Hexbin: apply function for every bin

问题:

I would like to build the hexbin plot where for every bin is the "ratio between class 1 and class2 points falling into this bin" is plotted (either log or not).

x <- rnorm(10000)
y <- rnorm(10000)
h <- hexbin(x,y)
plot(h)
l <- as.factor(c( rep(1,2000), rep(2,8000) ))

Any suggestions on how to implement this? Is there a way to introduce function to every bin based on bin statistics?

回答1:

@cryo111's answer has the most important ingredient - IDs = TRUE. After that it's just a matter of figuring out what you want to do with Inf's and how much do you need to scale the ratios by to get integers that will produce a pretty plot.

library(hexbin)
library(data.table)

set.seed(1)
x = rnorm(10000)
y = rnorm(10000)

h = hexbin(x, y, IDs = TRUE)

# put all the relevant data in a data.table
dt = data.table(x, y, l = c(1,1,1,2), cID = h@cID)

# group by cID and calculate whatever statistic you like
# in this case, ratio of 1's to 2's,
# and then Inf's are set to be equal to the largest ratio
dt[, list(ratio = sum(l == 1)/sum(l == 2)), keyby = cID][,
     ratio := ifelse(ratio == Inf, max(ratio[is.finite(ratio)]), ratio)][,
     # scale up (I chose a scaling manually to get a prettier graph)
     # and convert to integer and change h
     as.integer(ratio*10)] -> h@count

plot(h)

回答2:

You can determine the number of class 1 and class 2 points in each bin by

library(hexbin)
library(plyr)
x=rnorm(10000)
y=rnorm(10000)
#generate hexbin object with IDs=TRUE
#the object includes then a slot with a vector cID
#cID maps point (x[i],y[i]) to cell number cID[i]
HexObj=hexbin(x,y,IDs = TRUE)

#find count statistics for first 2000 points (class 1) and the rest (class 2)
CountDF=merge(count(HexObj@cID[1:2000]),
              count(HexObj@cID[2001:length(x)]),
              by="x",
              all=TRUE
             )
#replace NAs by 0
CountDF[is.na(CountDF)]=0
#check if all points are included
sum(CountDF$freq.x)+sum(CountDF$freq.y)

But printing them is another story. For instance, what if there are no class 2 points in one bin? The fraction is not defined then. In addition, as far as I understand hexbin is just a two dimensional histogram. As such, it counts the number of points that fall into a given bin. I do not think that it can handle non-integer data as in your case.