Skip to contents

Write a taupatch run config

Source: R/config.R
generate_config.Rd

Defaults reproduce a working ECOMON run over the Northeast US shelf, so a new config is normally one call with a name. The default bounding box matches the extent the original pipeline cropped its projections to (original/load_covars.R:147).

Usage

generate_config(
  name,
  zoop_file = "data/zooplankton_database.csv",
  output_dir = file.path("output", name),
  species = default_species_catalog(),
  active_species = "cfin",
  years = c(2003, 2017),
  months = c(1, 12),
  bbox = list(xmin = -76, xmax = -65, ymin = 35, ymax = 45),
  covariate_source = "copernicus",
  selected = c("SST", "SSS", "BOTT", "MLD", "CHL"),
  bathymetry = c("DEPTH", "SLOPE"),
  prejoin = list(),
  climate = character(),
  derivoce = list(),
  transform = list(log1p = c("CHL", "DEPTH")),
  normalize = TRUE,
  type = "rf",
  trees = 500,
  cv_folds = 10,
  dir = "inst/configs"
)

Arguments

name

config name; the file is written to <dir>/<name>.yaml

zoop_file

path to the zooplankton database CSV

output_dir

where run outputs are written

species

named list of species definitions; defaults to the three taxa in the original database (cfin, ctyp, pseudo) at the 90th percentile

active_species

which species to model

years

two-element c(start, end) year range

months

two-element c(start, end) month range

bbox

named list with xmin/xmax/ymin/ymax

covariate_source

one of "copernicus", "local_netcdf", "mock"

selected

time-varying covariate names; see copernicus_covariates()

bathymetry

static seafloor covariate names; see bathymetry_covariates()

prejoin

list of per-covariate steps applied before products are joined; see prejoin_steps()

climate

climate index names; see climate_index_covariates()

derivoce

list of derived-covariate steps; see derivoce_covariates()

transform

named list of transform to covariate names; see covariate_transforms()

normalize

whether to center and scale the predictors

type

which model to fit; see model_types()

trees

number of trees in the random forest

cv_folds

number of cross-validation folds

dir

directory to write the config into

Value

the path written, invisibly

Details

Covariates are named the way the rest of the package names them — SST, not thetao on cmems_mod_glo_phy_my_0.083deg_P1M-m — so a generated config reads like the shipped example and can be edited without consulting the Copernicus catalog. covariate_info() lists the names; the raw covariates.copernicus block is still accepted by hand for a dataset the catalog does not cover.

The file is loaded back and validated before the path is returned, so a mistyped covariate or transform is an error now rather than five minutes into a run.

See also

covariate_info() for the covariate names, load_config() to read the result back

Examples

if (FALSE) { # \dontrun{
generate_config("cfin_gom")
generate_config("ctyp_shelf", active_species = "ctyp",
                selected = c("SST", "SSS", "CHL"),
                transform = list(fourth_root = "CHL"))
} # }