Skip to content

Commit

Permalink
add data back to fleet
Browse files Browse the repository at this point in the history
* add data error checks in information
* add get and set id functions to fleet interface and expose in rcpp_interface
* update demo and tests
* add distribution helper functions and tests
* add tests to fleet interface
* Fix formatting for tidyverse style.
* Increase some of the documentation to better explain parameters.
* Share argument documentation across functions.
* Use match.arg().
* use && and ||
* add documentation
* add examples
* add check validity function
* update error message to use cli formatting
* add new tests
* helper function for expected names
  • Loading branch information
Andrea-Havron-NOAA authored and kellijohnson-NOAA committed Oct 23, 2024
1 parent b8ad1c6 commit 5c2c721
Show file tree
Hide file tree
Showing 13 changed files with 912 additions and 108 deletions.
4 changes: 4 additions & 0 deletions NAMESPACE
Original file line number Diff line number Diff line change
Expand Up @@ -28,9 +28,13 @@ export(get_log)
export(get_random)
export(is.FIMSFit)
export(is.FIMSFits)
export(lognormal)
export(m_agecomp)
export(m_index)
export(m_landings)
export(multinomial)
export(new_data_distribution)
export(new_process_distribution)
export(run_gtest)
export(setup_and_run_gtest)
export(setup_gtest)
Expand Down
371 changes: 371 additions & 0 deletions R/distribution_formulas.R
Original file line number Diff line number Diff line change
@@ -0,0 +1,371 @@
#' Validity checks for distributions
#'
#' Check the validity of arguments passed to [new_data_distribution()] and
#' [new_process_distribution()].
#'
#' @param args A list of input arguments.
#' @noRd
#' @return
#' Nothing is returned if the check is successful, and error messages are
#' returned if the checks are unsuccessful.
check_distribution_validity <- function(args) {
family <- args[["family"]]
families <- args[["families"]]
sd <- args[["sd"]]
if (class(family) != "family") {
cli::cli_abort(c(
"x" = "{.code {family}} is the incorrect type.",
"i" = "{.var family} needs to be specified as a family class, e.g.,
`family = gaussian()`."
))
}
if (!(family[["family"]] %in% families)) {
fam_name <- family[["family"]]
cli::cli_abort(c(
"x" = "FIMS currently does not offer the family, {.code {fam_name}}, for
this distribution type.",
"i" = "The families available for this distribution type are: {.code
{families}}."
))
}
if (!is.null(args[["data_type"]])) {
data_type <- args[["data_type"]]
data_type_names <- c("index", "agecomp", "lengthcomp")

if (!(data_type %in% data_type_names)) {
cli::cli_abort(c(
"x" = "The data type is incorrect.",
"i" = "Correct values are: {.code {data_type_names}}."
))
}

if (grepl("comp", data_type) &&
(family[["family"]] == "lognormal" ||
family[["family"]] == "gaussian")) {
fam_name <- family[["family"]]
cli::cli_abort(c(
"x" = "The family, {.code {fam_name}} is not available for the data
type, {.code {data_type}}.",
"i" = "The available families for this data type are: `multinomial`."
))
}
if (data_type == "index" &&
family[["family"]] == "multinomial") {
fam_name <- family[["family"]]
cli::cli_abort(c(
"x" = "The family, {.code {fam_name}} is not available for the data
type, {.code {data_type}}.",
"i" = "The available families for this data type are: `lognormal` or
`gaussian`."
))
}
}

if (!all(sd[["value"]] > 0)) {
sd_value <- sd[["value"]]
cli::cli_abort(c(
"x" = "At least one value for sd, {.code {sd_value}} is out of bounds.",
"i" = "All standard deviation values need to be positive."
))
}

if (length(sd[["estimated"]]) > 1) {
if (length(sd[["value"]]) != length(sd[["estimated"]])) {
sd_length <- length(sd[["value"]])
est_length <- length(sd[["estimated"]])
cli::cli_abort(c(
"x" = "The length of `sd$value` is {.code {sd_length}} and the length
of `sd$estimated` is {.code {est_length}}.",
"i" = "The size of `value` and `estimated` must match."
))
}
}
}

#' Return name of expected value
#'
#' @inheritParams new_data_distribution
#' @noRd
#' @return
#' A string specifying the name of the expected value.
#'
get_expected_name <- function(family, data_type) {
expected_name <- dplyr::case_when(
data_type == "index" && grepl("lognormal|gaussian", family[["family"]]) &&
family[["link"]] == "log" ~ "log_expected_index",
data_type == "index" && grepl("lognormal|gaussian", family[["family"]]) &&
family[["link"]] == "identity" ~ "expected_index",
grepl("comp", data_type) ~ "proportion_catch_numbers_at_age"
)
# Check combination of entries was okay and led to valid name
if (is.na(expected_name)) {
cli::cli_abort(c(
"x" = "Error, expected_name of distribution is {.var {expected_name}}.",
"i" = "Some combination of the family and data_type are incompatible."
))
}
return(expected_name)
}

#' Set up a new distribution for a data type
#'
#' @param module An identifier to a C++ fleet module that is linked to the data
#' of interest.
#' @param family A description of the error distribution and link function to
#' be used in the model. The argument takes a family class, e.g.,
#' `stats::gaussian(link = "identity")`.
#' @param sd A list of length two. The first entry, `"value"`, (default = 1)
#' stores the initial values (scalar or vector) for the relevant standard
#' deviations. The second entry, `"estimated"` (default = FALSE) is a scalar
#' or vector of booleans indicating whether or not standard deviation is
#' estimated. If `"value"` is a vector and `"estimated"` is a scalar, the
#' `"estimated"` value will be applied to the entire vector, otherwise, the
#' dimensions of the two must match.
#' @param data_type A single string specifying the type of data that the
#' distribution will be fit to. Options are listed in the function call,
#' where the first option listed, i.e., `"index"` is the default.
#' @return
#' Reference Class. Use [show()] to view Rcpp class fields, methods, and
#' documentation.
#' @export
#' @examples
#' \dontrun{
#' nyears <- 30
#' # Create a new fleet module
#' fleet <- methods::new(Fleet)
#' # Create a distribution for the fleet module
#' fleet_distribution <- new_data_distribution(
#' module = fishing_fleet,
#' family = lognormal(link = "log"),
#' sd = list(value = rep(sqrt(log(0.01^2 + 1)), nyears),
#' estimated = rep(FALSE, nyears),
#' data_type = "index"
#' )
#' }
new_data_distribution <- function(
module,
family,
sd = list(value = 1, estimated = FALSE),
data_type = c("index", "agecomp", "lengthcomp")
) {
data_type <- match.arg(data_type)
families <- c("lognormal", "gaussian", "multinomial")

# validity check on user input
args <- list(
family = family,
sd = sd,
data_type = data_type,
families = families
)
check_distribution_validity(args)

# assign name of observed data based on data_type
obs_id_name <- glue::glue("observed_{data_type}_data_id")

# Set up distribution based on `family` argument`
if (family[["family"]] == "lognormal") {
# create new Rcpp module
new_module <- new(TMBDlnormDistribution)

# populate logged standard deviation parameter with log of input
new_module$log_logsd <- new(
ParameterVector,
log(sd$value),
length(sd$value)
)
# setup whether or not sd parameter is estimated
if (length(sd$value) > 1 && length(sd$estimated) == 1) {
new_module$log_logsd$set_all_estimable(sd$estimated)
} else {
for (i in 1:seq_along(sd$estimated)) {
new_module$log_logsd[i]$estimated <- sd$estimated[i]
}
}
}

if (family[["family"]] == "gaussian") {
# create new Rcpp module
new_module <- new(TMBDnormDistribution)

# populate logged standard deviation parameter with log of input
new_module$log_sd <- new(ParameterVector, log(sd$value), length(sd$value))

# setup whether or not sd parameter is estimated
if (length(sd$value) > 1 && length(sd$estimated) == 1) {
new_module$log_sd$set_all_estimable(sd$estimated)
} else {
for (i in 1:seq_along(sd$estimated)) {
new_module$log_sd[i]$estimated <- sd$estimated[i]
}
}
}

if (family[["family"]] == "multinomial") {
#create new Rcpp module
new_module <- new(TMBDmultinomDistribution)
}

# setup link to observed data
if (data_type == "index") {
new_module$set_observed_data(module$GetObservedIndexDataID())
}
if (data_type == "agecomp") {
new_module$set_observed_data(module$GetObservedAgeCompDataID())
}

# set name of expected values
expected <- get_expected_name(family, data_type)
# setup link to expected values
new_module$set_distribution_links("data", module$field(expected)$get_id())

return(new_module)
}

#' Sets up a new distribution for a process
#' @inheritParams new_data_distribution
#' @param par A string specifying the parameter name the distribution applies
#' to. Parameters must be members of the specified module. Use
#' `methods::show(module)` to obtain names of parameters within the module.
#' @param is_random_effect A boolean indicating whether or not the process is
#' estimated as a random effect.
#' @seealso
#' * [new_data_distribution()]
#' @export
#' @return
#' Reference Class. Use `show()` to view Rcpp class fields, methods, and
#' documentation.
#' @examples
#' \dontrun{
#' # Create a new fleet module
#' recruitment <- methods::new(BevertonHoltRecruitment)
#' # view parameter names of the recruitment module
#' methods::show(BevertonHoltRecruitment)
#' # Create a distribution for the recruitment module
#' recruitment_distribution <- new_process_distribution(
#' module = recruitment,
#' par = "log_devs",
#' family = gaussian(),
#' sd = list(value = 0.4, estimated = FALSE),
#' is_random_effect = FALSE
#' )
#' }
new_process_distribution <- function(module,
par,
family,
sd = list(value = 1, estimated = FALSE),
is_random_effect = FALSE) {
families <- c("lognormal", "gaussian")

# validity check on user input
args <- list(family = family, sd = sd, families = families)
check_distribution_validity(args)

# Set up distribution based on `family` argument`
if (family[["family"]] == "lognormal") {
# create new Rcpp module
new_module <- new(TMBDlnormDistribution)

# populate logged standard deviation parameter with log of input
new_module$log_logsd <- new(
ParameterVector,
log(sd$value),
length(sd$value)
)
#setup whether or not sd parameter is estimated
if (length(sd$value) > 1 && length(sd$estimated) == 1) {
new_module$log_logsd$set_all_estimable(sd$estimated)
} else {
for (i in 1:seq_along(sd$estimated)) {
new_module$log_logsd[i]$estimated <- sd$estimated[i]
}
}
}

if (family[["family"]] == "gaussian") {
# create new Rcpp module
new_module <- new(TMBDnormDistribution)

# populate logged standard deviation parameter with log of input
new_module$log_sd <- new(ParameterVector, log(sd$value), length(sd$value))

#setup whether or not sd parameter is estimated
if (length(sd$value) > 1 && length(sd$estimated) == 1) {
new_module$log_sd$set_all_estimable(sd$estimated)
} else {
for (i in 1:seq_along(sd$estimated)) {
new_module$log_sd[i]$estimated <- sd$estimated[i]
}
}
}

# indicate whether or not parameter is treated as a random effect in the model
module$field(par)$set_all_random(is_random_effect)

n_dim <- length(module$field(par))

# create new Rcpp modules
new_module$x <- new(ParameterVector, n_dim)
new_module$expected_values <- new(ParameterVector, n_dim)

# initialize values with 0
# these are overwritten in the code later by user input
for (i in 1:n_dim) {
new_module$x[i]$value <- 0
new_module$expected_values[i]$value <- 0
}

# setup links to parameter
new_module$set_distribution_links(
"random_effects",
module$field(par)$get_id()
)

return(new_module)
}

#' Distributions not available in {stats}
#'
#' Family objects provide a convenient way to specify the details of the models
#' used by functions such as [stats::glm()]. These functions within this
#' package are not available within {stats} but are designed in a similar
#' manner.
#'
#' @inheritParams lognormal
#' @return
#' An object of class `family` (which has a concise print method). This
#' particular family has a truncated length compared to other distributions in
#' [stats::family()].
#' \item{family}{character: the family name.}
#' \item{link}{character: the link name.}
#'
#' @seealso
#' * [stats::family()]
#' * [stats::gaussian()]
#' * [stats::glm()]
#' * [stats::power()]
#' * [stats::make.link()]
#' @export
#' @examples
#' a_family <- multinomial()
#' a_family[["family"]]
#' a_family[["link"]]
lognormal <- function(link = "log") {
family_class <- c(
list(family = "lognormal", link = link),
stats::make.link(link)
)
class(family_class) <- "family"
return(family_class)
}

#' @rdname lognormal
#' @export
multinomial <- function(link = "logit") {
family_class <- c(
list(family = "multinomial", link = link),
stats::make.link(link)
)
class(family_class) <- "family"
return(family_class)
}
Loading

0 comments on commit 5c2c721

Please sign in to comment.