solve_subproblem_penalized: Solve subproblems via coordinate descent with Stochastic...
In pengminshi/cFIT: cFIT (common Factor Space Integration & Transfer)

View source: R/data_integration_sketched.R

Solve the subproblem given which parameter set to update. For each subproblem, the exact solution is obtained. The sketched objective function is solve w.r.t the set of parameter with additional penalty in the form of

\frac{1}{μ_t} \|w-w^{}t-1\|

, where the step size

mu_t = 0.1\times iter

solve_subproblem_penalized(
  params.to.update = c("W", "lambda", "H"),
  X.list,
  W,
  H.list,
  b.list,
  lambda.list,
  iter,
  params.list.last,
  verbose = T
)

`params.to.update`	a characteristic scalar, choice of ('W','lambda','b','H'), specifying which set of parameters to update
`X.list`	a list of ncells-by-ngenes gene expression matrix
`W`	ngenes-by-r numeric matrix.
`H.list`	A list of factor loading matrix of size ncells-by-r
`b.list`	A list of shift vector of size p (ngenes).
`lambda.list`	A list of scaling vector of size p (ngenes).
`iter`	integer, the current iteration
`params.list.last`	The parameters from last iteration, for SPP
`verbose`	boolean scalar, whether to show extensive program logs (default TRUE)