Pyinterpolate

The major release of **pyinterpolate**!

It brings many changes:
- First, from now on, we will name releases after the most prominent cities of Ukraine. [The first, and the biggest, is Kyiv, the Ukrainian capital](https://en.wikipedia.org/wiki/Kyiv)
- Calculations are faster, much faster than in release 0.2.5,
- API is enhanced, there are new classes and functions for semivariogram modeling and analysis, and for the complex data structures that are storing blocks and point support for Poisson Kriging and areal deconvolution,
- API is cleaner,
- paths are shorter,
- and the package works with any Python version from 3.7 up!

Generally speaking, it is the best surprise for the end of Summer.

If you are a newbie, I recommend going to tutorials to check what you can do. If you are a seasoned user, I recommend digging into the API (and tutorials) - some functions and classes are renamed, some are removed, and there are new additions.

The new release with the major fix of bug related to the prediction error variance calculation. It was wrong in the previous releases of the package (but it didn't affect predictions, just error variance terms).

Changes:

- neighbors selection (lags counting) has been changed,
- TheoreticalSemivariogram searches for optimal sill in a grid search algorithm,
- corrected error in Krige class; now calculation of error variance is correct,
- updated paper and package docs.

Predicted values have lower variance of predictions, and they are better because optimal sill is derived from the grid search. (In the previous releases, the sill was fixed and equal to data variance).

The new release with the major fix of bug related to the prediction error variance calculation. It was wrong in the previous releases of the package (but it didn't affect predictions, just error variance terms).

Changes:

* neighbors selection (lags counting) has been changed,
* `TheoreticalSemivariogram` searches for optimal sill in a grid search algorithm,
* corrected error in `Krige` class; now calculation of error variance is correct.

Now predicted values have a lower variance of predictions and they are better due to the fact, that optimal sill is derived from the grid search. (In the previous releases sill was fixed and equal to data variance).

In general, this update fixes bugs related to the Kriging interpolation. The algorithm has chosen n-closest neighbors without consideration about their range. Now it is fixed but `oridnary_kriging()` and `simple_kriging()` methods are invoked with three new parameters, one that controls the range of neighbors search, the other that controls the minimum number of neighbors if the algorithm doesn't detect any neighbor in the range and the last parameter that controls max number of neighbors (it is important from the computational time). Additionally, a few changes created by the new contributors during Hacktoberfest, are included in this release.

Changelog:

* `self.points_values` chenged to `self.points_array` in `TheoreticalSemivariogram` class,
* `NaN` values are tested and checked in `calc_semivariance_from_pt_cloud()` function,
* new semivariogram models included in the package: **cubic**, **circular**, **power**,
* corrected calculation of the closest neighbors for kriging interpolation,
* changed `prepare_kriging_data()` function,
* the new optional parameter `check_coordinates` (**bool**) of `calc_point_to_point_distance()` function to control the coordinates uniqueness tests. This test is very resource-consuming and should be avoided in a normal work and it should be performed before data injection into the modeling pipeline.
* the new `dev/profiling/` directory to test and profile parts of a code.

* more parameters to store (and access) in TheoreticalSemivariogram class,
* error weighting against the linear regression model (ax + b),
* global mean for Simple Kriging as a required parameter,
* tqdm progress bar to `RegularizedSemivariogram.transform()` and `interpolate_raster()` functions,
* refactored Semivariogram Regularization: ranges are controlled by algorithm, not an user,
* added pull request template,
* added issues templates,
* bug in spherical semivariogram model,
* experimental variogram as points (not a solid line),
* inverse distance weighting function: algorithm, tests, documentation and new tutorial,
* changed output names of regularized data (`ArealKriging.regularize_data`) from **estimated value** to **reg.est** and from **estimated prediction error** to **reg.err**,
* error related to the id column as a string removed,
* TheoreticalSemivariogram `params` attribute changed to `nugget`, `sill` and `range` attributes.

The first release of pyinterpolate package! This package originated from public health / urban spaces analysis and it's role is to provide access to Poisson Kriging and Kriging interpolation techniques for wider audience.

The most important features of a package:
- semivariogram deconvolution procedure: you can upsample your choropleth maps to the resolution of point support grid,
- Area to Point and Area to Area Poisson Kriging,
- Ordinary and Simple Kriging,
- Smooth surface interpolation,
- Semivariogram estimation and analysis,
- Variogram cloud calculations and outliers removal.

In future package will be updated with the new interpolation techniques and semivariogram analysis tools up.

Name of release **Arabia Terra** (https://en.wikipedia.org/wiki/Arabia_Terra) is taken from Martian region which lies in the north part of Mars.

UPDATES:

**2021-03-10**: Directional semivariograms, corrected broken dependency in README.md.
**2021-03-03**: New tutorials, variogram point cloud analysis, outliers detection and removal.
**2020-12-26**: Module *io* renamed to *io_ops* to avoid any conflicts with Python's core package *io*. Semivariance calculation functions do not require passing arrays or lists of lags by an user,
**2020-11-25**: Updated package paths and corrected typos in _docstrings_.