Changelog#

Version 1.1-rc#

Version 1.0.11#

[models] the models now use @numba.jit(nopython=True) as this will be the default behaviour starting with numba==0.59.0

Version 1.0.10#

[MetricSpace] now supports the Mahalanobis distance as well

Version 1.0.9#

[binning] fixed a deprecation warning raised by a future change to KMeans to preserve the current behavior
[data] Added a data generator for creating random multivariate data
[tests] Added tests for random multivariate data

Version 1.0.8#

[util] added cross_variograms for calculating cross-variograms for all combinations of N input variables. Variograms are returned in a 2D List (matrix) with all primary variograms on the diagonal.
[util] added support for DirectionalVariogram in case azimuth, tolerance or bandwidth are passed as keyword arguments to cross_variograms
[util] added tests for new cross_variograms function to be implemented.

Version 1.0.7#

[Variogram] The Variogram instance now removed all usages of the deprecated Variogram.residuals property
[tests] All warnings in the tests have been resolved, mostly by updating imports
[tests] Expected warnings in unittests are ignored now.

Version 1.0.6#

This is technically the same as the 1.0.5 as I screwed up and uploaded to PyPI without merging the changes.

Version 1.0.5#

[Variogram] The variogram now accepts 2D values in (n_samples, 2) shape. The second column will be interpreted as a co-variable co-located to the observations and the Variogram then calculates the cross-variogram for values[:, 0] ~ values[:, 1] Note that this only implements the basic calculation of cross-variograms. Utility functions and plotting still need to be updated. A tutorial will also be added in future.

Version 1.0.4#

[Variogram] added Variogram.pairwise_diffs property for accessing Variogram._diff
[Variogram] added Variogram.model_residuals as a replacement for Variogram.residuals
[Variogram] deprecated Variogram.residuals as it could be confused with Variogram.pairwise_diffs, which are also residuals

Version 1.0.3#

[Variogram] Variogram.pairwise_diffs wraps around old _diff and should be used instead of directly accessing _diff
[Variogram] Variogram.model_residuals will replace Variogram.residuals which has been deprecated

Version 1.0.1#

[Variogram] documentation added to use_nugget
[Variogram] Variogram.fit(method='manual') now implicitly sets use_nugget=True if a nugget is passed to fit.

Version 1.0.0#

[plotting] the 3D surface plot is now handling the opacity settings correctly.
[utils] the utils now include the likelihood submodule, which includes a get_likelihood function factory. The returned function can be minimized using SciPy to perform maximum likelihood fits.

Version 0.6.14#

[plotting] plotly contour plots are showing a colorbar label by default
[DirectionalVariogram] the constructor now sets an empty uncertainty array to prevent the class from throwing and error if no confidence interval is given.

Version 0.6.13#

[docs] the sphinx recipe knitting a TeX file from docs is now ignored on fail Reason is that the current build is too lage and any kind of buffer is overflowing
[docs] The jupyter notebook tutorials for the Docker image are now at root level.
[docs] The documentation tutorials are now sphinx-gallery builds of the notebook tutorial examples
[docs] all tutorials have been updated (spelling, code style etc.)
[docs] all tutorials now use the data submodule to be reproducible

Version 0.6.12#

[data] the dataset loader can now return pandas.DataFrame objects
[Dockerfile] some cleanups for making future tutorials work.

Version 0.6.11#

[Variogram] The kriging based leave-one-out cross validation is now omitting NaN.

Version 0.6.10#

[Variogram] The KMeans based binning function is now raising a value error if a ConvergenceWarning is found. The reason is, that the original settings for binning were not valid if KMeans did not converge and thus, the bins array might not be in a well defined state.

Version 0.6.9#

SciKit-GStat is now tested for Python 3.9. Finally. All unittest are now automatically run for Python 3.6, 3.7, 3.8 and 3.9

Version 0.6.8#

[RasterMetricSpace] small bugfix for Exceptions raised with small sample sizes

Version 0.6.7#

[RasterMetricSpace] a new class is introduced: RasterEquidistantMetricSpace. An instance can be passed as coordinates. It samples a given Raster image at concentric rings, to derive a more uniformly distributed distance matrix.

Version 0.6.6#

[Variogram] The automatic fitting of a theoretical variogram model is now optional. You can pass None as fit_method parameter, which will suppress the fitting.

Version 0.6.5#

[Variogram] now supports custom bin edges for the experimental variogram. Variogram.bins now accepts a list or array of upper bin edges.
[Variogram] has a new property called bin_count which returns the number of point pairs within each lag class

Version 0.6.4#

[Kriging] OrdinaryKriging.sigma <skgstat.OrdinaryKriging> is now initialized as a NaN-filled array.
[Kriging] OrdinaryKriging._estimator <skgstat.OrdinaryKriging> handles the error variance matrix index now correctly. On error during kriging, the index was not incremented, which lead to malformed error variance field output.

Version 0.6.3#

[interfaces] If any of the gstools interfaces are used, the Variogram will call fit without forcing a full preprocessing cycle. This fixes edge cases, where a parameter was mutated, but the fitting not performed before the instance was exported. This should only have happened in very rare occasions.
[data] added the meuse dataset from the R-package 'sp'

Version 0.6.2#

[Variogram] the fitting method is now implemented as Variogram.fit_method property. It will drop fitting parameters if the fit method is changed to something else than 'manual'.
[Variogram] If an invalid Variogram.fit_method is set, an AttributeError will instantly be raised. Beforehand it was only raised on the next call of fit

Version 0.6.1#

The Dockerfile was completely rewritten. A user can now specify the used Python version at build time of the docker image.
The Dockerfile is now part of the python package

Version 0.6.0#

The util and data submodule are now always loaded at top-level
fixed a potential circular import
added uncertainty tools to util. This is not yet finished and may change the signature before it gets stable with Version 1.0 or 1.1

Note

The current implementation of uncertainty propagation is not stable. It will be changed until version 0.7. The entry-point obs_sigma will stay stable and persist, but currently the uncertainty propagation will not be updated and invalidated as the Variogram instance changes.

Version 0.5.6#

[Variogram] the internal MetricSpace instance used to calculate the distance matrix is now available as the Variogram.metric_space property.
[Variogram] Variogram.metric_space is now read-only.
[unittest] two unittests are changed (linting, not functionality)

Version 0.5.5#

[data] new submodule data contains sample random fields and methods for sampling these fields in a reproducible way at random locations and different sample sizes.

Version 0.5.4#

[util] added a new cross_validation utility module to cross-validate variograms with leave-one-out Kriging cross validations.

Version 0.5.3#

[MetricSpace] new class ProbabilisticMetricSpace that extends the metric space by a stochastic element to draw samples from the input data, instead of using the full dataset.

Version 0.5.2#

[interface] new interface function added: to_gs_krige. This interface will return a gs.Krige instance from the fitted variogram.
some typos were corrected
some code refactored (mainly linting errors)

Version 0.5.1#

[plotting] the spatio-temporal 2D and 3D plots now label the axis correctly.
[plotting] fixed swapped plotting axes for spatio-temporal plots.

Version 0.5.0#

[MetricSpace] A new class MetricSpace was introduced. This class can be passed to any class that accepted coordinates so far. This wrapper can be used to pre-calculate large distance matrices and pass it to a lot of Variograms.
[MetricSpacePair] A new class MetricSpacePair was introduced. This is a pair of two MetricSpaces and pre-calculates all distances between the two spaces. This is i.e. used in Kriging to pre-calcualte all distance between the input coordinates and the interpolation grid only once.

Version 0.4.4#

[models] the changes to matern introduced in 0.3.2 are reversed. The Matérn model does not adapt the smoothness scaling to effective range anymore, as the behavior was too inconsistent.
[interface] minor bugfix of circular import in variogram_estimator interface
[models] matern(0, ...) now returns the nugget instead of numpy.NaN
[models] stable(0, ...) now returns the nugget instead of numpy.NaN or a ZeroDivisionError.

Version 0.4.3#

[Variogram] dim now returns the spatial dimensionality of the input data.
[Variogram] fixed a numpy depreaction warning in _calc_distances

Version 0.4.2#

[Variogram] bins now cases manual set bin edges automatically to a numpy.array().
[Variogram] get_empirical returns the empirical variogram. That is a tuple of the current bins and experimental arrays, with the option to move the bin to the lag classes centers.

Version 0.4.1#

[Variogram] moved the bin function setting into a wrapper instance method, which was an anonymous lambda before. This makes the Variogram serializable again.
[Variogram] a list of pylint errors were solved. Still enough left.

Version 0.4.0#

[binning] added 'stable_entropy' option that will optimize the lag class edges to be of comparable Shannon Entropy.

Version 0.3.11#

[Variogram] A new method is introduced to calculate fitting weights. Works for all but the manual fit method. By setting fit_sigma-'entropy', the fitting weights will be adjusted according to the lag classes’ Shannon entropy. That will ignore lag classes of high uncertainty and emphasize lags of low uncertainty.

Version 0.3.10#

[binning] added a median aggregation option to ward. This can be enabled by setting binning_agg_func to 'median'. The cluster centroids will be derived from the members median value, instead of mean value.
[Variogram] added fit_method-'ml' - a maximum likelihood fitting procedure to fit the theoretical variogram to the experimental
[Variogram] added fit_method-'manual'. This is a manual fitting method that takes the variogram parameters either at instantiation prefixed by fit_, or as keyword arguments by fit.
[Variogram] the manual fitting method will preserve the previous parameters, if the Variogram was fitted before and the fitting parameters are not manually overwritten.

Version 0.3.9#

[binning] added kmeans and ward for forming non-equidistant lag classes based on a distance matrix clustering
[Kriging] Kriging now stores the last interpolated field as z. This is the first of a few changes in future releases, which will ultimately add some plotting methods to Kriging.

Version 0.3.8#

[plotting] minor bugfixes in plotting routines (wrong arguments, pltting issues)
[docs] added a tutorial about plotting
[binning] added auto_derived_lags for a variety of different methods that find a good estimate for either the number of lag classes or the lag class width. These can be used by passing the method name as bin_func parameter: Freedman-Diaconis ('fd'), Sturge’s rule ('sturges'), Scott’s rule ('scott') and Doane’s extension to Sturge’s rule ('doane'). Uses histogram_bin_edges <numpy.histogram_bin_edges> internally.

Version 0.3.7#

[Variogram] now accepts arbitrary kwargs. These can be used to further specify functional behavior of the class. As of Version 0.3.7 this is used to pass arguments down to the entropy and percentile estimators.
[Variogram] the describe now adds the init arguments by default to the output. The method can output the init params as a nested dict inside the output or flatten the output dict.

Version 0.3.6#

Warning

There is some potential breaking behaviour

[Variogram] some internal code cleanup. Removed some unnecessary loops
[Variogram] setting the n_lags property now correctly forces a recalculation of the lag groupings. So far they were kept untouches, which might result in old experimental variogram values for the changed instance. This is a potential breaking change.
[Variogram] The lag_classes generator now yields empty arrays for unoccupied lag classes. This will result in NaN values for the semi-variance. This is actually a bug-fix. This is a potential breaking change

Version 0.3.5#

[plotting] The location_trend can now add trend model lines to the scatter plot for the 'plotly' backend and calculate the R² for the trend model.
[Variogram] the internal attribute holding the name of the current distance function was renamed from _dict_func to _dist_func_name

Version 0.3.4#

[plotting] The scattergram functions color the plotted points with respect to the lag bin they are originating from. For matplotlib, this coloring is suppressed, but can activated by passing the argument scattergram(single_color-False).

Version 0.3.3#

[plotting] a new submodule is introduced: skgstat.plotting. This contains all plotting functions. The plotting behavior is not changed, but using skgstat.plotting.backend(), the used plotting library can be switched from matplotlib to plotly
[stmodels] some code cleanup
[SpaceTimeVariogram] finally can fit the product-sum model to the experimental variogram

Version 0.3.2#

[models] Matérn model now adapts effective range to smoothness parameter
[models] Matérn model documentation updated
[models] some minor updates to references in the docs

Version 0.3.1#

[Variogram] - internal distance calculations were refactored, to speed things up
[Kriging] - internal distance calculations were refactored, to speed things up

Version 0.3.0#

[Variogram] some internal calculations were changed.
[DirectionalVariogram] - the circular search are is removed and raises a NotImplementedError
[DirectionalVariogram] - direction mask data is calculated way faster and without shapely involved.
shapely is not a dependency anymore
[unittests] - more unittests were added.

Version 0.2.8#

[Variogram] is now pickle.dump()-able, by removing lambda usage (thanks to @redhog!)
[Variogram] now raises a Warning if all input values are the same
[DOCS] Tutorial added and Dockerfile finalized
[Variogram] normalize default value changed to normalize-False
[Variogram] harmonize parameter is removed
[Variogram] Monotonization (old harmonize par) is available as a new theoretical model function. Can be used by setting model-'harmonize'
[interfaces] gstools interface implemented. gstools_cov_model takes a skgstat.Variogram instance and returns a fitted gstools.CovModel.

Version 0.2.7#

[Kriging] Little performance gains due to code cleanup.
[Variogram] The normalize-True default in __init__ will change to normalize-False in a future version. A DeprecationWarning was included.
[tests] The Variogram class fitting unit tests are now explicitly setting the normalize parameter to handle the future deprecation.
[tests] More unittests were added to increase coverage
[interfaces] The new submodule skgstat.interfaces is introduced. This submodule collects interfacing classes to use skgstat classes with other Python modules.
[interfaces] The first interfacing class is the VariogramEstimator. This is a scikit-learn compatible Estimator class that can wrap a Variogram. The intended usage is to find variogram hyper-parameters using GridSearchCV. This is also the only usecase covered in the unit tests.
[interfaces] Implemented pykrige_as_kwargs. Pass a Variogram object and a dict of parameters is returned that can be passed to pykrige Kriging classes using the double star operator.
Added Dockerfile. You can now build a docker container with scikit-gstat installed in a miniconda environment. On run, a jupyter server is exposed on Port 8888. In a future release, this server will serve tutorial notebooks.
[stmodels] small bugfix in product model
[stmodels] removed variogram wrapper and added stvariogram wrapper to correctly detect space and time lags

Version 0.2.6#

[OrdinaryKriging]: widely enhanced the class in terms of performance, code coverage and handling.
- added mode property: The class can derive exact solutions or estimate the kriging matrix for high performance gains
- multiprocessing is supported now
- the solver property can be used to choose from 3 different solver for the kriging matrix.
[OrdinaryKriging]: calculates the kriging variance along with the estimation itself. The Kriging variance can be accessed after a call to OrdinaryKriging.transform and can be accessed through the OrdinaryKriging.sigma attribute.
[Variogram] deprecated Variogram.compiled_model. Use Variogram.fitted_model instead.
[Variogram] added a new and much faster version of the parameterized model: Variogram.fitted_model
[Variogram] minor change in the cubic model. This made the adaption of the associated unit test necessary.

Version 0.2.5#

added OrdinaryKriging for using a Variogram to perform an interpolation.

Version 0.2.4#

added SpaceTimeVariogram for calculating dispersion functions depending on a space and a time lag.

Version 0.2.3#

[severe bug] A severe bug was in Variogram.__vdiff_indexer was found and fixed. The iterator was indexing the Variogram._diff array different from Variogram.distance. This lead to wrong semivariance values for all versions > 0.1.8!. Fixed now.
[Variogram] added unit tests for parameter setting
[Variogram] fixed fit_sigma setting of 'exp': changed the formula from e^{\left(\frac{1}{x}\right)} to 1. - e^{\left(\frac{1}{x}\right)} in order to increase with distance and, thus, give less weight to distant lag classes during fitting.

Version 0.2.2#

added DirectionalVariogram class for direction-dependent variograms
[Variogram] changed default values for estimator and model from function to string

Version 0.2.1#

added various unittests

Version 0.2.0#

completely rewritten Variogram class compared to v0.1.8