Skip to contents

Compare interpolation settings (size & build time) for a model

Source: R/model_size.R
compare_interp_settings.Rd

Interpolates mod in memory under each of several setting combinations and tabulates how big the resulting model is: total value-parameter rows, the parameter / map / set counts, the variable & constraint estimate (from model_size()), the in-memory size, and the build time. A quick way to see how fold / sparse / prune trade off model size and speed before committing to a solve.

Usage

compare_interp_settings(
  mod,
  settings = NULL,
  ...,
  name = "cmp",
  verbose = FALSE,
  top_n = 12L,
  keep_scen = FALSE
)

Arguments

mod

a model.

settings

a named list of setting combinations; each element is a list of arguments forwarded to interpolate_model(), e.g. list(fold = TRUE, sparse = TRUE, prune = TRUE). When NULL a default grid (dense / sparse / sparse+prune / fold+sparse+prune / fold-all) is used.

...

arguments forwarded to EVERY interp_mod() call, e.g. horizon = newHorizon(period = 2024).

name

base scenario name; each build is "<name>_<setting>".

verbose

forwarded to interpolate_model() (per-build progress).

top_n

number of largest parameters to keep per build.

keep_scen

if TRUE, the interpolated scenarios are returned in $scen (handy to feed a later solve step). Off by default to save memory.

Value

an object of class interp_settings_cmp: a list with

  • summary — one row per setting (ok, seconds, mb, param_rows, counts, variable/constraint estimates, error),

  • top — wide data.frame of the largest parameters' row counts per setting,

  • details — named list of per-build model_size() objects,

  • scen — named list of scenarios when keep_scen = TRUE,

  • settings — the settings grid used.

Details

All builds use ondisk = FALSE (so sizes are measured in memory and are directly comparable) and overwrite = TRUE.

See also

model_size(), interpolate_model()

Examples

if (FALSE) { # \dontrun{
cmp <- compare_interp_settings(mod, horizon = newHorizon(period = 2024))
cmp                       # prints the comparison table
# custom grid:
compare_interp_settings(mod,
  settings = list(
    none    = list(fold = FALSE, sparse = FALSE, prune = FALSE),
    all     = list(fold = c("region","slice","year","comm","tech","stg","trade"),
                   sparse = TRUE, prune = TRUE)),
  horizon = newHorizon(period = 2024))
} # }