Skip to content

3d barplot in rgl

9 messages · Chris Stubben, Hadley Wickham, Duncan Murdoch +1 more

Original
Chris Stubben
# 
Is there anyway to plot a matrix using a 3d bar plot.  Something like  
bar3 in matlab?

The example in demo hist3d does a 3d barplot for binned data, but has
anyone tried something for a simple matrix with spaces betwen bars
and axis labels using matrix dimnames or 1,2,3?


stages<-letters[1:3]

A<-matrix(c(
0.21, 0.21,0.03,
0.55, 0.58, 0.09,
1.30, 1.35, 0.22), nrow=3, byrow=TRUE, dimnames=list(stages,stages) )

## I can get a surface plot, but that's about it.

persp3d(A, col="red", alpha=0.7,
xlab="fate", ylab="stage", zlab="Sensitivity", box=FALSE)


Thanks,

Chris Stubben

--
-------------------

Los Alamos National Lab
BioScience Division
MS M888
Los Alamos, NM 87545
Hadley Wickham
# 
Why do you want a 3d barchart?  They are generally a bad way to
present information as tall bars can obscure short bars, and it is
hard to accurately read off the height of a bar.  While adding
rotation can reduce some of these problems, why not create a graphic
that your viewers can take in with a glance?

Hadley
On 9/25/07, stubben <stubben at lanl.gov> wrote:

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Duncan Murdoch
    

    

      
      #
    

  


  

      
On 9/25/2007 2:34 PM, stubben wrote:

            

      
      
      
    

        
That demo gives you the basics of the code, so it shouldn't be too hard 
to put your own together:  just strip out the counting part.

Duncan Murdoch

  
  


  


      

    

    

      
      

        


  

        

      Chris Stubben
    

    

      
      #
    

  


  

      
hadley wickham <h.wickham <at> gmail.com> writes:

            

      
      
      
    

        
3d barplots are a common way to display sensitvity/elasticity matrices in
stage-structured demography.  Here's a few other options from Caswell's Matrix
population models book (2001) - I definitely  prefer 3d barcharts to these
alternatives.


heatmap(log10(A[3:1,]), Rowv = NA,  Colv = NA, scale="none")

plot(log10(c(A)), type="s")


Thanks,

Chris

  
  


  


      

    

    

      
      

        


  

        

      Chris Stubben
    

    

      
      #
    

  


  

      
Duncan Murdoch <murdoch <at> stats.uwo.ca> writes:

            

      
      
      
    

        
Thanks, I did download the source to check the hist3d demo, but honestly it
didn't look very easy to simplify.  I switched to R from matlab, and 3d barplots
are the only thing I still use my student edition of matlab for anymore.  

I  would like to see a barplot3d added to your rgl package in the future, but
was hoping someone else had tried already.


Chris

  
  


  


      

    

    

      
      

        


  

        

      Hadley Wickham
    

    

      
      #
    

  


  

      
On 9/25/07, Chris Stubben <stubben at lanl.gov> wrote:

            

      
      
      
    

        
Both of those look like equally awful alternatives.  Perhaps you could
explain more about your data (perhaps with a sample?) so that we could
suggest better alternatives.

Hadley

  
  


  


      

    

    

      
      

        


  

        

      Duncan Murdoch
    

    

      
      #
    

  


  

      
On 9/25/2007 3:13 PM, Chris Stubben wrote:

            

      
      
      
    

        
They probably will someday, but not until someone who needs them sits 
down and writes the code.  I tend to agree with Hadley that they aren't 
a particularly effective sort of graph so it'll likely be someone else.

Duncan Murdoch

  
  


  


      

    

    

      
      

        


  

        

      Ben Bolker
    

    

      
      #
    

  


  

      
hadley wrote:

            

      
      
      
    

        
Transition matrices are Markov transition matrices among different
life stages of organisms -- in the simplest case (Leslie matrices,
tracking age-structured populations), the top row of the matrix represents
fecundity and the first off-diagonal (below the diagonal) represents
survival.    The elasticities and sensitivities are differential (absolute
or proportional) responses
of long-run population growth rate [i.e., the largest eigenvalue] to changes
in
matrix elements.

  In the Leslie case the matrix is quite sparse and it would be easy to just
plot sensitivities/elasticities of fecundities and survivorship.

  In the more complicated case (Lefkovitch matrices), the categories
represent
stages rather than ages so individuals can stay in the same class from year
to year,
or even revert to the previous class, so the matrix is less sparse and the
problem
is a little more challenging.

   Here are some data from the popbio package on CRAN
for playing with.

library(popbio)
example(teasel)   ## shows 2 plots -- heatmap and stair-step
A <- tea$sensitivities ##

Here's one possibility, setting 0 sensitivities to NA to omit
them from the plot:

z <- as.data.frame.table(A)
names(z) <- c("From","To","Sensitivity")
z$Sensitivity[z$Sensitivity==0] <- NA
dotplot(From~Sensitivity|To,data=z)
dotplot(From~Sensitivity|To,data=z,scales=list(x=list(log=TRUE)))

  I turned the plot horizontally to make it easier to draw labels
(as lattice often does), but this might be too confusing.

  Sorry not to use ggplot2, but I haven't sat down to try to figure
it out yet ....

  
    

      
      
    

  


  


      

    

    

      
      

        


  

        

      Chris Stubben
    

    

      
      #
    

  


  

          

      
      
      
    

        
Ben,

Thanks for your clear explanation and plot examples.  I like the dotplots alot
and added a few modifications below. Since I often compare rows and columns as
well as groups of elements representing growth (lower left trianlge), stasis
(diagonal), fertility etc., I like to preserve the matrix structure in the plot
if possible.

            

      
      
      
    

        
#switch labels here- stage at time t in columns and fate at time t+1 in rows
names(z) <- c("To","From","Sensitivity")

# reverse order on rows to match matrix
z$To <- ordered(z$To, levels = rev(levels(z$To)))

dotplot(To~Sensitivity|From, data=z,
scales=list(alternating=c(1,0), tck=c(1,0)), layout=c(6,1), aspect=2)

# or log-scale 
dotplot(To~Sensitivity|From,data=z,scales=list(x=list(log=TRUE)), as.table=TRUE)


Thanks again,

Chris