I tried to use ARPACK using some examples (here), but I could not even figure out how to input my matrix. From this question, It seems that the implementations in Python and Matlab are the only solution to avoid the complexity of ARPACK. Is there some (Fortran-based) way to do this, also avoiding to calculate all the eigenvalues/eigenvectors?
I wrote a wrapper which needs only to set the matrix A and the number of eigenvalues (e.g. nev = 1 if you need only the first eigenvalue). The code is adapted from the specific application which can be found here. My module-like version needs only to set the matrix A, the number of features, the number of eigenvalues and which of them should be taken ('LM' - Largest Magnitude'.
The code is available as a Gist here.
NB: ARPACK is different from LAPACK because it is usually optimized for dense matrices while this is optimized for large sparse matrices.
ARPACK package is designed to compute a few eigenvalues and corresponding eigenvectors of large sparse or structured matrices, using the Implicitly Restarted Arnoldi Method (IRAM) or, in the case of symmetric matrices, the corresponding variant of the Lanczos algorithm. It is used by many popular numerical computing environments such as SciPy, Mathematica, GNU Octave and MATLAB to provide this functionality. (source: Wikipedia)