Create separate vignette for 'old friends': ggmice equivalents for …

…the `mice` plotting functions

NAMESPACE

@@ -4,10 +4,10 @@ export("%>%")
 export(bwplot)
 export(densityplot)
 export(ggmice)
-export(plot_chains)
 export(plot_corr)
 export(plot_pattern)
 export(plot_pred)
+export(plot_trace)
 export(stripplot)
 export(theme_mice)
 export(xyplot)

R/corr.R

@@ -0,0 +1,56 @@
+#' Plot correlations between (incomplete) variables
+#'
+#' @param dat A dataset of class `data.frame`, `tibble`, or `matrix`.
+#' @param vrb String or vector with variable name(s), default is "all".
+#' @param label Logical indicating whether correlation values should be displayed.
+#' @param square Logical indicating whether the plot tiles should be squares.
+#' @param diagonal Logical indicating whether the correlation of each variable with itself should be displayed.
+#' @param rotate Logical indicating whether the variable name labels should be rotated 90 degrees.
+#'
+#' @return An object of class `ggplot`.
+#'
+#' @examples
+#' plot_corr(mice::nhanes, label = TRUE)
+#' @export
+plot_corr <- function(dat, vrb = "all", label = FALSE, square = TRUE, diagonal = FALSE, rotate = FALSE) {
+  if (!is.data.frame(dat) & !is.matrix(dat)) {
+    stop("Dataset should be a 'data.frame' or 'matrix'.")
+  }
+  if (vrb[1] == "all") {
+    vrb <- names(dat)
+  }
+  p <- length(vrb)
+  corrs <- data.frame(
+    vrb = rep(vrb, each = p),
+    prd = vrb,
+    corr = matrix(round(stats::cov2cor(stats::cov(data.matrix(dat[, vrb]), use = "pairwise.complete.obs")), 2), nrow = p * p, byrow = TRUE)
+  )
+  if (!diagonal) {
+    corrs[corrs$vrb == corrs$prd, "corr"] <- NA
+  }
+  gg <- ggplot2::ggplot(corrs, ggplot2::aes(x = .data$prd, y = .data$vrb, label = .data$corr, fill = .data$corr)) +
+    ggplot2::geom_tile(color = "black") +
+    ggplot2::scale_x_discrete(limits = vrb, position = "top") +
+    ggplot2::scale_y_discrete(limits = rev(vrb)) +
+    ggplot2::scale_fill_gradient2(low = "deepskyblue", mid = "lightyellow", high = "orangered", na.value = "white", limits = c(-1, 1)) +
+    ggplot2::labs(
+      x = "Imputation model predictor",
+      y = "Variable to impute",
+      fill = "Correlation*",
+      caption = "*pairwise complete observations"
+    ) +
+    theme_minimice()
+  if (label) {
+    gg <- gg + ggplot2::geom_text(color = "black", show.legend = FALSE)
+  }
+  if (square) {
+    gg <- gg + ggplot2::coord_fixed()
+  }
+  if (rotate) {
+    gg <- gg + ggplot2::theme(axis.text.x.top = ggplot2::element_text(angle = 90))
+  }
+  return(gg)
+}
+
+# TODO: add plot for missingness indicators predictors
+# TODO: maybe add model.matrix argument to correlation plot?

R/flux.R

@@ -0,0 +1,18 @@
+# plot_flux <- function(dat) {
+#   # escape function if dataset is complete
+#   # if(!any(is.na(dat))){return(plot_a_mouse())}
+#   # plot in and outflux
+#   flx <- mice::flux(dat) %>% cbind(variable = rownames(.))
+#   gg <- flx %>%
+#     ggplot2::ggplot(ggplot2::aes(x = influx,
+#                                  y = outflux,
+#                                  label = variable)) +
+#     ggplot2::geom_abline(intercept = 1,
+#                          slope = -1,
+#                          linetype = "dashed") +
+#     ggplot2::geom_text(position = ggplot2::position_jitter(width = 0.01, height = 0.01)) +
+#     ggplot2::lims(x = c(-0.01, 1.01), y = c(-0.01, 1.01)) +
+#     ggplot2::theme_classic()
+#   # output
+#   return(gg)
+# }

R/ggmice.R

@@ -21,8 +21,8 @@ ggmice <- function(data = NULL, mapping = ggplot2::aes()) {
   if (is.character(mapping$x) | is.character(mapping$y)) {
     stop("The mapping argument requires variable name(s) of type 'quosure', typically created with ggplot2::aes(). To supply a string instead, try using ggplot2::aes_string()")
   }
-  if (any(c("colour", "fill") %in% mapping_args)) {
-    warning("The aes() arguments 'colour', 'fill' and 'group' have a special use in ggmmice() and will be overwritten. Try using 'shape' or 'linetype' for additional mapping, or use faceting.")
+  if ("colour" %in% mapping_args) {
+    warning("The aes() argument 'colour' has a special use in ggmmice() and will be overwritten. Try using 'shape' or 'linetype' for additional mapping, or use faceting.")
   }
   # extract variable names from mapping object
   if (mice::is.mids(data)) {
@@ -48,7 +48,7 @@ ggmice <- function(data = NULL, mapping = ggplot2::aes()) {
       ),
       .imp = factor(.imp, ordered = TRUE)
     )
-    mice_mapping <- utils::modifyList(mapping, ggplot2::aes(colour = .where, fill = .where))
+    mice_mapping <- utils::modifyList(mapping, ggplot2::aes(colour = .where)) #, fill = .where
     mice_colors <- c("observed" = "#006CC2B3", "imputed" = "#B61A51B3")
   } else {
     where_xy <- rowSums(is.na(as.matrix(data[, c(vrb_x, vrb_y)]))) > 0L
@@ -60,13 +60,13 @@ ggmice <- function(data = NULL, mapping = ggplot2::aes()) {
       }),
       .where = factor(where_xy, levels = c(FALSE, TRUE), labels = c("observed", "missing"), ordered = TRUE)
     )
-    mice_mapping <- utils::modifyList(mapping, ggplot2::aes(colour = .where, fill = .where))
+    mice_mapping <- utils::modifyList(mapping, ggplot2::aes(colour = .where)) #, fill = .where
     mice_colors <- c("observed" = "#006CC2B3", "missing" = "#B61A51B3")
   }
   # create plot
   gg <- ggplot2::ggplot(data = mice_data, mapping = mice_mapping) +
     ggplot2::scale_color_manual(values = mice_colors, drop = TRUE, name = "") +
-    ggplot2::scale_fill_manual(values = mice_colors, drop = TRUE, name = "") +
+    #ggplot2::scale_fill_manual(values = mice_colors, drop = TRUE, name = "") +
     theme_mice()
   if (!mice::is.mids(data)) {
     gg <- gg +

R/pattern.R

@@ -67,3 +67,5 @@ plot_pattern <- function(dat, square = FALSE, rotate = FALSE) {
 
   return(gg)
 }
+
+

R/pred.R

@@ -47,61 +47,4 @@ plot_pred <- function(pred, label = FALSE, square = TRUE, rotate = FALSE) {
   return(gg)
 }
 
-#' Plot correlations between (incomplete) variables
-#'
-#' @param dat A dataset of class `data.frame`, `tibble`, or `matrix`.
-#' @param vrb String or vector with variable name(s), default is "all".
-#' @param label Logical indicating whether correlation values should be displayed.
-#' @param square Logical indicating whether the plot tiles should be squares.
-#' @param diagonal Logical indicating whether the correlation of each variable with itself should be displayed.
-#' @param rotate Logical indicating whether the variable name labels should be rotated 90 degrees.
-#'
-#' @return An object of class `ggplot`.
-#'
-#' @examples
-#' plot_corr(mice::nhanes, label = TRUE)
-#' @export
-plot_corr <- function(dat, vrb = "all", label = FALSE, square = TRUE, diagonal = FALSE, rotate = FALSE) {
-  if (!is.data.frame(dat) & !is.matrix(dat)) {
-    stop("Dataset should be a 'data.frame' or 'matrix'.")
-  }
-  if (vrb[1] == "all") {
-    vrb <- names(dat)
-  }
-  p <- length(vrb)
-  corrs <- data.frame(
-    vrb = rep(vrb, each = p),
-    prd = vrb,
-    corr = matrix(round(stats::cov2cor(stats::cov(data.matrix(dat[, vrb]), use = "pairwise.complete.obs")), 2), nrow = p * p, byrow = TRUE)
-  )
-  if (!diagonal) {
-    corrs[corrs$vrb == corrs$prd, "corr"] <- NA
-  }
-  gg <- ggplot2::ggplot(corrs, ggplot2::aes(x = .data$prd, y = .data$vrb, label = .data$corr, fill = .data$corr)) +
-    ggplot2::geom_tile(color = "black") +
-    ggplot2::scale_x_discrete(limits = vrb, position = "top") +
-    ggplot2::scale_y_discrete(limits = rev(vrb)) +
-    ggplot2::scale_fill_gradient2(low = "deepskyblue", mid = "lightyellow", high = "orangered", na.value = "white", limits = c(-1, 1)) +
-    ggplot2::labs(
-      x = "Imputation model predictor",
-      y = "Variable to impute",
-      fill = "Correlation*",
-      caption = "*pairwise complete observations"
-    ) +
-    theme_minimice()
-  if (label) {
-    gg <- gg + ggplot2::geom_text(color = "black", show.legend = FALSE)
-  }
-  if (square) {
-    gg <- gg + ggplot2::coord_fixed()
-  }
-  if (rotate) {
-    gg <- gg + ggplot2::theme(axis.text.x.top = ggplot2::element_text(angle = 90))
-  }
-  return(gg)
-}
-
 # TODO: add imputation method to pred plot
-# TODO: add plot for missingness indicators predictors
-# TODO: maybe add model.matrix argument to correlation plot?
-# TODO: add argument to rotate/shorten variable names

R/theme.R

@@ -27,3 +27,5 @@ theme_minimice <- function() {
       panel.grid.minor = ggplot2::element_blank()
     )
 }
+
+# TODO: make facets in plot_trace() look more pretty

R/chains.R → R/trace.R

@@ -1,5 +1,5 @@
 # plot convergence
-#' Convergence plot for Multiply Imputed Data Sets
+#' Plot the trace lines of the MICE algorithm for convergence evaluation
 #'
 #' @param imp An object of class `mids`.
 #' @param vrb String or vector with variable name(s), default is "all".
@@ -8,9 +8,9 @@
 #'
 #' @examples
 #' imp <- mice::mice(mice::nhanes, print = FALSE)
-#' plot_chains(imp)
+#' plot_trace(imp)
 #' @export
-plot_chains <- function(imp, vrb = "all") {
+plot_trace <- function(imp, vrb = "all") {
   if (!mice::is.mids(imp)) {
     stop("argument 'imp' must be a 'mids' object", call. = FALSE)
   }
@@ -56,3 +56,5 @@ plot_chains <- function(imp, vrb = "all") {
     ) +
     theme_mice()
 }
+
+# TODO: make iterations and statistic arguments as well

vignettes/ggmice.Rmd

@@ -88,7 +88,7 @@ The `ggmice` package contains functions to evaluate observed and imputed data.
 ## Algorithmic convergence
 
 ```{r convergence}
-plot_chains(imp, "bmi")
+plot_trace(imp, "bmi")
 ```
 
 ## Box and whiskers plot

vignettes/old_friends.Rmd

@@ -0,0 +1,113 @@
+---
+title: "old_friends"
+output: rmarkdown::html_vignette
+vignette: >
+  %\VignetteIndexEntry{old_friends}
+  %\VignetteEngine{knitr::rmarkdown}
+  %\VignetteEncoding{UTF-8}
+---
+
+```{r, include = FALSE}
+knitr::opts_chunk$set(
+  collapse = TRUE,
+  comment = "#>"
+)
+```
+
+```{r setup}
+library(ggmice)
+imp <- mice::mice(mice::boys, method = "pmm", printFlag = FALSE)
+```
+
+How to re-create the plotting functions from `mice` with `ggmice`, in alphabetical order.
+
+# `mice::bwplot()`
+
+Box-and-whisker plot of observed and imputed data.
+
+```{r bwplot}
+# original plot
+mice::bwplot(imp, bmi ~ .imp)
+# ggmice equivalent
+ggmice(imp, ggplot2::aes(x = .imp, y = bmi)) +
+      ggplot2::geom_boxplot() +
+      ggplot2::labs(x = "Imputation number")
+# extended reproduction with ggmice
+ggmice(imp, ggplot2::aes(x = .imp, y = bmi)) +
+  ggplot2::stat_boxplot(geom = 'errorbar', linetype = "dashed") +
+  ggplot2::geom_boxplot(outlier.colour = "grey", outlier.shape = 1) +
+  ggplot2::labs(x = "Imputation number")
+```
+
+# `mice::densityplot()`
+
+Density plot of observed and imputed data.
+
+```{r densityplot}
+# original plot
+mice::densityplot(imp, ~bmi)
+# ggmice equivalent
+ggmice(imp, ggplot2::aes(x = bmi, group = .imp)) +
+      ggplot2::geom_density() 
+# extended reproduction with ggmice
+ggmice(imp, ggplot2::aes(x = bmi, group = .imp, size = .where)) +
+      ggplot2::geom_density() +
+      ggplot2::scale_size_manual(values = c("observed" = 1, "imputed" = 0.5), guide = "none")
+```
+
+# `mice::flux()`
+
+[To be added]
+
+# `mice::md.pattern()`
+
+Missing data pattern plot.
+
+```{r md.pattern}
+# original plot
+mice::md.pattern(imp$data)
+# ggmice equivalent
+plot_pattern(imp$data)
+```
+
+# `mice::plot.mids()`
+
+Plot the trace lines of the MICE algorithm.
+
+```{r plot.mids}
+# original plot
+plot(imp, bmi ~ .it | .ms)
+# ggmice equivalent
+plot_trace(imp, "bmi")
+```
+
+# `mice::stripplot()`
+
+Stripplot of observed and imputed data.
+
+```{r stripplot}
+# original plot
+mice::stripplot(imp, bmi ~ .imp)
+# ggmice equivalent
+ggmice(imp, ggplot2::aes(x = .imp, y = bmi)) +
+      ggplot2::geom_jitter(width = 0.25) +
+      ggplot2::labs(x = "Imputation number")
+# extended reproduction with ggmice
+
+```
+
+# `mice::xyplot()`
+
+Scatterplot of observed and imputed data.
+
+```{r}
+# original plot
+mice::xyplot(imp, bmi ~ age)
+# ggmice equivalent
+ggmice(imp, ggplot2::aes(age, bmi)) +
+    ggplot2::geom_point()
+# extended reproduction with ggmice
+ggmice(imp, ggplot2::aes(age, bmi)) +
+    ggplot2::geom_point(shape = 1)
+```
+