Fit a network connection matrix using smoothness regularized least-squares regression. This solves the regularized regression problem:
min_W>0 ||W*X-Y||^2 + lambda*(ninj/nx)*||W*Lx + Ly*W||^2
or, if Omega is given, solve:
min_W>0 ||P(W*X-Y)||^2 + lambda*(ninj/nx)*||W*Lx + Ly*W||^2,
where P=P_Omega^c is the projection onto the complement of Omega.
The code implements checkpointing and can making an initial guess by solving the unregularized problem.
Requirements:
- HDF5
- Armadillo C++ linear algebra library linked to HDF5 (http://arma.sourceforge.net)
- mmio.c and mmio.h (http://math.nist.gov/MatrixMarket/mmio-c.html)
- L-BFGS-B-C library (https://github.com/kharris/L-BFGS-B-C) You will have to edit the Makefile to point to these libraries.
Optional requirements:
- optimized BLAS (MKL, OpenBLAS, etc.). This will require editing the Makefile
to point to your installed libraries. An example with this structure is in
Makefile.hyak.
Building: make all
To run a test case, try running test. The final lines should be:
At iterate 160, f(x)= 4.85e+01, ||proj grad||_infty = 1.26e-02
At iterate 161, f(x)= 4.85e+01, ||proj grad||_infty = 9.73e-03
* * *
Tit = total number of iterations
Tnf = total number of function evaluations
Tnint = total number of segments explored during Cauchy searches
Skip = number of BFGS updates skipped
Nact = number of active bounds at final generalized Cauchy point
Projg = norm of the final projected gradient
F = final function value
* * *
N Tit Tnf Tnint Skip Nact Projg F
120000 161 165 154333 0 107113 9.73e-03 4.84880e+01
F(x) = 4.848795812e+01
22
Cauchy time 1.322e+00 seconds.
Subspace minimization time 2.035e+00 seconds.
Line search time 5.812e+00 seconds.
Total User time 1.045e+01 seconds.
CONVERGED!