Heat map R with a wide range

Question

I am producing a heat map from this table

x

   .     0       1       2       3       4       5       6       7       8

  0    12820     720     807     879    1051     824     587     732     874
  1      557   38417   41289   44380   57301   42992   30805   41092   45616
  2       62   59575   83247   72433   95751   76113   50002   92921   72773
  3       23   45346  100836   57101   57903   50625   35223   52695   47868
  4        4   14718   40000   13135    5985   13188   19252    8044    7095
  5        0    4459    6828     674     890    5251    4959     399     563
  6        0     216     333     142     115     189     202     128     128
  7        0     188      97      30      57     255      19      51      29
  8        1      20      38      13       7       4      11      44      17
  9        0      11       9       1       0       8      12     102       6
  10    7992    9620   16841   11065    9917    9619    8133    9291    8554

           9      10

  0        6   10804
  1       55   26041
  2       33   45915
  3       17   35198
  4        1   10071
  5        0    2092
  6        0     102
  7        0      29
  8        0       1
  9        1       1
  10     487 1319070

With this code

hv <- heatmap.2(x, col=cm.colors(255), scale="none",trace='none', density="none",lmat=rbind( c(0, 3), c(2,1), c(0,4) ), lhei=c(1.5, 4, 2 ),Rowv = FALSE, Colv = FALSE,dendrogram = "none",margins=c(5,5),xlab="ICOADS ship speed indicator",ylab="ICOADS course indicator")

I produce this plot

which is not very useful. Anyway, if I don't want to put scale ="row" or scale="column" because it will be hard to interpret the result. Is there a way of using scale="none" and seeing different ranges of colours?

Thanks

Answer 1

Well, here's one way using ggplot.

library(reshape2)       # for melt
library(RColorBrewer)   # for brewer.pal(...)
library(ggplot2)

x  <- cbind(id=as.numeric(rownames(x)),x)
gg <- melt(x,id="id")
gg$variable <- as.numeric(substr(gg$variable,2,4))
brks   <- c(0,10^rep(0:7))
gg$bin <- cut(gg$value,breaks=brks,labels=0:7,include.lowest=T)

ggplot(gg) + 
  geom_tile(aes(x=factor(id),y=factor(10-variable),fill=bin))+
  scale_fill_manual(name="",labels=brks,values=rev(brewer.pal(8,"Spectral")))+
  scale_y_discrete(labels=10:0)+
  labs(x="",y="")+
  theme_bw()+theme(panel.border=element_blank())

The basic idea is to use a logarithmic scale for the colors. This is a bit of a problem for you because you have zeros. So a work-around sets bins using the cut(...) function. The bin breaks are set to 0,1,10,100,...,1e7.

Your data is in "wide" format: different values of "y" in different columns, whereas ggplot needs the data in "long" format: all the data in one column with x-values and y-values in separate columns. We convert wide to long using the melt(...) function. The rest of the code sets up the bins and the labels, and formats the plot to make it look as close to yours as I could manage.