Hi,
I tried to enter GSoC 2013 with that project but I dropped it because I had other activities. As far as I got it, the idea is that the current "eig" function in Octave does not perform by default the balancing process which improves the eigenvalue calculation, while ML does and also offers other options such as choosing whether to use Cholesky factorization or the QZ algorithm.
What I found in the Octave code is that the eigenvalue calculation is done by calling a FORTRAN routine, but I also found that ML calls several other routines (ZGEEV, ZGEEV,...) to perform the calculations with the cases considered by the parameters. I haven't followed a lot this community in the last months and I don't know how much a ML-compliant eig function is desired (eigs also?), but it would be nice to have that compatibility in these functions since they are very widely used.
So, the project would consist in include the "eig" function the capability to evaluate the input arguments, perform the corresponding FORTRAN subroutines and compare with results given by ML's "eig" function.
I don't know what is required or if I'm eligible to be Arpit's mentor, but I'd like to see this function in future versions and I am willful to help.
Best regards