Let ggplot handle histogran binning. Fix #198

Author: cpsievert

R/ggplotly.R

@@ -242,7 +242,52 @@ gg2list <- function(p){
     if (!all(barmodes == barmodes[1]))
       warning(paste0("You have multiple barcharts or histograms with different positions; ",
                      "Plotly's layout barmode will be '", layout$barmode, "'."))
+    # for stacked bar charts, plotly cumulates bar heights, but ggplot doesn't
+    if (layout$barmode == "stack") {
+      # could speed up this function with environments or C/C++
+      unStack <- function(vec) {
+        n <- length(vec)
+        if (n == 1) return(vec)
+        seq.n <- seq_len(n)
+        names(vec) <- seq.n
+        vec <- sort(vec)
+        for (k in seq(2, n)) {
+          vec[k] <- vec[k] - sum(vec[seq(1, k-1)])
+        }
+        as.numeric(vec[as.character(seq.n)])
+      }
+      ys <- lapply(trace.list, "[[", "y")
+      xs <- lapply(trace.list, "[[", "x")
+      x.vals <- unique(unlist(xs))
+      # if there is more than one y-value (for a particular x value)
+      # then 
+      # 
+      for (val in x.vals) {
+        zs <- lapply(xs, function(x) which(x == val))
+        ys.given.x <- Map(function(x, y) y[x], zs, ys)
+        if (length(unlist(ys.given.x)) < 2) next
+        st <- unStack(unlist(ys.given.x))
+        lens <- sapply(ys.given.x, length)
+        trace.seq <- seq_along(zs)
+        ws <- split(st, rep(trace.seq, lens))
+        for (tr in trace.seq) {
+          idx <- zs[[tr]]
+          if (length(idx)) trace.list[[tr]]$y[idx] <- ws[[tr]][idx]
+        }
+      }
+    }
   }
+    
+#     lens <- sapply(ys, length)
+#      && length(trace.list) > 1 && any(lens > 1)) {
+#       xs <- unlist(xs)
+#       trace.seq <- seq_along(trace.list)
+#       idx <- rep(trace.seq, lens)
+#       
+#       
+#       browser()
+#       diffs <- tapply(unlist(ys), INDEX = xs, unStack)
+#       for (k in trace.seq) trace.list[[k]]$y <- as.numeric(sapply(diffs, "[", k))
 
   # Bar Gap for histograms should be 0
   bargaps <- do.call(c, lapply(trace.list, function (x) x$bargap))

R/trace_generation.R

@@ -41,24 +41,14 @@ layer2traces <- function(l, d, misc) {
       g$geom <- "smoothLine"
     }
   }
-  # Barmode and bargap
-  barmode <- "group"
-  if (g$geom == "bar" || g$geom == "histogram") {
-    if (l$stat$objname == "bin") {
-      if (g$geom != "histogram") {
-        warning("You may want to use geom_histogram.")
-      }
-    } else {
-      bargap <- "default"
-    }
-    g$geom <- "bar"  # histogram is just an alias for geom_bar + stat_bin
-    pos <- l$position$.super$objname
-    if (pos == "identity") {
-      barmode <- "overlay"
-    } else if (pos == "stack") {
-      barmode <- "stack"
-    }
+  # histogram is essentially a bar chart with no gaps (after stats are computed)
+  if (g$geom == "histogram") {
+    g$geom <- "bar"
+    bargap <- 0
+    misc$hist <- TRUE
   }
+
+  # TODO: remove this once we reimplement density as area
   if (g$geom == "density") {
     bargap <- 0
   }
@@ -176,18 +166,19 @@ layer2traces <- function(l, d, misc) {
       matched.names <- names(basic$data)[data.i]
       name.i <- name.names %in% matched.names
       invariable.names <- cbind(name.names, mark.names)[name.i,]
+      # fill can be variable for histograms
+      #if (misc$hist) 
+      #  invariable.names <- invariable.names[!grepl("fill", invariable.names)]
       other.names <- !names(basic$data) %in% invariable.names
       vec.list <- basic$data[is.split]
       df.list <- split(basic$data, vec.list, drop=TRUE)
       lapply(df.list, function(df){
         params <- basic$params
         params[invariable.names] <- if (ncol(x <- df[1, invariable.names]) > 0) x else NULL
-        list(data=df[other.names],
-             params=params)
+        list(data=df[other.names], params=params)
       })
     }
   }
-  
   # Split hline and vline when multiple panels or intercepts:
   # Need multiple traces accordingly.
   if (g$geom == "hline" || g$geom == "vline") {
@@ -216,7 +207,6 @@ layer2traces <- function(l, d, misc) {
   }
   traces <- NULL
   names.in.legend <- NULL
-  
   for (data.i in seq_along(data.list)) {
     data.params <- data.list[[data.i]]
     data.params$params$stat.type <- l$stat$objname
@@ -260,18 +250,21 @@ layer2traces <- function(l, d, misc) {
     if (is.null(tr$name) || tr$name %in% names.in.legend)
       tr$showlegend <- FALSE
     names.in.legend <- c(names.in.legend, tr$name)
-    
-    if (g$geom == "bar")
-      tr$barmode <- barmode
-    
-    # Bar Gap
-    if (exists("bargap")) {
-      tr$bargap <- bargap
+
+    # special handling for bars
+    if (g$geom == "bar") {
+      tr$bargap <- if (exists("bargap")) bargap else "default"
+      pos <- l$position$.super$objname
+      tr$barmode <- if (pos == "identity") {
+         "overlay"
+      } else if (pos %in% c("stack", "fill")) {
+        "stack"
+      } else "group"
     }
+    
     traces <- c(traces, list(tr))
   }
 
-  
   sort.val <- sapply(traces, function(tr){
     rank.val <- unlist(tr$sort)
     if(is.null(rank.val)){
@@ -357,16 +350,9 @@ toBasic <- list(
     g$data <- g$prestats.data
     g
   },
-  bar=function(g) {
-    if (any(is.na(g$prestats.data$x)))
-      g$prestats.data$x <- g$prestats.data$x.name
-    for(a in c("fill", "colour")){
-      g$prestats.data[[a]] <-
-        g$data[[a]][match(g$prestats.data$group, g$data$group)]
-    }
-    g$params$xstart <- min(g$data$xmin)
-    g$params$xend <- max(g$data$xmax)
-    g$data <- g$prestats.data
+  bar=function(g){
+    g <- group2NA(g, "bar")
+    g$data <- g$data[!is.na(g$data$y), ]
     g
   },
   contour=function(g) {
@@ -591,40 +577,19 @@ geom2trace <- list(
     L
   },
   bar=function(data, params) {
-    L <- list(x=data$x,
+    #data <- data[order(data$y), ]
+    x <- if ("x.name" %in% names(data)) data$x.name else data$x
+    L <- list(x=x,
+              y=data$y,
+              type="bar",
               name=params$name,
               text=data$text,
               marker=list(color=toRGB(params$fill)))
-    
     if (!is.null(params$colour)) {
       L$marker$line <- list(color=toRGB(params$colour))
       L$marker$line$width <- if (is.null(params$size)) 1 else params$size
     }
-    
-    if (!is.null(params$alpha))
-      L$opacity <- params$alpha
-    
-    if (params$stat.type == "bin") {
-      L$type <- "histogram"
-      if (is.null(params$binwidth)) {
-        L$autobinx <- TRUE
-      } else {
-        L$autobinx <- FALSE
-        L$xbins=list(start=params$xstart,
-                     end=params$xend,
-                     size=params$binwidth)
-        if (inherits(data$x.name, "POSIXt")) {
-          # Convert seconds into milliseconds
-          L$xbins <- lapply(L$xbins, function(x) x * 1000)
-        } else if (inherits(data$x.name, "Date")) {
-          # Convert days into milliseconds
-          L$xbins <- lapply(L$xbins, function(x) x * 24 * 60 * 60 * 1000)
-        }
-      }
-    } else {
-      L$y <- data$y
-      L$type <- "bar"
-    }
+    if (!is.null(params$alpha)) L$opacity <- params$alpha
     L
   },
   step=function(data, params) {

tests/testthat.R

@@ -2,4 +2,4 @@ library(testthat)
 save_outputs <- function(gg, name, ignore_ggplot=FALSE) {
   print(paste("running", name))
 }
-test_check("plotly")
+test_check("plotly", filter = "histogram")

tests/testthat/test-ggplot-bar.R

@@ -1,5 +1,20 @@
 context("bar")
 
+expect_traces <- function(gg, n.traces, name){
+  stopifnot(is.ggplot(gg))
+  stopifnot(is.numeric(n.traces))
+  save_outputs(gg, paste0("bar-", name))
+  L <- gg2list(gg)
+  is.trace <- names(L) == ""
+  all.traces <- L[is.trace]
+  no.data <- sapply(all.traces, function(tr) {
+    is.null(tr[["x"]]) && is.null(tr[["y"]])
+  })
+  has.data <- all.traces[!no.data]
+  expect_equal(length(has.data), n.traces)
+  list(traces=has.data, kwargs=L$kwargs)
+}
+
 researchers <-
   data.frame(country=c("Canada", "Canada", "Germany", "USA"),
              name=c("Warren", "Andreanne", "Stefan", "Toby"),
@@ -10,75 +25,46 @@ gg <- ggplot(researchers, aes(country, papers, fill=field))
 
 test_that("position_dodge is translated to barmode=group", {
   gg.dodge <- gg + geom_bar(stat="identity", position="dodge")
-  L <- gg2list(gg.dodge)
-  expect_equal(length(L), 3)
-  trace.names <- sapply(L[1:2], "[[", "name")
+  info <- expect_traces(gg.dodge, 3, "dodge")
+  trs <- info$traces
+  trace.names <- sapply(trs[1:2], "[[", "name")
   expect_true(all(c("Math", "Bio") %in% trace.names))
-  expect_identical(L$kwargs$layout$barmode, "group")
+  expect_identical(info$kwargs$layout$barmode, "group")
   # Check x values
-  expect_identical(as.character(L[[1]]$x[1]), "Canada")
-  expect_identical(as.character(L[[1]]$x[2]), "Germany")
-  expect_identical(as.character(L[[2]]$x[1]), "Canada")
-  expect_identical(as.character(L[[2]]$x[2]), "USA")
-  
-  save_outputs(gg.dodge, "bar-dodge")
+  expect_identical(as.character(trs[[1]]$x), c("Canada", "Germany"))
+  expect_identical(as.character(trs[[2]]$x), c("Canada", "USA"))
 })
 
 test_that("position_stack is translated to barmode=stack", {
   gg.stack <- gg + geom_bar(stat="identity", position="stack")
-  L <- gg2list(gg.stack)
-  expect_equal(length(L), 3)
-  trace.names <- sapply(L[1:2], "[[", "name")
+  info <- expect_traces(gg.stack, 3, "stack")
+  trs <- info$traces
+  trace.names <- sapply(trs[1:2], "[[", "name")
   expect_true(all(c("Math", "Bio") %in% trace.names))
-  expect_identical(L$kwargs$layout$barmode, "stack")
-  
-  save_outputs(gg.stack, "bar-stack")
+  expect_identical(info$kwargs$layout$barmode, "stack")
 })
 
 test_that("position_identity is translated to barmode=overlay", {
   gg.identity <- gg + geom_bar(stat="identity", position="identity")
-  L <- gg2list(gg.identity)
-  expect_equal(length(L), 3)
-  trace.names <- sapply(L[1:2], "[[", "name")
+  info <- expect_traces(gg.identity, 3, "identity")
+  trs <- info$traces
+  trace.names <- sapply(trs[1:2], "[[", "name")
   expect_true(all(c("Math", "Bio") %in% trace.names))
-  expect_identical(L$kwargs$layout$barmode, "overlay")
-  
-  save_outputs(gg.identity, "bar-identity")
+  expect_identical(info$kwargs$layout$barmode, "overlay")
 })
 
 test_that("dates work well with bar charts", {
-  
   researchers$month <- c("2012-01-01", "2012-01-01", "2012-02-01", "2012-02-01")
   researchers$month <- as.Date(researchers$month)
-  
   gd <- ggplot(researchers, aes(month, papers, fill=field)) +
     geom_bar(stat="identity")
-  
-  L <- gg2list(gd)
-  
-  expect_equal(length(L), 3)  # 2 traces + layout
-  expect_identical(L$kwargs$layout$xaxis$type, "date")
-  expect_identical(L[[1]]$x[1], "2012-01-01 00:00:00")
-  expect_identical(L[[1]]$x[2], "2012-02-01 00:00:00")
-  
-  save_outputs(gd, "bar-dates")
+  info <- expect_traces(gd, 3, "dates")
+  trs <- info$traces
+  expect_identical(info$kwargs$layout$xaxis$type, "date")
+  expect_identical(trs[[1]]$x[1], "2012-01-01 00:00:00")
+  expect_identical(trs[[1]]$x[2], "2012-02-01 00:00:00")
 })
 
-expect_traces <- function(gg, n.traces, name){
-  stopifnot(is.ggplot(gg))
-  stopifnot(is.numeric(n.traces))
-  save_outputs(gg, paste0("bar-", name))
-  L <- gg2list(gg)
-  is.trace <- names(L) == ""
-  all.traces <- L[is.trace]
-  no.data <- sapply(all.traces, function(tr) {
-    is.null(tr[["x"]]) && is.null(tr[["y"]])
-  })
-  has.data <- all.traces[!no.data]
-  expect_equal(length(has.data), n.traces)
-  list(traces=has.data, kwargs=L$kwargs)
-}
-
 ## http://www.cookbook-r.com/Graphs/Bar_and_line_graphs_%28ggplot2%29/
 df <- data.frame(time = factor(c("Lunch","Dinner"), levels=c("Lunch","Dinner")),
                  total_bill = c(14.89, 17.23))
@@ -182,3 +168,23 @@ test_that("guides(fill=FALSE) does not affect colour legend", {
   expect_true(info$kwargs$layout$showlegend)
 })
 
+
+base <- ggplot(mtcars, aes(factor(vs), fill=factor(cyl))) 
+
+test_that("geom_bar() stacks counts", { 
+  info <- expect_traces(base + geom_bar(), 3, "position-stack")
+  expect_identical(info$kwargs$layout$barmode, "stack")
+  trs <- info$traces
+  test <- colSums(t(sapply(trs, "[[", "y")), na.rm = TRUE)
+  true <- as.numeric(colSums(with(mtcars, table(cyl, vs))))
+  expect_identical(test, true)
+})
+
+test_that("geom_bar(position = 'fill') stacks proportions", {
+  info <- expect_traces(base + geom_bar(position = "fill"), 3, "position-fill")
+  expect_identical(info$kwargs$layout$barmode, "stack")
+  trs <- info$traces
+  props <- colSums(t(sapply(trs, "[[", "y")), na.rm = TRUE)
+  expect_identical(props, c(1, 1))
+})
+