Statistical Learning with Kernel for Persistence Diagrams

Provides tools for exploiting topological information into standard statistical learning algorithms. To this aim, this package contains the most popular kernels defined on the space of persistence diagrams, and persistence images. Moreover, it provides a solver for kernel Support Vector Machines problems, whose kernels are not necessarily positive semidefinite, based on the C++ library 'LIBSVM' < https://www.csie.ntu.edu.tw/~cjlin/libsvm/>. Additionally, it allows to compute Wasserstein distance between persistence diagrams with an arbitrary ground metric, building an R interface for the C++ library 'HERA' < https://bitbucket.org/grey_narn/hera/src/master/>.

kernelTDA

This R-package provides an implementation of the most famous kernels found in the framework of Topological Data Analysis (TDA), more specifically:

• Persistence Scale Space Kernel
• Sliced Wasserstein Kernel
• Persistence Fisher Kernel
• Geodesic Wasserstein Kernel(s)
• Persistence Images

Here you can also find an R interface to the C++ library HERA, which contains an efficient implementation of the L_p q-Wasserstein distance between persistence diagrams.

Finally, this package contains a solver for kernelized Support Vector Machine problems with indefinite kernels, based on the algorithm proposed by Loosli et al.. The implementation is largely based on the C++ library LIBSVM, and on its R interface in the package e1071.

This package is now on CRAN, you can install it with:

install.packages("kernelTDA")