Solves the constrained lasso problem using the CLASSO module in Python. The optimization problem is
sparse_log_contrast( Z, y, C = NULL, fraclist = NULL, nlam = 20, min_frac = 1e-04 )
| Z | n by p matrix containing log(X) |
|---|---|
| y | n vector (response) |
| C | m by p matrix. Default is a row vector of ones. |
| fraclist | (optional) vector of tuning parameter multipliers. Should be in (0, 1]. |
| nlam | number of tuning parameters (ignored if fraclist non-NULL) |
| min_frac | smallest value of tuning parameter multiplier (ignored if fraclist non-NULL) |
minimize_beta, beta0 1/(2n) || y - beta0 1_n - Zagg_clr beta ||^2 + lamda_max * frac || beta ||_1 subject to C beta = 0
Default is C = 1_p^T, but C can be a general matrix.
Observe that the tuning parameter is specified through "frac", the fraction of lamda_max (which is the smallest value for which beta is nonzero).