R/cross_by.R

In crosstable: Crosstables for Descriptive Analyses

#' @importFrom cli cli_abort cli_warn
#' @importFrom dplyr filter pull
#' @importFrom glue glue glue_data
#' @importFrom purrr imap list_rbind
#' @importFrom rlang env
#' @importFrom stats na.omit
#' @keywords internal
#' @noRd
cross_by = function(data_x, data_y, funs, funs_arg, percent_pattern, total, percent_digits,
                    showNA, label, test, times, followup, drop_levels,
                    test_args, cor_method, effect, effect_args){
  if(!is.null(data_y) && ncol(data_y)>1) cli_abort(glue("data_y has {ncol(data_y)} columns (max=1)"))
  errors = rlang::env()

  by_levels = length(unique(na.omit(data_y[[1]])))
  if(!is.numeric(data_y[[1]]) && isTRUE(effect) && by_levels!=2){
    cli_warn(c("Cannot calculate crosstable effects as there is not exactly 2 groups in `by`.",
               i="`by` has {by_levels} level{?s}"),
             class = "crosstable_effect_2groups_warning",
             call = crosstable_caller$env)
    effect = FALSE
  }

  rtn_tbl = imap(data_x, ~{
    if(is.period(.x)){
      .x = as.numeric(.x) %>% copy_label_from(.x) %>% structure(is_period=TRUE)
    }

    if(anyNA(.x) && "NA" %in% .x) {
      .x[.x=="NA"] = "\"NA\""
    }
    if(!is.list(.x)){ #TODO is.list pour les erreurs, mieux vaudrait une classe spéciale?
      data_x[.y] = .x
      # errors[[.y]] = data.frame(name=.y, class="list")
      # return(NULL)
    }


    if(showNA=="no"){
      if(is.null(data_y)){
        data_x = data_x %>% filter(!is.na(data_x[[.y]]))
      } else {
        cc = !is.na(data_x[[.y]]) & !is.na(data_y[[1]])
        data_x = data_x %>% filter(cc)
        data_y = data_y %>% filter(cc)
      }
    }

    if(is.list(.x)){
      rtn=NULL
    } else if(is.numeric.and.not.surv(.x) || is.date(.x) || is.period(.x)){
      rtn=cross_numeric(data_x[.y], data_y, funs=funs, funs_arg=funs_arg,
                        showNA=showNA, total=total, label=label,
                        cor_digits=percent_digits, cor_method=cor_method,
                        test=test, test_args=test_args, effect=effect, effect_args=effect_args)
    } else if(is.character.or.factor(.x)){
      rtn=cross_categorical(data_x[.y], data_y, percent_pattern=percent_pattern,
                            showNA=showNA, total=total, label=label, percent_digits=percent_digits,
                            drop_levels=drop_levels,
                            test=test, test_args=test_args, effect=effect, effect_args=effect_args)
    } else if(is.Surv(.x)){
      rtn=cross_survival(data_x[.y], data_y, times=times, followup=followup,
                         showNA=showNA, total=total, label=label, surv_digits=percent_digits,
                         test=test, test_args=test_args, effect=effect, effect_args=effect_args)
    } else {
      rtn=NULL
    }
    if(is.null(rtn)){
      errors[[.y]] = data.frame(name=.y, class=paste_classes(.x))
    }

    rtn
  }) %>% list_rbind()


  errors = as.list(errors) %>% list_rbind()
  if(nrow(errors)>0){
    errors_s = glue_data(errors, "'{name}' ({class})")
    by_col = glue("'{names(data_y[1])}' ({paste_classes(data_y[[1]])})")
    if(is.null(data_y)){
      cli_warn("Could not describe column{?s} of wrong class: {errors_s}",
               class = "crosstable_wrong_col_class_warning",
               call = crosstable_caller$env)
    } else {
      cli_warn("Could not cross {qty(errors_s)} column{?s} {errors_s} by column {by_col})",
               class = "crosstable_wrong_col_class_by_warning",
               call = crosstable_caller$env)
    }
  }

  if("effect" %in% names(rtn_tbl) && any(rtn_tbl$effect=="No effect?")){
    x=rtn_tbl %>% filter(effect=="No effect?") %>% pull(.data$.id) %>% unique()
    cli_warn("Cannot calculate crosstable effects for variable{?s} {.var {x}}",
             class = "crosstable_effect_other_warning")
  }

  rownames(rtn_tbl)=NULL
  return(rtn_tbl)
}