scale_color_manual is sometimes not ordering factor levels correctly

[mariellep]

Please see my Stack Overflow post for more details! Very briefly, I checked to make sure that my factor was properly ordered and that the break and color vectors were correct, but the color sequence is still wrong in one instance.

I hope that the above link is enough to go on ( also see the pastebin for more detailed code); I can try to send an .Rdata file your way but it might be rather large.

[karawoo]

Can you please provide a minimal reprex (reproducible example)? The goal of a reprex is to make it as easy as possible for me to recreate your problem so that I can fix it: please help me help you!

If you've never heard of a reprex before, start by reading "What is a reprex", and follow the advice further down the page.

[mariellep]

I'm trying to generate some pseudocode that will have the same error, but so far I'm not able to reproduce the bug. I'll include what I did to show you what I've tried, but like I said, it's not generating the error that is noted in the Stackoverflow post. Short of giving you my data I'm not sure how I can get it to work so that you can debug it!

library(ggplot2)
library(ggmap)
library(maps)
library(mapdata)
library(RColorBrewer)

#x and y axes
lons<-seq(0,100)
lats<-seq(80,-80,-1)

#Psuedo data for the grid
longdata<-expand.grid(lons,lats)
longdata$value <- abs(longdata[,2]*2.1)
colnames(longdata)<-c("lon","lat","value")

brks_z<-seq(50,210,10)
contour_cols_z<-colorRampPalette(c("yellow","green","blue","purple","red"))(length(brks_z))

longdata$breaks<-cut(longdata$value,breaks=brks_z,include.lowest=TRUE)
brk_lev<-levels(as.factor(longdata$breaks))

longdata_n<-longdata[(longdata$lat<=75 & longdata$lat>=25),]
longdata_s<-longdata[(longdata$lat<=-25 & longdata$lat>=-75),]


m<-map_data("world2")
gplot_z<-ggplot()+ 
  coord_cartesian(xlim=c(min(lons),max(lons)),
                  ylim=c(min(lats),max(lats))) +
  geom_map(data= m, map = m, aes(map_id=region)) +
  stat_contour(data=longdata_n,aes(x=lon, y=lat, z = value),breaks=brks_z)+     
  stat_contour(data=longdata_s,aes(x=lon, y=lat, z = value),breaks=brks_z)+  
  geom_tile(data=longdata_n,aes(x=lon,y=lat,fill=breaks),alpha=0.5) +
  geom_tile(data=longdata_s,aes(x=lon,y=lat,fill=breaks),alpha=0.5) +
  scale_fill_manual(breaks=brk_lev,values=contour_cols_z)
print(gplot_z)

It produces the following:

[mariellep]

The one possible difference between the pseudocode and the actual code is the fact that I used melt on the data to get it into long format, rather than expand.grid as done here-- but would that make a difference?...

[mariellep]

The only other possible explanation I can think of is that it's because it's inside a function. I'll test that out and provide an update.

In the original code, I even tried changing the range of the breaks to include values that aren't in the plot. So for example, levels(as.factor(longdata$breaks)) produces the following:

  [1] "[50,60]"   "(60,70]"   "(70,80]"   "(80,90]"  
 [5] "(90,100]"  "(100,110]" "(110,120]" "(120,130]"
 [9] "(130,140]" "(140,150]" "(150,160]" "(160,170]"
[13] "(170,180]" "(180,190]" "(190,200]" "(200,210]"

and the contour_cols vector is

 [1] "#FFFF00" "#BBFF00" "#77FF00" "#32FF00"
 [5] "#00EE10" "#00AA54" "#006599" "#0021DD"
 [9] "#1504FD" "#3F0CF9" "#6A15F5" "#951DF1"
[13] "#B319BF" "#CC117F" "#E5083F" "#FF0000"

The very first contour interval is the fifth element in the breaks vector. But look at the plot that was produced:

[mariellep]

Update: wrapping the plot in a function in the pseudocode didn't make a difference, it still produced the same plot that I included in my pseudocode post.

At this point I'm completely stumped.

[mariellep]

Update: It's fixed by adding names to the contour color vector. See here for what I did. Can't figure out why it's happening still-- but there's a workaround...