R/utils.R

In strengejacke/ggeffects: Create Tidy Data Frames of Marginal Effects for 'ggplot' from Model Outputs

.get_model_function <- function(model) {
  # check class of fitted model
  lm_models <- c(
    "wblm", "wbm", "biglm", "speedlm", "gls", "ols", "ivreg", "gee", "plm", "lm",
    "rqss", "lmRob", "lm_robust", "lme", "truncreg", "nlmerMod", "glmgee",
    "lmerMod", "merModLmerTest", "rlmerMod", "bayesx", "mclogit"
  )

  info <- insight::model_info(model, verbose = FALSE)
  if (insight::is_multivariate(model) && !inherits(model, c("vglm", "vgam"))) {
    info <- info[[1]]
  }

  if (inherits(model, lm_models) && !inherits(model, "glm")) {
    "lm"
  } else if (inherits(model, "coxph")) {
    "coxph"
  } else if (inherits(model, "betareg")) {
    "betareg"
  } else if (isTRUE(info$is_linear)) {
    "lm"
  } else {
    "glm"
  }
}


.data_frame <- function(...) {
  x <- data.frame(..., stringsAsFactors = FALSE)
  rownames(x) <- NULL
  x
}


.check_vars <- function(terms, model) {
  if (missing(terms) || is.null(terms)) {
    insight::format_error("`terms` needs to be a character vector with at least one predictor name: one term used for the x-axis, more optional terms as grouping factors.") # nolint
  }

  # check for correct length of vector
  if (length(terms) > 5) {
    insight::format_alert("`terms` must have indicate more than five focal terms. Using first five variable names as focal terms now.")
    terms <- terms[1:5]
  }

  out_msg <- NULL
  msg <- tryCatch(
    {
      pv <- insight::find_predictors(
        model,
        effects = "all",
        component = "all",
        flatten = TRUE,
        verbose = FALSE
      )
      clean.terms <- .clean_terms(terms)
      if (!all(clean.terms %in% pv)) {
        out_msg <- c(
          "Some of the specified `terms` were not found in the model.",
          .misspelled_string(pv, clean.terms, "Maybe misspelled?")$msg
        )
      }
      out_msg
    },
    error = function(x) NULL
  )

  if (!is.null(out_msg)) {
    insight::format_error(out_msg)
  }

  terms
}


.offset_term <- function(model, condition = NULL, verbose = TRUE) {
  tryCatch(
    {
      off <- insight::safe_deparse(model$call$offset)
      if (identical(off, "NULL")) {
        return(NULL)
      }
      cleaned_off <- insight::clean_names(off)
      if (!identical(off, cleaned_off) && isTRUE(verbose) && !inherits(model, "glmmTMB") && !cleaned_off %in% names(condition)) { # nolint
        insight::format_alert(
          "Model uses a transformed offset term. Predictions may not be correct.",
          sprintf("It is recommended to fix the offset term using the `condition` argument, e.g. `condition = c(%s = 1)`.", cleaned_off), # nolint
          sprintf("You could also transform the offset variable before fitting the model and use `offset(%s)` in the model formula.", cleaned_off) # nolint
        )
      }
      cleaned_off
    },
    error = function(e) {
      NULL
    }
  )
}


.get_raw_data <- function(model, mf, terms) {
  # sanity check - could data be extracted from model frame?
  if (is.null(mf)) {
    mf <- .safe(insight::get_data(model, source = "environment", verbose = FALSE))
  }

  # for matrix variables, don't return raw data
  if (any(vapply(mf, is.matrix, TRUE)) && !inherits(model, c("coxph", "coxme"))) {
    return(NULL)
  }

  if (!all(insight::find_response(model, combine = FALSE) %in% colnames(mf))) {
    return(NULL)
  }

  # add rownames, for labelling data points in plots
  .safe({
    mf <- datawizard::rownames_as_column(mf)
  })

  # sanity check, make sure we have rownames as variable
  if (is.null(mf$rowname)) {
    mf$rowname <- as.character(seq_len(nrow(mf)))
  }

  # get response and x-value
  response <- insight::get_response(model, verbose = FALSE)

  # sanity check - has response correct length? May differ from model frame due to missings
  if (length(response) != nrow(mf)) {
    response <- .safe(mf[[insight::find_response(model)]])
  }

  # for cox-models, modify response
  if (inherits(model, "coxph")) {
    response <- response[[2]]
  }

  # character vectors to factors
  for (i in terms) {
    if (is.character(mf[[i]])) {
      mf[[i]] <- factor(mf[[i]], levels = unique(mf[[i]]))
    }
  }
  x <- .factor_to_numeric(mf[[terms[1]]])


  # add optional grouping variable
  if (length(terms) > 1) {
    group <- .as_label(
      mf[[terms[2]]],
      prefix = FALSE,
      drop.na = TRUE,
      drop.levels = !is.numeric(mf[[terms[2]]])
    )
  } else {
    group <- as.factor(1)
  }

  if (length(terms) > 2) {
    facet <- .as_label(
      mf[[terms[3]]],
      prefix = FALSE,
      drop.na = TRUE,
      drop.levels = !is.numeric(mf[[terms[3]]])
    )
  } else {
    facet <- as.factor(1)
  }

  if (length(terms) > 3) {
    panel <- .as_label(
      mf[[terms[4]]],
      prefix = FALSE,
      drop.na = TRUE,
      drop.levels = !is.numeric(mf[[terms[4]]])
    )
  } else {
    panel <- as.factor(1)
  }

  # return all as data.frame
  tryCatch(
    .data_frame(
      response = response,
      x = x,
      group = group,
      facet = facet,
      panel = panel,
      rowname = mf$rowname
    ),
    error = function(x) NULL,
    warning = function(x) NULL,
    finally = function(x) NULL
  )
}


.prettify_data <- function(conditional_terms, original_model_frame, terms,
                           use_all_values = FALSE, show_pretty_message = FALSE) {
  lapply(conditional_terms, function(.x) {
    pr <- original_model_frame[[terms[.x]]]
    if (is.numeric(pr)) {
      if (.x > 1 && .n_distinct(pr) >= 10) {
        values_at(pr)
      } else if (.n_distinct(pr) < 20 || isTRUE(use_all_values)) {
        sort(stats::na.omit(unique(pr)))
      } else {
        if (show_pretty_message) {
          insight::format_alert(sprintf(
            "Data were 'prettified'. Consider using `terms=\"%s [all]\"` to get smooth plots.", terms[.x]
          ))
          show_pretty_message <- FALSE
        }
        pretty_range(pr)
      }
    } else if (is.factor(pr)) {
      levels(droplevels(pr))
    } else {
      stats::na.omit(unique(pr))
    }
  })
}


.get_residual_variance <- function(x) {
  out <- .safe(insight::get_sigma(x, ci = NULL, verbose = FALSE)^2, 0)
  if (!length(out)) {
    return(0)
  }
  out
}


.frac_length <- function(x) {
  if (is.numeric(x)) {
    max(nchar(gsub(pattern = "(.\\.)(.*)", "\\2", sprintf("%f", abs(x) %% 1))))
  } else {
    0
  }
}


is.whole <- function(x) {
  (is.numeric(x) && isTRUE(all.equal(x, round(x)))) || is.character(x) || is.factor(x)
}


is.whole.number <- function(x) {
  (is.numeric(x) && isTRUE(all.equal(x, round(x))))
}


is_brms_trial <- function(model) {
  is.trial <- FALSE

  if (inherits(model, "brmsfit") && is.null(stats::formula(model)$responses)) {
    is.trial <- tryCatch(
      {
        rv <- insight::safe_deparse(stats::formula(model)$formula[[2L]])
        trimws(sub("(.*)\\|(.*)\\(([^,)]*).*", "\\2", rv)) %in% c("trials", "resp_trials")
      },
      error = function(x) {
        FALSE
      }
    )
  }

  is.trial
}


.get_model_info <- function(model) {
  faminfo <- insight::model_info(model, verbose = FALSE)
  if (!is.null(faminfo)) {
    if (insight::is_multivariate(model) && !inherits(model, c("vglm", "vgam"))) {
      faminfo <- faminfo[[1]]
    }
    faminfo$is_brms_trial <- is_brms_trial(model)
  }
  faminfo
}


.compact_list <- function(x) {
  if (is.data.frame(x)) {
    x <- x[stats::complete.cases(x), ]
  }
  x[!vapply(x, function(i) {
    !insight::is_model(i) && (length(i) == 0 || is.null(i) || (!is.function(i) && any(as.character(i) == "NULL", na.rm = TRUE)))
  }, TRUE)]
}


is.gamm <- function(x) {
  inherits(x, c("list", "gamm")) && all(names(x) %in% c("lme", "gam"))
}


is.gamm4 <- function(x) {
  inherits(x, "list") && all(names(x) %in% c("mer", "gam"))
}


.is_delta_sdmTMB <- function(x) {
  ret <- FALSE
  if (inherits(x, "sdmTMB") && isTRUE(x$family$delta)) {
    ret <- TRUE
  }
  ret
}


.n_distinct <- function(x, na.rm = TRUE) {
  if (na.rm) x <- x[!is.na(x)]
  length(unique(x))
}


# select rows where values in "variable" match "value"
.select_rows <- function(data, variable, value) {
  data[which(data[[variable]] == value), , drop = FALSE]
}


# remove column
.remove_column <- function(data, variables) {
  a <- attributes(data)
  if (!length(variables) || is.null(variables)) {
    return(data)
  }
  if (is.numeric(variables)) variables <- colnames(data)[variables]
  data <- data[, -which(colnames(data) %in% variables), drop = FALSE]
  remaining <- setdiff(names(a), names(attributes(data)))
  if (length(remaining)) attributes(data) <- c(attributes(data), a[remaining])
  data
}


.convert_numeric_factors <- function(x) {
  num_facs <- vapply(x, .is_numeric_factor, TRUE)
  if (any(num_facs)) {
    x[num_facs] <- lapply(x[num_facs], function(i) as.numeric(as.character(i)))
  }
  x
}


.is_numeric_factor <- function(x) {
  is.factor(x) && !anyNA(suppressWarnings(as.numeric(levels(x))))
}


.is_numeric_character <- function(x) {
  is.character(x) && !anyNA(suppressWarnings(as.numeric(x)))
}


.is_pseudo_numeric <- function(x) {
  if (is.factor(x)) {
    to_check <- levels(x)
  } else if (is.character(x)) {
    to_check <- x
  } else {
    return(FALSE)
  }
  # if we have a leading zero, check if length of string is greater than 1
  # only then we assume a pseudo-numeric
  any(startsWith(to_check, "0") & nchar(to_check) > 1)
}


.factor_to_numeric <- function(x, lowest = NULL) {
  if (is.numeric(x)) {
    return(x)
  }

  if (is.logical(x)) {
    return(as.numeric(x))
  }

  if (anyNA(suppressWarnings(as.numeric(as.character(stats::na.omit(x)))))) {
    if (is.character(x)) {
      x <- as.factor(x)
    }
    x <- droplevels(x)
    levels(x) <- 1:nlevels(x)
  }

  out <- as.numeric(as.character(x))

  if (!is.null(lowest)) {
    difference <- min(out) - lowest
    out <- out - difference
  }

  out
}


.check_returned_se <- function(se.pred) {
  !is.null(se.pred) && length(se.pred) > 0 && !is.null(se.pred$se.fit) && length(se.pred$se.fit) > 0
}


.mvrnorm <- function(n = 1, mu, Sigma, tol = 1e-06) {
  p <- length(mu)
  if (!all(dim(Sigma) == c(p, p))) {
    insight::format_error("Can't calculate simulated predictions. Incompatible arguments to calculate multivariate normal distribution.") # nolint
  }
  eS <- eigen(Sigma, symmetric = TRUE)
  ev <- eS$values
  if (!all(ev >= -tol * abs(ev[1L]))) {
    insight::format_error("`Can't calculate simulated predictions. Sigma` is not positive definite.")
  }
  X <- drop(mu) + eS$vectors %*% diag(sqrt(pmax(ev, 0)), p) %*% t(matrix(stats::rnorm(p * n), n))
  nm <- names(mu)
  dn <- dimnames(Sigma)
  if (is.null(nm) && !is.null(dn)) {
    nm <- dn[[1L]]
  }
  dimnames(X) <- list(nm, NULL)
  if (n == 1) {
    drop(X)
  } else {
    t(X)
  }
}


.safe <- function(code, on_error = NULL) {
  tryCatch(code, error = function(e) on_error)
}


.misspelled_string <- function(source, searchterm, default_message = NULL) {
  if (is.null(searchterm) || length(searchterm) < 1) {
    return(default_message)
  }
  # used for many matches
  more_found <- ""
  # init default
  msg <- ""
  # remove matching strings
  same <- intersect(source, searchterm)
  searchterm <- setdiff(searchterm, same)
  source <- setdiff(source, same)
  # guess the misspelled string
  possible_strings <- unlist(lapply(searchterm, function(s) {
    source[.fuzzy_grep(source, s)] # nolint
  }), use.names = FALSE)
  if (length(possible_strings)) {
    msg <- "Did you mean "
    if (length(possible_strings) > 1) {
      # make sure we don't print dozens of alternatives for larger data frames
      if (length(possible_strings) > 5) {
        more_found <- sprintf(
          " We even found %i more possible matches, not shown here.",
          length(possible_strings) - 5
        )
        possible_strings <- possible_strings[1:5]
      }
      msg <- paste0(msg, "one of ", datawizard::text_concatenate(possible_strings, last = " or ", enclose = "\""))
    } else {
      msg <- paste0(msg, "\"", possible_strings, "\"")
    }
    msg <- paste0(msg, "?", more_found)
  } else {
    msg <- default_message
  }
  # no double white space
  list(msg = insight::trim_ws(msg), possible_strings = possible_strings)
}


.fuzzy_grep <- function(x, pattern, precision = NULL) {
  if (is.null(precision)) {
    precision <- round(nchar(pattern) / 3)
  }
  if (precision > nchar(pattern)) {
    return(NULL)
  }
  p <- sprintf("(%s){~%i}", pattern, precision)
  grep(pattern = p, x = x, ignore.case = FALSE)
}


.dynEval <- function(x,
                     ifnotfound = NULL,
                     minframe = 1L,
                     inherits = FALSE,
                     remove_n_top_env = 0) {
  n <- sys.nframe() - remove_n_top_env
  x <- insight::safe_deparse(x)
  while (n > minframe) {
    n <- n - 1L
    env <- sys.frame(n)
    r <- try(eval(str2lang(x), envir = env), silent = TRUE)
    if (!inherits(r, "try-error") && !is.null(r)) {
      return(r)
    }
  }
  ifnotfound
}