[R] pipe data from plot(). was: ROCR.plot methods, cross validation averaging

[Tim Howard]

David, 
Thank you for your reply. Yes, I can access the y-values slot with perf at y-values
but, note that in the cross-validation example (ROCR.xval), the plot function 
averages across the list of ten vectors in the y-values slot. 

I might be able to create a function to average across these ten vectors, but, since 
the plot function already does it for me, I thought it most efficient to get the values
from the function.  The compounding factor is that averaging needs to incorporate 
some kind of complex (to me at least) equalization based on the third slot (alpha.values). 

I don't know how to average vectors (especially uneven-length vectors) that align
using the alpha-values (suggestions here welcome!). Again, the plot function does 
this for me... if I could just get those values. 


Tobias, 
You suggestion to change the plot.performance function is a good one. I'll see if 
I can get in there and tweak it. 


Thanks to both of you for the help.
Tim


>>> David Winsemius <dwinsemius at comcast.net> 9/24/2009 9:43 AM >>>

On Sep 24, 2009, at 9:09 AM, Tim Howard wrote:

> All,
> I'm trying again with a slightly more generic version of my first  
> question. I can extract the
> plotted values from hist(), boxplot(), and even plot.randomForest().  
> Observe:
>
> # get some data
> dat <- rnorm(100)
> # grab histogram data
> hdat <- hist(dat)
> hdat     #provides details of the hist output
>
> #grab boxplot data
> bdat <- boxplot(dat)
> bdat     #provides details of the boxplot output
>
> # the same works for randomForest
> library(randomForest)
> data(mtcars)
> RFdat <- plot(randomForest(mpg ~ ., mtcars, keep.forest=FALSE,  
> ntree=100), log="y")
> RFdat
>
>
> ##But, I can't use this method in ROCR
> library(ROCR)
> data(ROCR.xval)
> RCdat <- plot(perf, avg="threshold")
>
> RCdat
> ## output:  NULL
>
> Does anyone have any tricks for piping or extracting these data?
> Or, perhaps for steering me in another direction?

After looking at the examples in ROCR, my guess is that you really  
ought to examine the perf object itself. It's an S4 object so some of  
the access to internals are a bit different. In the example  
performance object I just created, the y-values slot values would ba  
obtainable with:

perf at y.values 

  The is also help from:
?"plot-methods"

-- 
David
>
> Thanks,
> Tim
>
>
> From: "Tim Howard" <tghoward at gw.dec.state.ny.us>
> Subject: [R] ROCR.plot methods, cross validation averaging
> To: <osander at mpi-sb.mpg.de>, <tobias.sing at mpi-sb.mpg.de>,
> 	<r-help at r-project.org>
> Message-ID: <4ABA1079.6D16.00D5.0 at gw.dec.state.ny.us>
> Content-Type: text/plain; charset=US-ASCII
>
> Dear R-help and ROCR developers (Tobias Sing and Oliver Sander) -
>
> I think my first question is generic and could apply to many methods,
> which is why I'm directing this initially to R-help as well as  
> Tobias and Oliver.
>
> Question 1. The plot function in ROCR will average your cross  
> validation
> data if asked. I'd like to use that averaged data to find a "best"  
> cutoff
> but I can't figure out how to grab the actual data that get plotted.
> A simple redirect of the plot (such as test <- plot(mydata)) doesn't  
> do it.
>
> Question 2. I am asking ROCR to average lists with varying lengths for
> each list entry. See my example below. None of the ROCR examples  
> have data
> structured in this manner. Can anyone speak to whether the averaging
> methods in ROCR allow for this? If I can't easily grab the data as  
> desired
> from Question 1, can someone help me figure out how to average the  
> lists,
> by threshold, similarly?
>
> Question 3. If my cross validation data happen to have a list entry  
> whose
> length = 2, ROCR errors out. Please see the second part of my example.
> Any suggestions?
>
> #reproducible examples exemplifying my questions
> ##part one##
> library(ROCR)
> data(ROCR.xval)
> # set up data so it looks more like my real data
> sampSize <- c(4, 55, 20, 75, 350, 250, 6, 120, 200, 25)
> testSet <- ROCR.xval
> # do the extraction
> for (i in 1:length(ROCR.xval[[1]])){
>  y <- sample(c(1:350),sampSize[i])
>  testSet$predictions[[i]] <- ROCR.xval$predictions[[i]][y]
>  testSet$labels[[i]] <- ROCR.xval$labels[[i]][y]
>  }
> # now massage the data using ROCR, set up for a ROC plot
> # if it errors out here, run the above sample again.
> pred <- prediction(testSet$predictions, testSet$labels)
> perf <- performance(pred,"tpr","fpr")
> # create the ROC plot, averaging by cutoff value
> plot(perf, avg="threshold")
> # check out the structure of the data
> str(perf)
> # note the ragged edges of the list and that I assume averaging
> # whether it be vertical, horizontal, or threshold, somehow
> # accounts for this?
>
> ## part two ##
> # add a list entry with only two values
> perf at x.values[[1]] <- c(0,1)
> perf at y.values[[1]] <- c(0,1)
> perf at alpha.values[[1]] <- c(Inf,0)
>
> plot(perf, avg="threshold")
>
> ##output results in an error with this message
> # Error in if (from == to) rep.int(from, length.out) else  
> as.vector(c(from,  :
> # missing value where TRUE/FALSE needed
>
>
> Thanks in advance for your help
> Tim Howard
> New York Natural Heritage Program
>
> ______________________________________________
> R-help at r-project.org mailing list
> https://stat.ethz.ch/mailman/listinfo/r-help 
> PLEASE do read the posting guide http://www.R-project.org/posting-guide.html 
> and provide commented, minimal, self-contained, reproducible code.

David Winsemius, MD
Heritage Laboratories
West Hartford, CT