0

Context

I am trying to generate a heatmap for the evaluation of an exercise on process scheduling. The rows (y-axis) correspond to processes and the columns (x-axis) to the current time step.

The data is available to me as a matrix corresponding to what I described above, since geom_tile expects a data frame, I use reshape::melt to transform the data into a data frame with one column denoting the process, one the current timestep and one the value (see below). | Process | Timestep | value | | ---------- | --------- | ----- | | Scheduler | 1 | 1 | | 4 | 1 | 13 | | 3 | 1 | 13 | | 2 | 1 | 13 | | 1 | 1 | 26 | | Scheduler | 2 | 10 | | 4 | 2 | 24 | | 3 | 2 | 84 | | 2 | 2 | 19 | | 1 | 2 | 314 |

[...]

What is going wrong?

Without any interference, the type of df$Timestep is integer.

Type of Timestep in df returned by melt, before plotting with ggplot

Passing the data frame like this to ggplot, yields the following result:

Result without any modifications

As you can see, it looks like ggplot adds a column for t=0 and t=32, so before and after the actual first and last timestep:

enter image description here

This is precisely the problem I would like to tackle.

What have I tried

Change data frame column type

If I change the type of the "Timestep" column in the data frame returned by reshape::melt to character by setting df$Timestep <- as.character(df$Timestep) ggplot gets confused about where to place the columns (seems to not consider natural alphabetical order), making the plot useless.

Result after convertif df$Timestep to character

Enforce data frame column type "character" during conversion

By prepending "t" to the timesteps, turning them into "t1","t2",... ggplot seems to be able to still understand the natural order of the axis values, but this makes the axis look messy (Timestep and t1,...) and thus not a desireable solution either.

Result after prepending a character

Code

This is the code used for drawing the plots that you can see in the images within this post:

drawHeatmap <- function(dataToDraw, solutionToHighlight, plotTitle) {
  rownames(dataToDraw) <- as.character(c("Scheduler", 4:1))

  # Used to prepend a character and enfore df column type that way
  # colnames(dataToDraw) <- paste("t", 1:ncol(dataToDraw), sep="")
  colnames(dataToDraw) <- c(1:ncol(dataToDraw)
  df <- melt(dataToDraw)
  colnames(df) <- c("Process","Timestep","value")
  
  # Modify df types as needed
  # Changing df column type here makes ggplot not know how to deal with the natural ordering
  # df$Timestep <- as.character(df$Timestep)
  df$value <- as.integer(df$value)  # Convert in order to use continuous scale
  
  solution_path <- highlightCorrectPath(t(solutionToHighlight))
 
  ggplot() +
    coord_fixed() +
    scale_fill_gradientn(name = "Occurences", colors = c("blue", "green", "yellow", "orange", "red")) +
    geom_tile(data = df, aes(x=Timestep, y=Process, fill = value)) +
    scale_color_manual(values=c("black"), labels = c("Solution cells")) + 
    geom_rect(aes(xmin = solution_path[,"x_left"],
                  xmax = solution_path[,"x_right"],
                  ymin =  solution_path[,"y_bottom"],
                  ymax =  solution_path[,"y_top"], color = "Solution Cells"),
              fill=NA, size=1.1) +
    labs(title=plotTitle) +
    theme(legend.position = "bottom")
}
Liqs
  • 137
  • 1
  • 9
  • Maybe adding ```+scale_x_discrete(expand = c(0, 0)``` or ```+scale_x_continuous(expand = c(0, 0)``` would erase the pads left and right. Try to play with the values c(0,0) to see what's working best. – Nimzo Nov 23 '22 at 11:48
  • 1
    Instead of converting to a character try with converting tzo a factor, i.e. try `df$Timestep <- factor(df$Timestep)`. That should preserve the order. – stefan Nov 23 '22 at 12:39
  • Thanks @stefan that worked. Do you have any idea why ggplot seems to add padding columns for integer axes or if this is a known/common issue? – Liqs Nov 23 '22 at 20:16
  • 1
    Yep. This is a known "issue". By default ggplot2 will choose approximately 5 pretty breaks for a continuous scale. Sometime more, sometimes less. A second approach to fix that would be to set the breaks using e.g. `scale_x_continuous(breaks = 1:31)`. – stefan Nov 23 '22 at 20:50

1 Answers1

0

Solution

Adding df$Timestep <- factor(df$Timestep) fixed the issue, thanks a lot @stefan for suggesting this in the comments. However, I am still unsure what exactly caused ggplot to behave this way.

For the full code and resulting plot see below.

Code

drawHeatmap <- function(dataToDraw, solutionToHighlight, plotTitle) {
  rownames(dataToDraw) <- as.character(c( "Dispatcher", "Scheduler", 3:1))
  colnames(dataToDraw) <- c(1:ncol(dataToDraw))
  df <- melt(dataToDraw)
  colnames(df) <- c("Process","Timestep","value")
  
  # Modify df types as needed
  df$value <- factor(df$value)
  df$Timestep <- factor(df$Timestep)
  
  plot_colors <- colorRampPalette(c("blue", "green", "yellow", "orange", "red"))(nlevels(df$value))
  plot_breaks <- seq(from = levels(df$value)[1], to = levels(df$value)[nlevels(df$value)], by = 10)
  
  solution_path <- highlightCorrectPath(t(solutionToHighlight))
 
  ggplot() +
    coord_fixed() +
    scale_fill_discrete(name = "Occurences", type = plot_colors, breaks = plot_breaks) +
    geom_tile(data = df, aes(x=Timestep, y=Process, fill = value)) +
    scale_color_manual(values=c("black"), labels = c("Solution cells")) + 
    geom_rect(aes(xmin = solution_path[,"x_left"],
                  xmax = solution_path[,"x_right"],
                  ymin =  solution_path[,"y_bottom"],
                  ymax =  solution_path[,"y_top"], color = "Solution Cells"),
              fill=NA, size=1.1) +
    labs(title=plotTitle) +
    theme(legend.position = "bottom")
}

Plot

Plot after converting Timesteps column to factor

Liqs
  • 137
  • 1
  • 9