QUFFKA — QUadrature-based Features For Kernel Approximation. It applies simple quadrature rules to generate random features for kernel approximation.
Clone and install:
$ cd quffka
$ pip install .
Three simple steps to kernel approximation:
To prepare datasets, run:
$ ./download_datasets.sh
$ cook-datasets
To run kernel approximation (beware long computations! First please see examples), run:
$ approximate-kernels
To compare how fast explicit mapping is computed, run:
$ measure-time