in the source code.
2. A Frequently Asked Questions section in the documentation.
3. `Network.gui()` can now take a latitude/longitude bounding box as input.
4. Multiple timescales for individual `Exponential` & `Logarithmic` models, including an option for (all of) the solvers to enforce sign constraints.
5. A `DecayingSplineSet` that is one-sided (becoming a child class of the new parent `BaseSplineSet` class).
6. Network availability plots for regular `Network` objects.
7. Added support for UNR's highrate timeseries file format `.kenv`.
8. The model functions creating the mapping columns, i.e., the model coefficients, are now explicitly documented in the `Model.get_mapping_single` methods.

Minimum dependencies have also been updated. As a result and the major changes above, **a complete reinstall of the environment and the package is recommended**. This version will remain in pre-release state for a while to see if there are any leftover bugs. Please continue to report them either through a GitHub issue or via email.

**Full Changelog**: <https://github.com/tobiscode/disstans/compare/v1.1.2...v2.0rc0>

This is a beta release of version 1.2 which includes a couple of notable changes, including:

- Multiple timescales for individual `Exponential` & `Logarithmic` models, including an option for the `lasso_regression` solver to enforce sign constraints. (Support for other solvers is planned.)
- A `DecayingSplineSet` that is one-sided (becoming a child class of the new parent `BaseSplineSet` class).
- Network availability plots for regular `Network` objects.
- Added support for UNR's highrate timeseries file format `.kenv`.

Most of these changes are barely tested, and will therefore be in beta stage for now.

**Full Changelog**: https://github.com/tobiscode/disstans/compare/v1.1.1...v1.2-beta

This is a bugfix release. It addresses some minor as well as not-so-minor issues, notably:

- It fixes 1 by pinning the pandas version to <1.5, something that will be reversed in release 1.2.
- The trends shown in the GUI and calculated by `Station.get_trend()` apparently used to throw just wrong solutions, which were traced back to the use of [scipy.sparse.linalg.lsqr](https://docs.scipy.org/doc/scipy/reference/generated/scipy.sparse.linalg.lsqr.html) (possibly related due to the fact that the code was only using dense data). The switch to [numpy.linalg.lstsq](https://numpy.org/doc/stable/reference/generated/numpy.linalg.lstsq.html) fixes this.
- Showing trends in the GUI also failed when some stations did not have the required Timeseries or Model, now those stations are just ignored.
- `UNRTimeseries` no longer fails when loading data with multiple reference longitudes, a sign for bad data.

**Full Changelog**: https://github.com/tobiscode/disstans/compare/v1.1.1...v1.1.2

The paper describing DISSTANS is now accepted and published by Computers & Geosciences: [Decomposition and Inference of Sources through Spatiotemporal Analysis of Network Signals: The DISSTANS Python package](https://doi.org/10.1016/j.cageo.2022.105247). This version adds the updated links, and fixes some documentation issues. Please consider this version as the final version for the published study.

**Full Changelog**: https://github.com/tobiscode/disstans/compare/v1.1...v1.1.1