bigint_kernel_1:
This implementation is done using an array of unsigned shorts. It is also reference counted. For further details see the above link. Also note that kernel_2 should be faster in almost every case so you should really just use that version of the bigint object.
kernel_1ais a typedef for bigint_kernel_1 kernel_1a_cis a typedef for kernel_1a that checks its preconditions.
bigint_kernel_2:
This implementation is basically the same as kernel_1 except it uses the Fast Fourier Transform to perform multiplications much faster.
kernel_2ais a typedef for bigint_kernel_2 kernel_2a_cis a typedef for kernel_2a that checks its preconditions.
Random Maximum Margin Hashing by Alexis Joly and Olivier BuissonIn particular, we use a linear support vector machine to generate planes. We train it on randomly selected and randomly labeled points from the data to be hashed.
As an aside, the hash() for matrix objects is defined here. It has the same interface as all the others.
Gander, W. and W. Gautshi, "Adaptive Quadrature -- Revisited" BIT, Vol. 40, (2000), pp.84-101
An Introduction to the Kalman Filter by Greg Welch and Gary Bishop