Knitr: Use nested loops to generate multiple reports

Question

A previous post on StackOverflow had a very useful guide on how to use a template .Rnw file to produce multiple reports. I'd like to duplicate this, except I'd have 4 loops rather than just the one used in the example.

In my case, these are the loops that I'm using to produce the reports:

outermost loop: looping through a vector that contains the names of the variables I want to analyze
inner 3 loops: looping through different seasons/geographic locations

Here is a sample of the template code, which works without any issues when I supply all of the necessary variables that would be provided in the loops:

\begin{document}
This is a test in which the \texttt{\Sexpr{varname}} variable is used in this report. If successful, further reports can be generated using a loop with this same script!

First, a plot of the CDFs, with Kolmogorov-Smirnov statistics:
\begin{figure}[h]
\centering
<<cdf-plots,dev='png',out.width='0.5\\linewidth',echo=FALSE,warning=FALSE,fig.align='centering'>>=
 chart_stats(varname,data_vec,labs,season,s,h,colvec=colvec,cdf_plot=TRUE)
 cap1=paste('Cumulative distribution functions for the counts,',season,s,h)
@
\caption{\Sexpr{cap1}}
\end{figure}
<<ks_calc,warning=FALSE,results='asis',echo=FALSE>>=
library(xtable)
ks=chart_stats(varname,data_vec,labs,season,s,h,ks_test=TRUE,dval=TRUE)
cap=paste('D-values for',season,h,s)
print(xtable(ks,caption=cap))
@

For the sake of being able to see my output, each file is currently generated as a separate .tex file as shown:

library(knitr)
setwd("~/data_netcdf")
load("loaded_data.Rdata")
source("~/tempestextremes/test/chart_stats.R")

data_vec<-c("ERA", "climo", "2xCO2","SSTplus2","SSTplus2_2xCO2")

seasons_vec<-c("DJF","JJA","MAM","SON")
sec=c("ATL","PAC")
hemi=c("NH","SH")
var=c('centlat','centlon','area')

for (varname in var){
  }
  for (season in seasons_vec){
    for (h in hemi){
      for (s in sec){
        output_name=paste(varname,'_',season,'_',h,'_',s,'_report.tex',sep="")
        knit2pdf("~/data_netcdf/report_test.Rnw",output=output_name)
      }
    }   
  }
}

There is a knitr example that makes use of child files to combine outputs into one document, which sounds similar to what I'd like to do, and here's what I've tried:

<<test-main,include=FALSE>>= 
[removed for length]

#testing on just one variable name
var=c('centlat')
for (varname in var){
  out=NULL
  for (season in seasons_vec){
    for (h in hemi){
      for (s in sec){
        out=c(out,knit_child("~/data_netcdf/report_test.Rnw"))
      }
    }
  }
}
@

\Sexpr(paste(out,collapse='\n'))

but I'm getting the following error, probably due to the fact that I have multiple levels of loops which are all using the same chunk label names as references for the figures:

Error in parse_block(g[-1], g[1], params.src) : 
  duplicate label 'cdf-plots'
Calls: knit ... process_file -> split_file -> lapply -> FUN -> parse_block

Can someone please explain this error message? Do I need to somehow alter the label names so that it differentiates between each time it loops? Or should I implement the 3 inner loops within the template file? Or, perhaps combining all of the .tex files after the fact?

I would appreciate help in figuring out the best workflow for this.

You do need to differentiate the label names so that they are different every time it loops. Is this helping you http://stackoverflow.com/questions/12095113/r-knitr-possible-to-programmatically-modify-chunk-labels ? — Tutuchan, Aug 16 '16 at 09:11
I've used `knit_expand` to successfully evaluate a `prefix` variable that I set within the loop, but I can't figure out how I'd combine that with `knit_child` (or if I should be even using `knit_child`). Any thoughts? — pianoplunkster, Aug 17 '16 at 20:02
Looks like you found the solution! So as you saw, you don't need `knit_child` if you use `knit_expand`. — Tutuchan, Aug 18 '16 at 08:00

pianoplunkster · Accepted Answer · 2016-08-17T23:13:01.110

I have come up with a solution! Some notes:

1) the template file cannot have the \begin{document}...\end{document} tags or anything in the preamble, this must be in the main .Rnw file.

2) As of this moment, it only works for a single variable (rather than looping over the vector of variables), but it's a trivial matter to scale this up.

So here is an example of the template:

<<{{prefix}}-setup>>=
varname='{{varname}}'
season='{{season}}'
h='{{h}}'
s='{{s}}'
@

First, a plot of the CDFs, with Kolmogorov-Smirnov statistics:
\begin{figure}[h]
\centering
<<'{{prefix}}-cdf-plots',dev='png',fig.lp='{{prefix}}',out.width='0.5\\linewidth',echo=FALSE,warning=FALSE,fig.align='centering'>>=
colvec=c("blue","red","green","purple","pink")
chart_stats(varname,data_vec,labs,season,s,h,colvec=colvec,cdf_plot=TRUE)
cap1=paste('Cumulative distribution functions for the counts,',season,s,h)
@
\caption{\Sexpr{cap1}}
\end{figure}

And then the main document:

\documentclass{article}
\usepackage[margin=0.5in]{geometry}
\begin{document}

<<test-main,include=FALSE>>=
library(knitr)
setwd("~/data_netcdf")
load("loaded_data.Rdata")
source("~/tempestextremes/test/charts_stats.R")
[other stuff]
@
...
<<generate-code,echo=FALSE>>=
  varname='centlat'
  out=NULL
  for (season in seasons_vec){
    for (h in hemi){
      for (s in sec){
        prefix=paste(season,h,s,sep="_")
        out=c(out,knit_expand("~/data_netcdf/report_test_1.Rnw"))
      }
    }
  }
@

\Sexpr{paste(knit(text=out),collapse='\n')}
\end{document}

I should note that the inspiration came from the knitr-examples section, specifically this one.

Knitr: Use nested loops to generate multiple reports

1 Answers1