R/biglasso-dispatch.R

In biglasso: Extending Lasso Model Fitting to Big Data

# Dispatch table for biglasso()'s (family, screen) -> C routine mapping.
#
# Each leaf lists:
#   - routine: the .Call() routine name (see src/init.c for registration)
#   - args:    the exact, ordered argument tokens the routine expects. Token
#              values are supplied by biglasso() via a single named list
#              (built once per call, see `dispatch_values` in biglasso())
#              covering every token used anywhere in this table; tokens that
#              don't apply to a given family (e.g. 'd'/'d_idx' for
#              family != "cox") are simply never selected out of that list.
#   - out:     the names to assign, in order, to the routine's returned
#              list. This is what `res[[k]]` used to spell out at each call
#              site; it's now colocated with the routine name so the two can
#              be audited together instead of hunting for a matching
#              `res[[k]]` block elsewhere in biglasso.R.
#
# Every `screen` value that reaches biglasso_dispatch_lookup() is guaranteed
# to have a real entry here: biglasso() derives its match.arg() choices
# directly from names(biglasso_dispatch_table[[family]]) (excluding the "MM"
# pseudo-key described below), so an unrecognized combination is rejected up
# front rather than silently substituted.
#
# alg.logistic == "MM" is keyed as the pseudo-screen "MM" (binomial only):
# it dispatches to a different routine than any actual screen value, and
# biglasso() selects this key directly rather than via `screen`.

biglasso_dispatch_table <- list(
  gaussian = list(
    Adaptive = list(
      routine = "cdfit_gaussian_ada_edpp_ssr",
      args = c(
        "X",
        "yy",
        "row_idx",
        "lambda",
        "nlambda",
        "lambda_log_scale",
        "lambda_min",
        "alpha",
        "user_lambda",
        "eps",
        "max_iter",
        "penalty_factor",
        "dfmax",
        "ncores",
        "update_thresh",
        "verbose"
      ),
      out = c(
        "b",
        "center",
        "scale",
        "lambda",
        "loss",
        "iter",
        "rejections",
        "safe_rejections",
        "col.idx"
      )
    ),
    SSR = list(
      routine = "cdfit_gaussian_ssr",
      args = c(
        "X",
        "yy",
        "row_idx",
        "lambda",
        "nlambda",
        "lambda_log_scale",
        "lambda_min",
        "alpha",
        "user_lambda",
        "eps",
        "max_iter",
        "penalty_factor",
        "dfmax",
        "ncores",
        "verbose"
      ),
      out = c("b", "center", "scale", "lambda", "loss", "iter", "rejections", "col.idx")
    ),
    Hybrid = list(
      routine = "cdfit_gaussian_bedpp_ssr",
      args = c(
        "X",
        "yy",
        "row_idx",
        "lambda",
        "nlambda",
        "lambda_log_scale",
        "lambda_min",
        "alpha",
        "user_lambda",
        "eps",
        "max_iter",
        "penalty_factor",
        "dfmax",
        "ncores",
        "safe_thresh",
        "verbose"
      ),
      out = c(
        "b",
        "center",
        "scale",
        "lambda",
        "loss",
        "iter",
        "rejections",
        "safe_rejections",
        "col.idx"
      )
    )
  ),
  binomial = list(
    MM = list(
      routine = "cdfit_binomial_ssr_approx",
      args = c(
        "X",
        "yy",
        "row_idx",
        "lambda",
        "nlambda",
        "lambda_min",
        "alpha",
        "user_lambda",
        "eps",
        "max_iter",
        "penalty_factor",
        "dfmax",
        "ncores",
        "warn",
        "verbose"
      ),
      out = c("a", "b", "center", "scale", "lambda", "loss", "iter", "rejections", "col.idx")
    ),
    Hybrid = list(
      routine = "cdfit_binomial_slores_ssr",
      args = c(
        "X",
        "yy",
        "n_pos",
        "ylab",
        "row_idx",
        "lambda",
        "nlambda",
        "lambda_log_scale",
        "lambda_min",
        "alpha",
        "user_lambda",
        "eps",
        "max_iter",
        "penalty_factor",
        "dfmax",
        "ncores",
        "warn",
        "safe_thresh",
        "verbose"
      ),
      out = c(
        "a",
        "b",
        "center",
        "scale",
        "lambda",
        "loss",
        "iter",
        "rejections",
        "safe_rejections",
        "col.idx"
      )
    ),
    Adaptive = list(
      routine = "cdfit_binomial_ada_slores_ssr",
      args = c(
        "X",
        "yy",
        "n_pos",
        "ylab",
        "row_idx",
        "lambda",
        "nlambda",
        "lambda_log_scale",
        "lambda_min",
        "alpha",
        "user_lambda",
        "eps",
        "max_iter",
        "penalty_factor",
        "dfmax",
        "ncores",
        "warn",
        "safe_thresh",
        "update_thresh",
        "verbose"
      ),
      out = c(
        "a",
        "b",
        "center",
        "scale",
        "lambda",
        "loss",
        "iter",
        "rejections",
        "safe_rejections",
        "col.idx"
      )
    ),
    SSR = list(
      routine = "cdfit_binomial_ssr",
      args = c(
        "X",
        "yy",
        "row_idx",
        "lambda",
        "nlambda",
        "lambda_log_scale",
        "lambda_min",
        "alpha",
        "user_lambda",
        "eps",
        "max_iter",
        "penalty_factor",
        "dfmax",
        "ncores",
        "warn",
        "verbose"
      ),
      out = c("a", "b", "center", "scale", "lambda", "loss", "iter", "rejections", "col.idx")
    )
  ),
  cox = list(
    SSR = list(
      routine = "cdfit_cox_ssr",
      args = c(
        "X",
        "yy",
        "d",
        "d_idx",
        "row_idx_cox",
        "lambda",
        "nlambda",
        "lambda_log_scale",
        "lambda_min",
        "alpha",
        "user_lambda",
        "eps",
        "max_iter",
        "penalty_factor",
        "dfmax",
        "ncores",
        "warn",
        "verbose"
      ),
      out = c("b", "center", "scale", "lambda", "loss", "iter", "rejections", "col.idx")
    ),
    sscox = list(
      routine = "cdfit_cox_sscox",
      args = c(
        "X",
        "yy",
        "d",
        "d_idx",
        "row_idx_cox",
        "lambda",
        "nlambda",
        "lambda_log_scale",
        "lambda_min",
        "alpha",
        "user_lambda",
        "eps",
        "max_iter",
        "penalty_factor",
        "dfmax",
        "ncores",
        "warn",
        "safe_thresh",
        "verbose"
      ),
      out = c("b", "center", "scale", "lambda", "loss", "iter", "rejections", "col.idx")
    ),
    scox = list(
      routine = "cdfit_cox_scox",
      args = c(
        "X",
        "yy",
        "d",
        "d_idx",
        "row_idx_cox",
        "lambda",
        "nlambda",
        "lambda_log_scale",
        "lambda_min",
        "alpha",
        "user_lambda",
        "eps",
        "max_iter",
        "penalty_factor",
        "dfmax",
        "ncores",
        "warn",
        "safe_thresh",
        "verbose"
      ),
      out = c("b", "center", "scale", "lambda", "loss", "iter", "rejections", "col.idx")
    ),
    safe = list(
      routine = "cdfit_cox_safe",
      args = c(
        "X",
        "yy",
        "d",
        "d_idx",
        "row_idx_cox",
        "lambda",
        "nlambda",
        "lambda_log_scale",
        "lambda_min",
        "alpha",
        "user_lambda",
        "eps",
        "max_iter",
        "penalty_factor",
        "dfmax",
        "ncores",
        "warn",
        "safe_thresh",
        "verbose"
      ),
      out = c("b", "center", "scale", "lambda", "loss", "iter", "rejections", "col.idx")
    ),
    Adaptive = list(
      routine = "cdfit_cox_ada_scox",
      args = c(
        "X",
        "yy",
        "d",
        "d_idx",
        "row_idx_cox",
        "lambda",
        "nlambda",
        "lambda_log_scale",
        "lambda_min",
        "alpha",
        "user_lambda",
        "eps",
        "max_iter",
        "penalty_factor",
        "dfmax",
        "ncores",
        "warn",
        "safe_thresh",
        "update_thresh",
        "verbose"
      ),
      out = c(
        "b",
        "center",
        "scale",
        "lambda",
        "loss",
        "iter",
        "rejections",
        "safe_rejections",
        "col.idx"
      )
    ),
    None = list(
      routine = "cdfit_cox",
      args = c(
        "X",
        "yy",
        "d",
        "d_idx",
        "row_idx_cox",
        "lambda",
        "nlambda",
        "lambda_log_scale",
        "lambda_min",
        "alpha",
        "user_lambda",
        "eps",
        "max_iter",
        "penalty_factor",
        "dfmax",
        "ncores",
        "warn",
        "verbose"
      ),
      out = c("b", "center", "scale", "lambda", "loss", "iter", "rejections", "col.idx")
    )
  ),
  mgaussian = list(
    SSR = list(
      routine = "cdfit_mgaussian_ssr",
      args = c(
        "X",
        "yy",
        "row_idx",
        "lambda",
        "nlambda",
        "lambda_log_scale",
        "lambda_min",
        "alpha",
        "user_lambda",
        "eps",
        "max_iter",
        "penalty_factor",
        "dfmax",
        "ncores",
        "verbose"
      ),
      out = c("b", "center", "scale", "lambda", "loss", "iter", "rejections", "col.idx")
    ),
    Adaptive = list(
      routine = "cdfit_mgaussian_ada",
      args = c(
        "X",
        "yy",
        "row_idx",
        "lambda",
        "nlambda",
        "lambda_log_scale",
        "lambda_min",
        "alpha",
        "user_lambda",
        "eps",
        "max_iter",
        "penalty_factor",
        "dfmax",
        "ncores",
        "safe_thresh",
        "update_thresh",
        "verbose"
      ),
      out = c(
        "b",
        "center",
        "scale",
        "lambda",
        "loss",
        "iter",
        "rejections",
        "safe_rejections",
        "col.idx"
      )
    )
  )
)

# Looks up the dispatch spec for (family, screen). `screen_key` should
# already be "MM" for alg.logistic == "MM".
biglasso_dispatch_lookup <- function(family, screen_key) {
  spec <- biglasso_dispatch_table[[family]][[screen_key]]
  if (is.null(spec)) {
    stop("Invalid screening method!")
  }
  spec
}