0

I am trying to adapt a worked example from ggplot2 heatmaps: using different gradients for categories

However the values that I want to plot are discrete (I think). I have already standardised my values (in a stored procedure) into percent values between 0 and 1. If the percent_value is 0 then I was to show white. If the percent value is 1 then I want to show the full colour. The colour gradates from white to full. Each category has it own colour.

Here is my code...

library(RColorBrewer)
rm(list=ls())
yval <- c("51140/1234.5985/16:25:17" ,"51140/1234.5985/16:25:17" ,"51140/1234.5985/16:25:17" ,"51141/1234.5985/16:25:17" ,"51146/1234.5985/16:25:17" ,"51146/1234.5985/16:25:17" ,"51146/1234.5985/16:25:17" ,"51147/1234.5985/16:25:17" ,"51147/1234.5985/16:25:17" ,"51147/1234.5985/16:25:17" ,"51149/1234.5985/16:25:17" ,"51150/1234.5985/16:25:17" ,"51150/1234.5985/16:25:17" ,"51150/1234.5985/16:25:17" ,"51153/1234.5985/16:25:17" ,"51153/1234.5985/16:25:17" ,"51153/1234.5985/16:25:17")
cat <- c("cat1" ,"cat1" ,"cat1" ,"cat2" ,"cat1" ,"cat1" ,"cat1" ,"cat1" ,"cat1" ,"cat1" ,"cat2" ,"cat1" ,"cat1" ,"cat1" ,"cat1" ,"cat1" ,"cat1")
xval <- c("cat1.ant" ,"cat1.output3" ,"cat1.input5" ,"cat2.cat2_active_state" ,"cat1.input5" ,"cat1.output3" ,"cat1.ant" ,"cat1.ant" ,"cat1.output3" ,"cat1.input5" ,"cat2.cat2_active_state" ,"cat1.input5" ,"cat1.ant" ,"cat1.output3" ,"cat1.output3" ,"cat1.ant" ,"cat1.input5")
value <- c(0.75 ,1 ,1 ,0.1 ,1 ,1 ,0.75 ,0 ,1 ,1 ,1 ,1 ,0.75 ,1 ,1 ,0.75 ,1)
dat <- data.frame(xval, yval, cat, value)

n <- length(unique(dat$cat))
qual_col_pals = brewer.pal.info[brewer.pal.info$category == 'qual',]
col_vector = unlist(mapply(brewer.pal, qual_col_pals$maxcolors, rownames(qual_col_pals)))
sample_colours <- sample(col_vector, n)

# 2 categories. I've hard-coded the gradient ends in this example.
# I've tried translating the value up the number line to separate the categories into different colour bands.
gradientends <- c(0, 1, 2, 3)

interleave <- function(v1,v2)
{
  ord1 <- 2*(1:length(v1))-1
  ord2 <- 2*(1:length(v2))
  c(v1,v2)[order(c(ord1,ord2))]
}
colorends <- interleave(rep("white",n),sample_colours)

ggplot(dat, aes(x = xval, y = factor(yval))) + 
   geom_tile(aes(fill = value), colour = "grey80") + 
   geom_text(aes(label = value)) +
   scale_fill_gradientn(colours = colorends) + #, values = gradientends) +
   theme(axis.ticks = element_blank(), 
         axis.text.x = element_text(angle = 330, hjust = 0))

I have tried various approaches and it seems to me that scale_fill_gradient is probably not a good way to approach this. It appears that the scale function is "adjusting" values on the fly because depending on the values I am plotting then I get my heatmap looking correct or not.

Is there a way around it with this approach or perhaps there is a better approach?

Liam

Community
  • 1
  • 1
Liam Caffrey
  • 172
  • 1
  • 12

1 Answers1

0

I have figured out how to get my example working. It turns out I was getting the gradientends wrong and I should have been doing a rescale in scale_fill_gradientn(colours = colorends, values = rescale(gradientends)). To be honest, I'm not quite sure what is happening here! Presumeably the gradientends are getting rescaled in the same manner as scale_fill_ is filling in scaled rescaleoffset values, so everything lines up correctly with no overspill into neighbouring colour blocks.

Here is the working code. I have put the data in dput() ofrmat as suggested in SO guidelines. I've included the value and rescaloffset values in the geom_text (which helps debugging). I also added another category to complicate it a bit.

rm(list=ls())
library(RColorBrewer)

dat <- structure(list(xval = structure(c(5L, 3L, 2L, 4L, 2L, 3L, 1L, 
1L, 3L, 2L, 4L, 2L, 1L, 3L, 3L, 1L, 2L), .Label = c("cat1.ant", 
"cat1.input5", "cat1.output3", "cat2.cat2_active_state", "cat3.ant"
), class = "factor"), yval = structure(c(1L, 1L, 1L, 2L, 3L, 
3L, 3L, 4L, 4L, 4L, 5L, 6L, 6L, 6L, 7L, 7L, 7L), .Label = c("51140/1234.5985/16:25:17", 
"51141/1234.5985/16:25:17", "51146/1234.5985/16:25:17", "51147/1234.5985/16:25:17", 
"51149/1234.5985/16:25:17", "51150/1234.5985/16:25:17", "51153/1234.5985/16:25:17"
), class = "factor"), cat = structure(c(3L, 1L, 1L, 2L, 1L, 1L, 
1L, 1L, 1L, 1L, 2L, 1L, 1L, 1L, 1L, 1L, 1L), .Label = c("cat1", 
"cat2", "cat3"), class = "factor"), value = c(0.75, 1, 1, 0.1, 
1, 1, 0.75, 0, 1, 1, 1, 1, 0.75, 1, 1, 0.75, 1), rescaleoffset = c(200.75, 
1, 1, 100.1, 1, 1, 0.75, 0, 1, 1, 101, 1, 0.75, 1, 1, 0.75, 1
)), .Names = c("xval", "yval", "cat", "value", "rescaleoffset"
), row.names = c(NA, -17L), class = "data.frame")

n <- length(unique(dat$cat))
qual_col_pals = brewer.pal.info[brewer.pal.info$category == 'qual',]
col_vector = unlist(mapply(brewer.pal, qual_col_pals$maxcolors, rownames(qual_col_pals)))
sample_colours <- sample(col_vector, n)

dat$rescaleoffset <- dat$value + 100*(as.numeric(dat$cat)-1)

scalerange <- range(dat$value)
gradientends <- scalerange + rep(c(0,100,200), each=2)
colorends <- c("white", "red", "white", "green", "white", "blue")

ggplot(dat, aes(xval, yval)) + 
  geom_tile(aes(fill = rescaleoffset), colour = "white") + 
  geom_text(aes(label = paste(format(round(value, 5), nsmall = 5), format(round(rescaleoffset, 5), nsmall = 5), sep='\n'))) +
  scale_fill_gradientn(colours = colorends, values = rescale(gradientends)) + 
  scale_x_discrete("", expand = c(0, 0)) + 
  scale_y_discrete("", expand = c(0, 0)) + 
  theme_grey(base_size = 9) + 
  theme(axis.ticks = element_blank(), 
        axis.text.x = element_text(angle = 330, hjust = 0))+
  theme(legend.background = element_rect(fill="gray90", size=30, linetype="dotted"))

Although the values are numeric and would appear to be continuous, they actually represent discrete categoric values. Overall, I'm happy with this and is exactly what I am looking for, although some work required on the formatting and parametrisation.

EDIT: Now I am really baffled. Here is a similar set of data but it is not plotting as I expect. I expect the BCU1 catgeory to be darkviolet (not white) because it has a value of 1.0. There is something that I am not understanding with the scaling. Could anyone help?

dat <- structure(list(heatmap_row_display = structure(c(2L, 6L, 5L, 
8L, 4L, 3L, 7L, 9L, 1L, 3L, 7L, 9L, 4L, 1L, 4L, 1L, 3L, 7L, 9L
), .Label = c("051140/1084.8158/16:25:17", "051141/1084.8466/16:25:17", 
"051146/1084.8803/16:25:17", "051147/1084.8876/16:25:17", "051148/1084.8965/16:25:17", 
"051149/1084.9465/16:25:17", "051150/1084.9525/16:25:17", "051152/1084.9965/16:25:17", 
"051153/1085.0193/16:25:17"), class = "factor"), msg_no = c(51141L, 
51149L, 51148L, 51152L, 51147L, 51146L, 51150L, 51153L, 51140L, 
51146L, 51150L, 51153L, 51147L, 51140L, 51147L, 51140L, 51146L, 
51150L, 51153L), relative_time_ms = c(1084.8466, 1084.9465, 1084.8965, 
1084.9965, 1084.8876, 1084.8803, 1084.9525, 1085.0193, 1084.8158, 
1084.8803, 1084.9525, 1085.0193, 1084.8876, 1084.8158, 1084.8876, 
1084.8158, 1084.8803, 1084.9525, 1085.0193), pcan_rx_datetime_adjusted = structure(c(1487089517, 
1487089517, 1487089517, 1487089517, 1487089517, 1487089517, 1487089517, 
1487089517, 1487089517, 1487089517, 1487089517, 1487089517, 1487089517, 
1487089517, 1487089517, 1487089517, 1487089517, 1487089517, 1487089517
), class = c("POSIXct", "POSIXt"), tzone = ""), block_name = structure(c(1L, 
1L, 2L, 2L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 
3L, 3L), .Label = c("BCU1", "BCU2", "IDC1_status"), class = "factor"), 
    pcan_attribute = structure(c(1L, 1L, 2L, 2L, 3L, 3L, 3L, 
    3L, 3L, 4L, 4L, 4L, 4L, 4L, 5L, 5L, 5L, 5L, 5L), .Label = c("BCU1.BCU1_active_state", 
    "BCU2.BCU2_active_state", "IDC1_status.IDC1_ant", "IDC1_status.IDC1_input5", 
    "IDC1_status.IDC1_output3"), class = "factor"), data_value_as_string = c(1L, 
    1L, 1L, 1L, 0L, 3L, 3L, 3L, 3L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 
    1L, 1L, 1L), data_value = c(1L, 1L, 1L, 1L, 0L, 3L, 3L, 3L, 
    3L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L, 1L), value = c(1, 
    1, 1, 1, 0, 0.75, 0.75, 0.75, 0.75, 1, 1, 1, 1, 1, 1, 1, 
    1, 1, 1), colour = structure(c(2L, 2L, 1L, 1L, 3L, 3L, 3L, 
    3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L, 3L), .Label = c("cyan", 
    "darkviolet", "deeppink"), class = "factor"), rescaleoffset = c(1, 
    1, 101, 101, 200, 200.75, 200.75, 200.75, 200.75, 201, 201, 
    201, 201, 201, 201, 201, 201, 201, 201)), .Names = c("heatmap_row_display", 
"msg_no", "relative_time_ms", "pcan_rx_datetime_adjusted", "block_name", 
"pcan_attribute", "data_value_as_string", "data_value", "value", 
"colour", "rescaleoffset"), row.names = c(NA, 19L), class = "data.frame")

n <- length(unique(dat$block_name))
# Do it this way to avoid reordering the colours in the data frame
sample_colours <- levels(factor(dat$colour, levels=unique(dat$colour)))

# Rescale all the values into categories of 100
dat$rescaleoffset <- dat$value + 100*(as.numeric(dat$block_name)-1)

scalerange <- range(dat$value)
# Mark the end of each gradient for each category block.
gradientends <- scalerange + rep(seq(0, (n - 1) * 100, by = 100), each=2)

# Interleave two vectors, used to interleave "white" with each of the category colours. 
# "white" is used to colour the values on lowest end of each category's gradient range.
interleave <- function(v1,v2)
{
  ord1 <- 2*(1:length(v1))-1
  ord2 <- 2*(1:length(v2))
  c(v1,v2)[order(c(ord1,ord2))]
}
colorends <- interleave(rep("white",n),sample_colours)

p <- ggplot(dat, aes(pcan_attribute, heatmap_row_display)) + 
  geom_tile(aes(fill = rescaleoffset), colour = "white") + 
  geom_text(aes(label = paste(format(round(value, 1), nsmall = 1), sep='\n')), size=rel(2.0)) +
  scale_fill_gradientn(colours = colorends, values = rescale(gradientends)) + 
  scale_x_discrete("", expand = c(0, 0)) + 
  scale_y_discrete("", expand = c(0, 0)) + 
  theme_grey(base_size = 9) + 
  theme(axis.ticks = element_blank(), 
        axis.text.x = element_text(angle = 330, hjust = 0))+
  theme(legend.background = element_rect(fill="gray90", size=30, linetype="dotted")) 

print(p)
Liam Caffrey
  • 172
  • 1
  • 12