spNetwork 0.4.3.2

other changes

The package searchTrees is not on CRAN anymore. It has been replaced in spNetwork by home made c++ classes using the BH library.

example datasets are now stored in .rda file and can be loaded with the function data

It is possible now to choose between two strategies to deal with density at 0 when calculating LOO likelihood for several bandwidths. For more details, see the parameter zero_strat in the functions bw_cvl_calc, bw_cvl_calc.mc, bw_cv_likelihood_calc, bw_cv_likelihood_calc.mc, bws_tnkde_cv_likelihood_calc and bws_tnkde_cv_likelihood_calc.mc.
It is possible now to set directly the size of local bandwidths for the functions nkde, nkde.mc, tnkde and tnkde.mc. This feature is currently experimental, please submit an issue if you encounter a bug.
One can now use the identity matrix type for neighbour weighting (network_listw). It is useful if one wants to apply different function to calculate the weights and to compare the results.
Adaptive bandwidth for TNKDE can now be calculated separately or simultaneously. For more details, see the parameter adaptive_separate.

The vignette Details about NKDE” has been removed from the package because it could not be knitted on some CRAN machines. It is still accessible on the website.
A new vignette K-nearest neighbour adaptive bandwidth has also be added to the website. It presents how spNetwork can use user defined local bandwidths.

The new c++ function for adding vertex on lines was buggy and could break the geometry of the original lines. This has been fixed.
On Linux with gcc-12, the package couldn’t be compiled. This has been fixed.

the package has been reworked to use sf rather than rgeos, maptools and sp. All the tests passed and the documentation has been modified.

Added a set of functions to perform Temporal Network Kernel Density Estimation (see vignette TNKDE)

Performance was improved for data-driven bandwidth selection for NKDE
The function lines_points_along has been reworked with c++ and is now way faster
The function lines_center has been reworked with c++ and is now way faster

An error was raised when points were sharing the exact same locations in the function network_knn. It has been corrected an a warning was added to clarify the situation.
A clearer error message is now raised when multiple geometries are passed to the nkde and nkde.mc functions.
When using the discontinuous NKDE, NA values were returned if an event was located exactly at the extremity of an edge with no connexions. This has been corrected.

Minor release to fix a bug on SOLARIS for CRAN

Added two functions for data-driven bandwidth selection (see vignette NKDE)
Added a function to simplify network before processing (simplify_network), reducing network complexity by removing useless edges or merge continuous edges can improve performance. This function is still experimental and must be used only for undirected networks.
Added two functions to calculate k nearest neighbours: network_knn and network_knn.mc
Added a function to calculate isochrones on a geographical network: calc_isochrones and a Vignette (Calculating isochrones) to give an example.
Making some internal functions visible for the user to allow graph analysis on networks, see the new vignette Building networks.

Several vignettes were added to present the new features. A pkgdown website was also built to give a more attractive way to browse the documentation.

We follow now the code coverage. We want to reach 70% coverage before the next CRAN release.

issue 1 fixed by changing the function maptools::snapPointsToLines for a home-made function. The package BH including the c++ boost library is used to realize fast spatial query.
issue 2 fixed by handling the special case of lines of length 0.
an error was raised when using listw_network with polygons and the method = “pointsalong”. This was caused by a minor error in the code and works now as expected.
an error was present in the Diggle correction factor, leading to overestimation of the correction. This error has been corrected.
for the k functions, a minor error in the calculus was corrected, affecting only the scale of the result. These functions are now formally tested.
In the nkde function, when using an integer matrix, an error was crashing the R session in some case. This has been corrected.

Performance was improved for functions listw_network, listw_network.mc, kfunctions, kfunctions.mc, cross_kfunctions and cross_kfunctions.mc, by reducing the complexity of the network with no loss of accuracy.
Several geometrical functions have been accelerated by c++ code using the boost library
upgraded c++ compilation requirement to c++17
the functions used for the nkde calculation were reworked. Sensible speed gain can be observed.