v1.8.1

Released on: 2023/06/11

DOI: https://doi.org/10.5281/zenodo.10964877

Note: Verde v1.8.0 is the last release that is compatible with Python 3.7 and 3.8.

Breaking changes:

• Drop support for Python 3.7 and 3.8 (#418).

Bug fixes:

• Fix pandas warnings about .ravel() (#449)
• Fix dropping non-dimensional coordinates in grid_to_table (#441)

Maintenance:

• Remove dependence on pkg_resources (#448)
• Extend support for Python 3.12 (#442)
• Use Trusted Publishers for PyPI deployment (#436)
• Use Burocrata to check/add license notices (#435)
• Add Dependabot config to update GitHub Actions (#421)

Documentation:

• Replace Sphinx napoleon for numpydoc (#450)
• Refactor the Overview introductory tutorial (#429)
• Update the versions of documentation tools (#419)

This release contains contributions from:

• Leonardo Uieda
• Santiago Soler
• Gelson Ferreira Souza Junior

v1.8.0

Released on: 2023/05/08

DOI: https://doi.org/10.5281/zenodo.7907182

Note: Verde v1.7.0 is the last release that is compatible with Python 3.6.

New features:

• New interpolator verde.KNeighbors class for nearest neighbor interpolation (#378)
• New interpolator verde.Cubic gridder class based on SciPy (#374)
• New interpolator verde.Linear gridder class based on SciPy (#372)
• New function verde.line_coordinates, a 1D version of verde.grid_coordinates (#390)
• New scoring parameter for verde.SplineCV to specify the scoring function (#380)

Deprecations:

• Deprecate the engine argument of Spline/SplineCV (#373)
• Deprecate the engine argument in VectorSpline2D (#410)
• Deprecate verde.ScipyGridder in favor of the new Linear/Cubic/KNeighbors (#393)
• Deprecate the scatter method of all interpolators (#357)
• Undo deprecation of region/spacing/shape in the grid method (#394)
• Undo deprecation of verde.VectorSpline2D (#385)

Improvements:

• Remove the need for mindist in verde.Spline by using a better Green's function for small distances (#401)
• Fix behavior of coordinate generation in verde.line_coordinates if spacing >= 2 * interval (#406)
• Use the new classes Linear/Cubic/KNeighbors in verde.project_grid (#395)
• Default to not rescaling coordinates in Linear/Cubic (#391)
• Add option to return 1D arrays in grid_coordinates (#388)

Documentation:

• New logo and use sphinx-design in the docs (#367)
• Move deprecated APIs to their own docs section (#400)
• Improve docstring of verde.make_xarray_grid (#399)
• Fix typo in the verde.base.BaseGridder.fit docstring (#397)
• Add missing matplotlib scraper for sphinx-gallery (#389)
• Use PyGMT instead of Cartopy in the verde.Chain tutorial (#386)
• Use Markdown for the README instead of RST (#366)

Maintenance:

• Drop support for Python 3.6 (#364)
• Fail CI if codecov upload fails (#409)
• More informative warning messages by setting stacklevel=2 (#407)
• Remove the deprecated sample data gallery (#387)
• Set lower bounds for dependencies based on NEP29 (#384)
• Replace setup.py with PyPA "build" (#371)
• Replace deprecated numpy dtypes np.bool and np.int (#362)

This release contains contributions from:

• Sarah Margrethe Askevold
• James Sample
• Santiago Soler
• Matt Tankersley
• Leonardo Uieda

v1.7.0

Released on: 2022/03/25

DOI: https://doi.org/10.5281/zenodo.6384887

Deprecation:

• Move the CheckerBoard class to verde.synthetic (#353)
• Deprecate the verde.test function which will be removed in v2.0.0 (#344)
• Deprecate the datasets module, which will be replaced by Ensaio in the future (#277)
• Warn that the default score will change from R² to negative RMSE in v2.0 (#352)

New features:

• Add option to pass coordinates to the grid method instead of just region and spacing (#326)
• Add support for Python 3.9 (#323) and 3.10 (#346)

Documentation:

• Modernize the front page of the docs (#356)
• Modernize the Installing page (#355)
• Update the contact link in the docs (#347)
• Switch the docs theme to the sphinx-book-theme (#343)
• Update dims in example of make_xarray_grid (#329)
• Explicitly pass default arguments with their corresponding keywords on tests and examples (#327)

Maintenance:

• Replace Google Analytics for Plausible one to make our docs more privacy-friendly (#358)
• Move configuration from setup.py to setup.cfg (#348)
• Link CoC, Authorship, Contributing, and Maintainers guides back to the Fatiando-wide pages (#338)
• Replace pylint with more flake8 plugins (#337)
• Rename the main branch from "master" to "main" (#335)
• Remove normalize argument when creating scikit-learn solvers (#333)

This release contains contributions from:

• Santiago Soler
• Leonardo Uieda

v1.6.1

Released on: 2021/03/22

Minor changes:

• Allow make_xarray_grid to receive data=None instead of raising an error. This is used to create an empty xarray.Dataset (#318)

Maintenance:

• Fix use of wrong version numbers for PyPI releases (#317)

This release contains contributions from:

• Santiago Soler
• Leonardo Uieda

v1.6.0

Released on: 2021/03/18

New features:

• Allow specifing the scoring function in cross_val_score instead of always using the .score method of the gridder (#273)
• New function verde.make_xarray_grid to simplify the creation of xarray.Dataset from individual numpy arrays that represent a 2D grid (#282 and #300)

Enhancements:

• Raise informative errors for invalid verde.rolling_window arguments, like missing spacing or shape and invalid window sizes (#280)
• Replace DeprecationWarning with FutureWarning since these are intended for end-users, which allows us to avoid having to set warning.simplefilter (#305 and #293)

Documentation:

• Several typo fixes (#306 #303 #281)
• Update link to the GMT website in the Baja bathymetry example (#298)
• Fix issue with Cartopy 0.17 and require versions >= 0.18 for building the docs (#283)

Maintenance:

• Refactor internal function get_data_names and related check functions to simplify their logic and make them more useful (#295)
• Require Black >=20.8b1 (#284)
• Format the doc/conf.py sphinx configuration file with Black (#275)
• Add a license and copyright notice to every source file (#308)
• Replace versioneer for setuptools-scm (#307)
• Replace Travis and Azure with GitHub Actions (#309)
• Exclude Dask 2021.03.0 as a dependency. This release was causing the tests to fail under Python 3.8 on every OS. The problem seems to be originated in dask.distributed (#311)
• Use the OSI version of item 3 in the license (#299)

This release contains contributions from:

• Santiago Soler
• Leonardo Uieda
• Federico Esteban
• DC Slagel

v1.5.0

Released on: 2020/06/04

Bug fixes:

• Apply projections using only the first two coordinates instead all given coordinates. Projections only really involve the first two (horizontal) coordinates. Only affects users passing extra_coords to gridder methods. (#264)

New features:

• New blocked cross-validation classes BlockShuffleSplit and BlockKFold. These are scikit-learn compatible cross-validators that split the data into spatial blocks before assigning them to folds. Blocked cross-validation can help avoid overestimation of prediction accuracy for spatial data. The classes work with verde.cross_val_score and any other function/method/class that accepts a scikit-learn cross-validator. (#251 and #254)
• Add the option for block-wise splitting in verde.train_test_split by passing in a spacing or shape parameters. (#253 and #257)

Base classes:

• Add optional argument to verde.base.least_squares to copy Jacobian matrix. (#255)
• Add extra coordinates (specified by the extra_coords keyword argument to outputs of BaseGridder methods. (#265)

Maintenance:

• Update tests to repr changes in scikit-learn 0.23.0. (#267)

Documentation:

• Fix typo in README contributing section. (#258)

This release contains contributions from:

• Leonardo Uieda
• Santiago Soler
• Rowan Cockett

v1.4.0

DOI: https://doi.org/10.5281/zenodo.3739449

Bug fixes:

• Profile distances are now returned in projected (Cartesian) coordinates by the profile method of gridders if a projection is given. The method has the option to apply a projection to the coordinates before predicting so we can pass geographic coordinates to Cartesian gridders. In these cases, the distance along the profile is calculated by the profile_coordinates function with the unprojected coordinates (in the geographic case it would be degrees). The profile point calculation is also done assuming that coordinates are Cartesian, which is clearly wrong if inputs are longitude and latitude. To fix this, we now project the input points prior to passing them to profile_coordinates. This means that the distances are Cartesian and generation of profile points is also Cartesian (as is assumed by the function). The generated coordinates are projected back so that the user gets longitude and latitude but distances are still projected Cartesian meters. (#231)
• Function verde.grid_to_table now sets the correct order for coordinates. We were relying on the order of the coords attribute of the xarray.Dataset for the order of the coordinates. This is wrong because xarray takes the coordinate order from the dims attribute instead, which is what we should also have been doing. (#229)

Documentation:

• Generalize coordinate system specifications in verde.base.BaseGridder docstrings. Most methods don't really depend on the coordinate system so use a more generic language to allow derived classes to specify their coordinate systems without having to overload the base methods just to rewrite the docstrings. (#240)

New features:

• New function verde.convexhul_mask to mask points in a grid that fall outside the convex hull defined by data points. (#237)
• New function verde.project_grid that transforms 2D gridded data using a given projection. It re-samples the data using ScipyGridder (by default) and runs a blocked mean (optional) to avoid aliasing when the points aren't evenly distributed in the projected coordinates (like in polar projections). Finally, it applies a convexhul_mask to the grid to avoid extrapolation to points that had no original data. (#246)
• New function verde.expanding_window for selecting data that falls inside of an expanding window around a central point. (#238)
• New function verde.rolling_window for rolling window selections of irregularly sampled data. (#236)

Improvements:

• Allow verde.grid_to_table to take xarray.DataArray as input. (#235)

Maintenance:

• Use newer MacOS images on Azure Pipelines. (#234)

This release contains contributions from:

• Leonardo Uieda
• Santiago Soler
• Jesse Pisel

v1.3.0

DOI: https://doi.org/10.5281/zenodo.3620851

DEPRECATIONS (the following features are deprecated and will be removed in Verde v2.0.0):

• Functions and the associated sample dataset verde.datasets.fetch_rio_magnetic and verde.datasets.setup_rio_magnetic_map are deprecated. Please use another dataset instead. (#213)
• Class verde.VectorSpline2D is deprecated. The class is specific for GPS/GNSS data and doesn't fit the general-purpose nature of Verde. The implementation will be moved to the Erizo package instead. (#214)
• The client keyword argument for verde.cross_val_score and verde.SplineCV is deprecated in favor of the new delayed argument (see below). (#222)

New features:

• Use the dask.delayed interface for parallelism in cross-validation instead of the futures interface (dask.distributed.Client). It's easier and allows building the entire graph lazily before executing. To use the new feature, pass delayed=True to verde.cross_val_score and verde.SplineCV. The argument client in both of these is deprecated (see above). (#222)
• Expose the optimal spline in verde.SplineCV.spline_. This is the fitted verde.Spline object using the optimal parameters. (#219)
• New option drop_coords to allow verde.BlockReduce and verde.BlockMean to reduce extra elements in coordinates (basically, treat them as data). Default to True to maintain backwards compatibility. If False, will no longer drop coordinates after the second one but will apply the reduction in blocks to them as well. The reduced coordinates are returned in the same order in the coordinates. (#198)

Improvements:

• Use the default system cache location to store the sample data instead of ~/.verde/data. This is so users can more easily clean up unused files. Because this is system specific, function verde.datasets.locate was added to return the cache folder location. (#220)

Bug fixes:

• Correctly use parallel=True and numba.prange in the numba compiled functions. Using it on the Green's function was raising a warning because there is nothing to parallelize. (#221)

Maintenance:

• Add testing and support for Python 3.8. (#211)

Documentation:

• Fix a typo in the JOSS paper Bibtex entry. (#215)
• Wrap docstrings to 79 characters for better integration with Jupyter and IPython. These systems display docstrings using 80 character windows, causing our larger lines to wrap around and become almost illegible. (#212)
• Use napoleon instead of numpydoc to format docstrings. Results is slightly different layout in the website documentation. (#209)
• Update contact information to point to the Slack chat instead of Gitter. (#204)

This release contains contributions from:

• Santiago Soler
• Leonardo Uieda

v1.2.0

DOI: https://doi.org/10.5281/zenodo.3347076

Bug fixes:

• Return the correct coordinates when passing pixel_register=True and shape to verde.grid_coordinates. The returned coordinates had 1 too few elements in each dimension (and the wrong values). This is because we generate grid-line registered points first and then shift them to the center of the pixels and drop the last point. This only works when specifying spacing because it will generate the right amount of points. When shape is given, we need to first convert it to “grid-line” shape (with 1 extra point per dimension) before generating coordinates. (#183)
• Reset force coordinates when refitting splines. Previously, the splines set the force coordinates from the data coordinates only the first time fit was called. This means that when fitting on different data, the spline would still use the old coordinates leading to a poor prediction score. Now, the spline will use the coordinates of the current data passed to fit. This only affects cases where force_coords=None. It’s a slight change and only affects some of the scores for cross-validation. (#191)

New functions/classes:

• New class verde.SplineCV: a cross-validated version of Spline . that performs grid search cross-validation to automatically tune the parameters of a Spline. (#185)
• New function verde.longitude_continuity to format longitudes to a continuous range so that they can be indexed with verde.inside (#181)
• New function verde.load_surfer to load grid data from a Surfer ASCII file (a contouring, griding and surface mapping software from GoldenSoftware). (#169)
• New function verde.median_distance that calculates the median near neighbor distance between each point in the given dataset. (#163)

Improvements:

• Allow verde.block_split and verde.BlockReduce to take a shape argument instead of spacing. Useful when the size of the block is less meaningful than the number of blocks. (#184)
• Allow zero degree polynomials in verde.Trend, which represents a mean value. (#162)
• Function verde.cross_val_score returns a numpy array instead of a list for easier computations on the results. (#160)
• Function verde.maxabs now handles inputs with NaNs automatically. (#158)

Documentation:

• New tutorial to explain the intricacies of grid coordinates generation, adjusting spacing vs region, pixel registration, etc. (#192)

Maintenance:

• Drop support for Python 3.5. (#178)
• Add support for Python 3.7. (#150)
• More functions are now part of the base API: n_1d_arrays, check_fit_input and least_squares are now included in verde.base. (#156)

This release contains contributions from:

• Goto15
• Lindsey Heagy
• Jesse Pisel
• Santiago Soler
• Leonardo Uieda

v1.1.0

https://doi.org/10.5281/zenodo.1478245

New features:

• New verde.grid_to_table function that converts grids to xyz tables with the coordinate and data values for each grid point (#148)
• Add an extra_coords option to coordinate generators (grid_coordinates, scatter_points, and profile_coordinates) to specify a constant value to be used as an extra coordinate (#145)
• Allow gridders to pass extra keyword arguments (**kwargs) for the coordinate generator functions (#144)

Improvements:

• Don’t use the Jacobian matrix for predictions to avoid memory overloads. Use dedicated and numba wrapped functions instead. As a consequence, predictions are also a bit faster when numba is installed (#149)
• Set the default n_splits=5 when using KFold from scikit-learn (#143)

Bug fixes:

• Use the xarray grid’s pcolormesh method instead of matplotlib to plot grids in the examples. The xarray method takes care of shifting the pixels by half a spacing when grids are not pixel registered (#151)

New contributors to the project:

• Jesse Pisel