This versions of FAµST brings several enhancements and complements to BSR matrices API, power iteration/2-norm algorithms and matfaust.Faust in single/float precision.
– Integration of BSR matrices in DYNPROG algorithm (beta).
– More information displayed on BSR matrices when a Faust is printed out.
– Faust.factors is able to return a BSR matrix too with pyfaust and a CSR sparse matrix copy with matfaust (Matlab doesn’t support BSR matrices).
– matfaust/pyfaust.rand_bsr is capable to generate complex random BSR Fausts with nonzero imaginary part.
– Random vector initialization in power iteration algorithm, and workaround of the float accuracy issues when used on a Faust.
– Also securing the spectral norm/2-norm computation of a Faust (a null Faust has zero norm, no need to go further in calculations).
– matfaust.Faust.single allows to convert a double Faust to a single/float Faust (see also this FAQ entry).