Python code for Graph Basis Functions approximation.
This code is a (partial) porting of the Matlab toolbox GBFlearn writtten by Wolfgang Erb.
The implementation structure and the code of the greedy algorithms is taken from the VKOGA package.
All the approximants are implemented as scikit-learn Estimator
, and thus they can be combined with the parameter
optimization tools of the package.
You can start with one of the demos:
- demo_interpolation.ipynb: An introduction to the basic computation of a GBF approximant.
- demo_error_decay.ipynb: A comparison of the decay of the approximation error for different kernels.
- demo_parameter_optimization.ipynb: An introduction to the use of
sklearn
to optimize the paramters. - demo_interpolation_fgreedy.ipynb: An introduction to the basic computation of a GBF approximant via f-greedy.
- demo_point_selection_Pgreedy.ipynb: An introduction to the point selection via P-greedy.
- demo_point_selection_Pgreedy_par_optimization.ipynb: Point selection via P-greedy with optimization of parameters.
The code is organized as follows:
- approx.py: Definition of the GBF approximation models (GBFIntepolation, GBFGreedy).
- kernels.py: Definition of the graph kernels and of the GBFs.
- utils.py: Definition of various utility functions.
- graph_loaders.py: Definition of various utility to load some example graphs.