That sounds super cool! Do you have any examples that I can see online? If not, I'm wondering the brief idea behind these dynamic exams. Thanks!

Originally posted on 2012-03-02 20:28:48

Guest *Kihoro* @ 2012-03-22 18:47:32 originally posted:

System not yet hosted but the concept is simple.

Write all questions (with their answers/solutions) as child documents and have the main file to input them only if knitr code allows (random or systematic) selection. Knitting allows not manages the file/question selection but also sweaves the child files before a  standard exam draft with appended marking scheme thanks to answers.sty package.

Two questions

1. With the appended exams it would be nenessary to access the lyx/latex file containing the solutions. Is this possible. In other words, how can one access lyx temporary files immediately after compilation.

2. With Knitr, character zero (0) is displayed in scientific notation. How can I alter this?

2 is a bug and has been fixed in the latest version on CRAN; for 1, take a look at the object .cmdargs when you compile the lyx document -- it contains information about the temp directory (e.g. dirname(.cmdargs[1]) might be what you want).

Originally posted on 2012-03-22 20:01:12

1. I guess you mean the TeX ERT (Ctrl+L)  is the framed chunk, and the Chunk ERT is the unframed chunk ( https://github.com/downloads/yihui/lyx/sweave.pdf ). The former is more stable because it is a whole unit and the Chunk ERT is not, e.g. you cannot easily paste code into a Chunk ERT, and when you copy and paste a Chunk ERT, you can lose the indentation information. It is strongly recommended to use the TeX ERT only. You can collapse it by right click and "Close Inset".

2. I do not understand your situation, but you can turn off cache by SweaveOpts{cache=FALSE} which is the default of the cache option, so you do not really need to do it.

Originally posted on 2012-04-11 04:00:10

Guest *MartinInChemnitz* @ 2012-04-23 04:13:07 originally posted:

Thank you for the helpful clarification Yihui.

1. I did have ERT-frames in chunk formatted paragraphs that did not evaluate. Now I know what caused the problem.

2. The caching and the overwriting of figures by reused chunks are still my biggest stumbling blocks:
   a) If I make a little change to a chunk in a lyx file that I have compiled, like inserting a line with a "1" at the   end, I need to save the file under a different name or the output of the chunk will remain unchanged in the necxt compilation - in spite of an ERT-box at the very top with the command "SweaveOpts{cache=FALSE}" that you recommended above.  One filename one compilation is my work around routine fight against caching.
   b) I reuse a chunk that displays descriptive statistics for various columns in a data frame.
   like (ERT boxes not shown)
   SweaveOpts{cache=FALSE}
   <>=
   require(ggplot2)
   a<- data.frame(X=c(1,2,3),Y=c(4,5,6),Z=c(7,8,9))
   VarName<- 'Y'
   (at)
   <<displayvar,cache=f,results='asis',echo=f,fig.width=4,fig.height=3>>=
   d <- a[,c('X',VarName)]
   names(d)<-c('x','y')
   ggplot()+geom_point(data=d,mapping=aes(x=x,y=y))
   (at)
   <<2>>=
   VarName<- 'Z'
   (at)
   <<displayvar,cache=f,results='asis',echo=f,fig.width=4,fig.height=3>>=
   (at)

And the second call of the chunk overwrites the figure produced by the first call so that I get twice the same figure.
No idea why I cannot remove the following from this comment:

You need to convert < to < when writing comments; I'm using Disqus and this annoys me a lot (they should be able to automatically encode HTML entities).

For caching, you are recommended to use the ref.label option: http://yihui.name/knitr/demo/reference/

It will not work if you use the same label for two chunks which generate cache and figure files, because these filenames are from the chunk label. The latter will certain overwrite the former if they have the same name.

Originally posted on 2012-04-23 04:30:52

I recommend you to use the UTF-8 encoding for your LyX document; before trying knitr, you need to make sure this LyX document compiles successfully without R code. Besides, Windows is difficult to work with non-ASCII documents.

Originally posted on 2012-04-29 21:29:29

As I said, you need to create a lyx document with Portuguese which compiles correctly, then add the knitr module. I mean you have to make it clear if it is a LaTeX problem or a knitr problem.

Originally posted on 2012-04-30 00:19:12

Guest *Klebyn* @ 2012-04-30 00:31:44 originally posted:

The problem occur just when I load the module knitr... Portuguese compile very fine without module knitr. I make several tests. Thanks again.

I just tested under Ubuntu (Linux), and UTF8-encoded Portuguese worked fine. Windows is often hopeless in terms of encodings. Which version of knitr are you using? If the latest version on Github does not work for you ( https://github.com/yihui/knitr ), I probably will not have a good solution in the very near future.

BTW, does knitr-minimal.Rhtml work for you? http://yihui.name/knitr/demo/minimal/

Originally posted on 2012-04-30 00:42:44

Guest *Klebyn* @ 2012-04-30 01:16:38 originally posted:

 library(knitr)
knit('knitr-minimal.Rnw')it create tex file and compile without errors. My version is knitr_0.4. I make changes: add áéí innormal text and in the title with usepackage[utf8]{inputenc}. Compile without errors again.The problem ought be the processing in LyX. (version 2.0.3-2).Is there any schedule for to create knitr_0.5.zip package and to migrate for the CRAN???

knitr 0.5 has been submitted to CRAN this morning, and if it passes R CMD check, it will be on CRAN in two days.

Originally posted on 2012-04-30 01:25:31

Guest *Klebyn* @ 2012-04-30 01:42:22 originally posted:

 I used the suggestion:library(devtools); install_github('knitr', 'yihui')and I get the version 0.5. I get the same behaviour :-( .

Then it is real hacking time for you -- the R script used to call knitr is lyxknitr.R under the "scripts" directory of your library directory (see Help -- About). Open that script and tweak it to see if you can make it. I guess it may have something to do with options(encoding = ...).

Originally posted on 2012-04-30 01:51:26

Guest *Klebyn* @ 2012-04-30 02:44:25 originally posted:

 o/o/o/

Now it works!!! :-)

I just comment the line (in lyxknitr.R)
####options(encoding = .cmdargs[3])

I don't understand why, but works!

Thanks for your attention and patience

Good to know. I will update the web page accordingly. Thanks!

Originally posted on 2012-04-30 03:01:45

Thanks. That makes sense, and you can set opts_knit$set(progress = FALSE) before you knit() to suppress the progress bar. See http://yihui.name/knitr/options

Originally posted on 2012-05-01 15:09:58

Guest *Bastien Ferland-Raymond* @ 2012-05-01 18:37:24 originally posted:

 Thanks, it works great.  Sorry I didn't notice the option, I still new in sweave/knitr and I not there yet in my self teaching schedule...
Keep up the good work, your package is amazing.

No problem. I know there are too many options :)

Originally posted on 2012-05-01 18:43:39

Guest *Shmil T* @ 2012-05-20 17:33:53 originally posted:

OK, no worries.

I think that what can actually solve the problem is to hook a TeX command which forces LTR in the beginning of the R code, and cancel it (meaning to return to the previous writing direction) in the end of the code. 

As you can see, the whole document is right aligned (which happens only when I use knitr)
I tried to write Pi several times.
Attached is the output of this syntax:

% Preview source code

%% LyX 2.0.3 created this file. For more info, see http://www.lyx.org/.
%% Do not edit unless you really know what you are doing.
documentclass[english,hebrew]{article}
usepackage[T1]{fontenc}
usepackage[latin9,cp1255]{inputenc}
usepackage{fixltx2e}

makeatletter
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Textclass specific LaTeX commands.
usepackage{theorem}
theorembodyfont{upshape}
newtheorem{theorem}{R{משפט}}[section]
AtBeginDocument{make@lrthetheorem}

makeatother

usepackage{babel}
begin{document}
שלום, רציתי לומר מהו הערך של המספר האהוב עליי, פי: Sexpr{pi} . ואני
שואל אתכם, האין זה מספר יפה?

<<>>=
pi
@

inputencoding{latin9}Now, lets take a look of an english expression
such as Sexpr{pi}. Ain't it beautiful? %

end{document}

I'm afraid it is difficult for me to help in this case, because I know nothing about Hebrew. Can you provide an example to me? What is your OS and what is the version of your LyX as well as its LaTeX backend?

Originally posted on 2012-05-20 15:55:48

Guest *Shmil T* @ 2012-05-20 16:23:52 originally posted:

Hi, and thanks for the reply!
Using windows 7, Lyx 2.0.3, MikTeX 2.9.Using also the TEWAKED version of knitr (solution taken from other post in this site), as it is the only way to get it working.The attached code gave me 2 words in hebrew and a reversed R output (RTL).FYI, I don't get it to draw plots either.Thanks again!Shmulik% Preview source code

%% LyX 2.0.3 created this file. For more info, see http://www.lyx.org/.
%% Do not edit unless you really know what you are doing.
documentclass[english,hebrew]{article}
usepackage[T1]{fontenc}
usepackage[latin9,cp1255]{inputenc}
usepackage{fixltx2e}

makeatletter
%%%%%%%%%%%%%%%%%%%%%%%%%%%%%% Textclass specific LaTeX commands.
usepackage{theorem}
theorembodyfont{upshape}
newtheorem{theorem}{R{משפט}}[section]
AtBeginDocument{make@lrthetheorem}

makeatother

usepackage{babel}
begin{document}
שלום

<<>>=
1+1
@

%
תודה!
end{document}

I do not even know how to compile a normal Hebrew document with LaTeX; at least pdflatex fails under my Ubuntu system...

Originally posted on 2012-05-20 17:19:38

Guest *Shmil T* @ 2012-05-21 13:33:41 originally posted:

Sorry, I replied to the next thread by mistake :

An Rnw document is basically a LaTeX document, so just write whatever you write with ctex, and make sure the R code is marked up appropriately. If you are not familiar with Sweave, see this example: https://github.com/yihui/knitr/blob/master/inst/examples/knitr-minimal.Rnw

Originally posted on 2012-05-25 02:40:54

Sounds like a pure RStudio problem (so better report to them: http://support.rstudio.org/help/ ). What is your RStudio version? I thought RStudio should be intelligent enough to solve the bib problem; if you could provide a reproducible example, they should be glad to help you.

Originally posted on 2012-06-01 20:40:12

rapport is based on templates, and knitr is based on literate programming; programming allows you to do anything, including using templates. There are just too many features to describe, but I do not know your background, so it is hard to list the points that may interest you most. Anyway, one thing that I can sell to you is that you can easily get started with this R Markdown example in RStudio: https://raw.github.com/yihui/knitr/master/inst/examples/knitr-minimal.Rmd

Originally posted on 2012-06-01 20:36:57

This is a LaTeX question; you can redefine the knitrout environment, and here is an example: https://gist.github.com/2760395#file_knitrout_vspace.rnw

Originally posted on 2012-06-03 13:41:50

Guest *JMol* @ 2012-06-04 18:24:54 originally posted:

Thank you very much. Originally in what file is defined for first time the knitrout environment?. I have been looking for in knitr.sty and other files with out luck.

It was defined as an empty environment which allows you to redefine later: https://github.com/yihui/knitr/blob/master/inst/misc/knitr.sty

Originally posted on 2012-06-04 18:47:37

Guest *JMol* @ 2012-06-04 18:54:54 originally posted:

Thank you again for your great job.

Hi,

This is pretty simple to decide: the two formats have different goals. Markdown aims to be simple, so it has to be very limited; LaTeX is a comprehensive typesetting toolset, which allows you to do anything but requires more efforts too. If you need advanced layouts, you definitely should go with LaTeX.

Originally posted on 2012-06-04 23:29:38

Guest *Chris* @ 2012-06-05 08:01:02 originally posted:

Ok, good. I hope knitr will continue to support LaTex. Thanks again for it.

Absolutely! I worked much harder on LaTeX than Markdown. You can compare the amount of code on them respectively: https://github.com/yihui/knitr/blob/master/R/hooks-md.R and https://github.com/yihui/knitr/blob/master/R/hooks-latex.R There is not much to do with markdown now, and there may still be issues to fix/improve with LaTeX.

Originally posted on 2012-06-05 15:10:40

That is because the output is too wide; you should either make the page margin smaller, or cut the output width somehow (may not be possible with str()).

BTW, you should use size='small' instead of size=small; see http://yihui.name/knitr/options

Originally posted on 2012-06-05 17:55:26

Guest *JMol* @ 2012-06-05 18:33:01 originally posted:

I am a newbie both in R and LaTeX, so, please, excuse my insolence, but is it not possible to find a solution. For example, redefining knitrout environment

ifdefinedknitrout
  renewenvironment{knitrout}{
    parnoindentbegin{minipage}{textwidth}
  }{
    end{minipage}
  }
else
fi

Generating new problems, it resolves partially the overfull problem.

Nevertheless, I can break the line manually.

duplicate with http://stackoverflow.com/q/11042681/559676

Originally posted on 2012-06-15 22:17:29

The line is wrapped because it is too wide. You probably need a smaller width, e.g. options(width=40)

Originally posted on 2012-06-15 21:47:46

Guest *JMol* @ 2012-06-16 09:28:23 originally posted:

Thank you Yihui. Please forgive my stupidity, but I am not able to find the width option both in the package and chunk options.

I said options(width=40), so you should go to the documentation ?options instead of looking for that option in this package; options() is a function in base R.

Originally posted on 2012-06-17 17:37:59

How about install.packages('knitr', dependencies = TRUE, repos = 'http://rforge.net')?

Originally posted on 2012-06-21 13:44:05

see, e.g. http://superuser.com/questions/404060/configuring-notepad-to-run-knitr2pdf

You can write to Andrew directly about this (yes, he is the author of NppToR). Or you can try RStudio, which is well integrated with knitr.

Originally posted on 2012-06-22 04:56:54

Please see if http://yihui.name/knitr/demo/minimal/ answers your questions. Basically you have to let R know where is the computer code and where is the text for human.

For the last question, I recommend you to work under the directory of Rnw files, and do not set the 'output' argument unless you really have to and you understand the consequences well; setting it to a file under another directory will not work since figure files are written to the current working directory.

Originally posted on 2012-06-22 20:39:45

Guest *Jacob Wegelin* @ 2012-07-11 15:47:27 originally posted:

Thanks for your reply.

In the past I have developed R code for a given project within a single function, which I keep editing and enlarging. I place readline() and/or error() calls inside this function, so that the execution will pause or stop before certain not-well-developed (draft) code below it is executed. Is there a way to do something like this in knitr? 

Below a toy example of the kind of function I mean. 

Below that is example.Rnw, which shows that neither readline() nor error() will pause or stop execution. Thus, suppose I made a mistake which generated huge, nonsense output (such as an infinite or huge loop with a cat() command inside it). This would just keep running and pouring text into in a huge example.tex file. Is there a way to pause or stop execution in knitr?    myreadline <-function (mytext = "enter to continue:")
   {
       cat(mytext)
       readline()
   }

fnJunk <-function(){

 First some stuff I want to run

   x<-rnorm(5)
   print(x)

 Now some stuff I don't want to run yet 

   myreadline("Are you ready to do a big simulation?")
   stop("Hey! No you aren't! This is enough!")
   for(thisdog in 1:1e5) {
   #  Some time-consuming calculation ...
      cat(thisdog, "n")
   }
}

%% LyX 2.0.3 created this file.  For more info, see http://www.lyx.org/.%% Do not edit unless you really know what you are doing.documentclass{article}usepackage[sc]{mathpazo}usepackage[T1]{fontenc}usepackage{geometry}geometry{verbose,tmargin=2.5cm,bmargin=2.5cm,lmargin=2.5cm,rmargin=2.5cm}setcounter{secnumdepth}{2}setcounter{tocdepth}{2}usepackage{url}usepackage[unicode=true,pdfusetitle, bookmarks=true,bookmarksnumbered=true,bookmarksopen=true,bookmarksopenlevel=2, breaklinks=false,pdfborder={0 0 1},backref=false,colorlinks=false] {hyperref}hypersetup{ pdfstartview={XYZ null null 1}}usepackage{breakurl}setlength{parindent}{0.0in}setlength{parskip}{0.1in}newcommand{superscript}[1]{ensuremath{^{textrm{#1}}}}newcommand{subscript}[1]{ensuremath{_{textrm{#1}}}}begin{document}<<setup, include="FALSE," cache="FALSE">>=# this is equivalent to SweaveOpts{...}opts_chunk$set(fig.path='figure/minimal-', fig.align='center', fig.show='hold')options(replace.assign=TRUE,width=90)@<<loaddata, echo="TRUE">>=x<-rnorm(5)print(x)myreadline <-function (mytext = "enter to continue:") {    cat(mytext)    readline()}myreadline("Do you really want to continue?")stop("Nooo! This is the start of a long simulation!")#Suppose this was long, not just 1 to 5.for( dog in 1:5 ){cat(dog, "n")}@EnD{document}

见统计之都上的回答。

——原帖发布于 2012-07-03 03:43:02

An awkward way is to leave a blank line after each comment of arguments. All comments are wrapped in the units of paragraphs, so if a comment forms a paragraph, it will not be wrapped with the next/previous comments.

It is possible, though, to strop wrapping on certain types of comments, e.g. #' roxygen comments. If that is a desirable solution, I can implement it in the next version of formatR.

Originally posted on 2012-07-05 18:14:51

Guest *Ross Gayler* @ 2012-07-09 04:04:52 originally posted:

 Thanks, Yihui.  I had tried leaving a blank line and decided I didn't really like the way it looked when rendered.  I also tried adding two spaces at the end of the line (which works in the markdown text but not the embedded R chunks).

I'm not familiar with roxygen so can't really comment on the desirability of your suggestion.  I think the arguments would go something like this: Any mechanism would either be a very specific chunk option in knitr (e.g. respect.R.comment.break=TRUE) or be some sort of markup (which would be more flexible than a chunk option). If it is markup, it would either be some sort of pre-existing markup (e.g. #' roxygen comment) or some new markup that is specific to formatR.  All other things being equal I think I would prefer co-opting some pre-existing markup scheme rather than some new, ad hoc markup. (On the other hand, it probably makes little sense to drag in an isolated fragment of some pre-existing markup scheme, so you may have to support all the roxygen markup, for example.  Then I would be worried that the learning curve has been made even steeper. (That seems like a lot of effort just to get a line break.)

 OK, I think I will just stop wrapping roxygen comments, i.e. comments that begin with #'. I will not check the type of roxygen comments; as you said, that will make things complicated. The simple rules will be: # will be wrapped, and #' will not.

Originally posted on 2012-07-09 16:59:42

Guest *Ross Gayler* @ 2012-07-09 20:41:40 originally posted:

Thanks Yihui.
Ross

Guest *wegelinjacob* @ 2012-07-11 18:31:07 originally posted:

One advantage to errors (calls to stop()) embedded in the code, is to build in guarantees, along the lines of 

stopifnot( is.numeric(x))

Is there a way to design a *.Rnw file so that it will  refuse to continue if the condition (such as a condition set up by stopifnot() ) is notsatisfied? 

If you have stop() in a function, knitr will not run the code after stop() in that function. You can easily verify this by, say,

f = function() {
  stop('no, stop here')
  for (i in 1:10000) cat(i, 'n')
}
f()

However, if stop() is in a separate expression, the code after it will still be executed because all the code is evaluated with try(). If you want a full stop in that chunk, see https://gist.github.com/2629886#file_knitr_error.rmd

Note there is NO option named 'error' in opts_knit; I have no idea on where you saw it. All possible package options were documented here: http://yihui.name/knitr/options#package_options

readline() is a very different problem; see a discussion here: http://stackoverflow.com/q/11248705/559676

In your case, I would suggest you use a global variable to control your code rather than readline(), e.g. run = TRUE/FALSE, which is a better programming practice, I believe; there is no guarantee for the output of readline() but global constants are easier to control and have guaranteed values.

Originally posted on 2012-07-11 20:35:08

Guest *wegelinjacob* @ 2012-07-11 23:06:43 originally posted:

Dear Yihui,

Thank you for your reply and the link on how to make knitr completely stop in case of an error.
But is the code on that page an example of an error that stops knitr? Is the code
x<-ff(1:10)
print(x)

supposed to return an error?

Is this the "ff" function which is in the "ff" library? I loaded that package and ran ff(1:10) and it did not return an error. What would ff(1:10) have to do with causing knitr to stop?
Thank you also for the link regarding conceptual incompatibility between readline() and knitr.
Jake

That is just a coincidence; ff was not supposed to be found because it was not defined in the document. All my examples are self-contained; if I need add-on packages, I will write library() explicitly in the source document.

Originally posted on 2012-07-12 00:56:49

knitr will not stop no matter if error=TRUE or FALSE; the only way to stop it is showed in this example via an output hook: https://gist.github.com/2629886#file_knitr_error.rmd

NOTE: 'error' is NOT an option in opts_knit; it is in opts_chunk. In that page, there are two objects opts_chunk and opts_knit, one controlling chunks, the other controlling package options.

Originally posted on 2012-07-11 21:28:03

Guest *wegelinjacob* @ 2012-07-13 16:46:43 originally posted:

Dear Yihui:

Thank you for this information.

I believe the following is common practice for programmers building code: to step through code; or, more crudely, to run code up to a certain point and have it either pause or die at that point. Then when the programmer is confident that the results are good up to that point, he/she moves the "die point" further down the code and tests it again.
I do not mean that the final code would behave like this, but that the programmer uses this technique when building it.

This is my object in asking about a way to make knitr stop or pause.

At http://jacobwegelin.net/tmp/XieYihui/ is an example, 2012-07-13Fri12h30m-a.Rnw, where I tested the

knit_hooks$set(error = function(x, options) stop(x))

approach.

This approach presents two difficulties.

(1) The resulting 2012-07-13Fri12h30m-a.tex lacks an end{document} line, so that it cannot be compiled in LaTeX without being edited first.

(2) More difficult, 2012-07-13Fri12h30m-a.tex also lacks a definition of the knitrout environment.

In fact, it lacks even more than this, as I discovered by trying the following workarounds:

(1) First I added by hand end{document} to the end of 2012-07-13Fri12h30m-a.tex.
(2) I added by hand

newenvironment{knitrout}{}{} % an empty environment to be redefined in TeX
to 2012-07-13Fri12h30m-a.Rnw.

But then I found also that definecolor was missing from 2012-07-13Fri12h30m-a.tex.
Ideally, one could place a command inside the *.Rnw so that the code below that command would not be run, but the code up to that command would run; and so that one could compile *.tex, i.e., see the results up to that point.

I use that practice continually.

I understand that in the final product, one does not want the *.Rnw to be interactive. But when building the code, I want to interact with it and, at every step, see how well I've done so far.

Do you see my point?

Maybe there is some other way to do this in the context of knitr, other than by putting an error or readline() inside the code?

Thanks again. I'd be interested in your thoughts on this.

Jake

Guest *wegelinjacob* @ 2012-07-13 17:47:11 originally posted:

I think the solution is to add "eval=FALSE" to the code chunks that one does not want to run yet. An example is posted under today's date at http://jacobwegelin.net/tmp/XieYihui/.
---------- Forwarded message ----------
Date: Fri, 13 Jul 2012 12:46:32

When you throw a real error during the compilation, the result tex document is not supposed to be a complete document (because the process was interrupted), and I guess it is not surprising that some stuff like end{document} or the preamble are missing.

Instead of being as "rude" as asking the program to die by knit_hooks$set(error = function(x, options) stop(x)), I'd rather suggest you ask the program to skip the evaluation of the rest of chunks by setting opts_chunk$set(eval=FALSE) in the chunk after which you want the rest of the code not to be evaluated any more. This guarantees the process finishes normally and quits gracefully after a certain chunk. Otherwise you have to wrestle with a broken output document endlessly.

If you want to interact with your code, you just need a decent editor like Emacs/ESS or RStudio, both of which recognize code chunks in an Rnw document and support running code chunks in an interactive R session. Even if you do not use these editors, there is still another approach to interact with code: http://yihui.name/knitr/demo/externalization/ You can write all your pure R code in a separate script, in which you are free to explore/diagnose your code; after you are done, just add a label to the code and reference it in your Rnw document.

Originally posted on 2012-07-13 17:50:21

See my reply below: http://yihui.name/knitr/#comment-587156836

Originally posted on 2012-07-13 17:58:15

Guest *wegelinjacob* @ 2012-07-13 18:23:03 originally posted:

From the new example I just placed at http://jacobwegelin.net/tmp/XieYihui/ under "2012-07-13Fri14h03m example", it appears that putting
opts_chunk$set(eval=FALSE)

anywhere inside a chunk has the following effect:

(1) if eval=TRUE is specified in the chunk header, then eval remains true in spite of the opts_chunk$set(eval=FALSE) statement.
(2) if eval is left at the default value, then

opts_chunk$set(eval=FALSE)

turns off evaluation for the entire chunk, not only the code after the opts_chunk statement.
Is this correct?

Thus, the code

opts_chunk$set(eval=FALSE)

seems no more flexible than simply putting "eval=TRUE/FALSE" into the header. Is this correct?

opts_chunk sets options globally in the whole document after that chunk, whereas options in the chunk header are local

There is no way to set an option in the middle of a chunk and let it affect the rest of the code in the chunk; all the code in a chunk is evaluated as a whole unit. If you want to change options in the middle, you have to separate the chunk into two.

Originally posted on 2012-07-13 19:25:26

I have never seen that happens. Do you have a reproducible example? What is your sessionInfo() after library(knitr)?

Originally posted on 2012-07-12 16:38:17

Guest *wegelinjacob* @ 2012-07-13 19:53:14 originally posted:

Dear Yihui,

Thank you for your reply re: knitr hanging.

You suggested I send a reproducible example.

I cannot post the example publicly on the Web, because it depends on a dataset that has not been published. But if you would like to look at an example, I could show it to you individually.
The behavior is sporadic, but frequent enough to slow down my work and be a real annoyance. The same file might run on one occasion and then not run on another.
Just now, for instance, I got "try.Rnw" to run many times, and then it froze up again. Could this be caused by a memory leak? It tends to work for a while, then fail; then if I kill R and restart R, it works again.
Here is the sessionInfo() that you requested:

First I recommend you to update to the latest version of knitr (better to the development version: https://github.com/yihui/knitr#readme)

If the same thing still occurs, try to use some random data instead of the real data to make a reproducible example.

I do not know how you run knitr; ideally you should run it in a new R session each time instead of calling knit() in the same R session again and again.

Originally posted on 2012-07-13 20:22:39

Guest *Ikuyasu* @ 2012-07-12 21:53:46 originally posted:

I believe that that's the fastest way to publish online through blog system (blogger, wordpress etc) after creating a statistical report in R. If there is a way to link a post on Rpub to my blog entry, then this will be the fastest since Rstudio has a botton just to click to publish on Rpub.

To my understanding, base64 encoded images are just another type of representation of the original images, so there should not be quality loss. Do you have an example PNG that looks different when encoded?

Originally posted on 2012-07-12 23:46:40

I have seen quiet a few people trying to publish blog posts with knitr; you can search for "wordpress" or "blog" in the page http://yihui.name/knitr/demo/showcase/

For RPubs, you need to contact RStudio developers. I'm not the right person to talk to. You can, for example, suggest them make RPubs embeddable in other sites.

Originally posted on 2012-07-12 23:52:34

Guest *ikuyasu* @ 2012-07-13 21:14:00 originally posted:

Thank you for pointing out the direction that I can look at. I will keep you updated on what I find!

Guest *Loyolite270* @ 2012-07-23 10:10:47 originally posted:

 The problem is resolved, i just updated the knitr package in R. Thank you very much for this wonderful package

OK, I saw it on StackOverflow as well: http://tex.stackexchange.com/q/65497/9128

Originally posted on 2012-08-04 23:34:18

which problem are you talking about?

Originally posted on 2012-08-04 23:38:56

用results='asis'，因为你生成的是HTML代码。参见手册：http://cran.r-project.org/web/packages/googleVis/vignettes/googleVis.pdf

——原帖发布于 2012-08-13 16:29:33

Sounds like an RStudio problem: http://groups.google.com/group/knitr/browse_thread/thread/e1d6d71220bc5d06

Originally posted on 2012-08-26 01:17:05

Great question. This part (foreign language engines) is still not mature in knitr, and I need to think about a more general design. For the time being, it is simply a system() call to python, so R picks whatever version that comes first in the PATH variable. If you want one specific version of python to be used, you can probably use Sys.setenv(PATH = "change.the.order.of.paths").

I'll consider generalizing the framework in the next version so it will be easier to configure the engines. That being said, you can always reset the default engines, e.g. knit_engines$set(python = your_own_function); see https://github.com/yihui/knitr/blob/master/R/engine.R

Originally posted on 2012-09-05 21:09:54

1. 能检查一下你的TeX源代码究竟有什么问题吗？
2. 消息栏里没有任何消息？View-->View Messages

——原帖发布于 2012-09-11 21:40:35

访客 *Henry Shang* @ 2012-09-16 21:35:20 写道：

 我相信是LyX的问题。View-->View Messages里面仅仅提示预览pdf4时发生错误。经反复升降级Texlive，发现lyx问题越来越多了，甚至找不到xetex了。最后将~/.lyx文件夹删除即OK。显然LyX的重新配置命令存在缺陷。
最后给点鼓励的话，大家使用knitr感到很顺手，希望你继续努力！有这么給力的年轻人，不愁拿不下钓鱼岛。加油！

You can probably try to move the xcolor options before documentclass, e.g.

PassOptionsToPackage{svgnames,table}{xcolor}
documentclass[trans]{beamer}

Originally posted on 2012-09-12 18:25:43

Read the manual please (Section 3.1): https://github.com/downloads/yihui/knitr/knitr-manual.pdf (also FAQ6 https://github.com/yihui/knitr/blob/master/FAQ.md)

Originally posted on 2012-09-13 16:26:29

Sorry I cannot reproduce the problem. Attached is what I got from your code with the latest versions of all packages: http://i.imgur.com/BOPFZ.png

Originally posted on 2012-09-21 21:38:17

Guest *mat* @ 2012-09-24 14:16:35 originally posted:

 Yihui,
Thank you for looking at this. I loaded mgcv so our sessionInfo() call have the same output but I'm still not getting the nice output in your .png

Could the header on the .Rnw cause this effect? I'm using this one (amended from Coombes and Baggerly at MDAnderson).

% HEADER
documentclass{article}  % Use the standard article style by default
usepackage{graphicx}    %,color,fancyvrb}    % So we can include graphics easily
pagestyle{myheadings}   % put our own headings on all but the first page

% Reset the page size to use more of the paper
setlength{topmargin}{0in}
setlength{textheight}{8in}
setlength{textwidth}{6.5in}
setlength{oddsidemargin}{0in}
setlength{evensidemargin}{0in}

% meta-data
title{Learning ggplot}
markright{ggplot ch02.Rnw}   % this.Rnw NB, does not tolerate underscore use
author{mat}
date{21 sep 2012}

begin{document}
SweaveOpts{concordance=TRUE}
maketitle

<>=
opts_chunk$set(tidy = FALSE)
@f1ecdb63bdeec2a6b1134e2acf417c9b:disqus

Should I take this to stackoverflow?

Guest *mat* @ 2012-09-24 14:18:47 originally posted:

 I have no idea how that link icon got into the setopts chunk! (to clarify, it's not in the original .Rnw)
Thanks
Mat

Please post source code to pastebin.com or elsewhere and paste the link here. Disqus really sucks when dealing with program code. It has been driving me crazy.

Originally posted on 2012-09-24 16:45:19

You can pass arguments to R or Rscript directly as well, e.g.

R -e 'commandArgs()' --args a b c
Rscript -e 'commandArgs()' --args a b c
Rscript foobar.R --args a b c

Originally posted on 2012-10-01 00:14:39

Guest *Levi Waldron* @ 2012-10-01 02:51:54 originally posted:

 Thanks for your prompt reply, Yihui!  This is very convenient that the arguments passed to the knitr-invoking script are then just available to the .Rnw file.  Your suggestions do the trick, and also it is no problem to use optparse with either positional or flag arguments, for the knitr-invoking script.  Here is an example of a generic knitr-invoking script I am using, with a combination of flag arguments (for knitr input/output files) and positional arguments (passed on the the tangled R script). https://dl.dropbox.com/u/15152544/knit.R.  It is based on the script provided by http://www.tomtorsneyweir.com/research-notebook.  For dedicated scripts it may be more desirable to use all flag arguments, but this seems like a useful script to put in /usr/local/bin.

Thanks for sharing the script. I have a simplified version in the package as well: https://github.com/yihui/knitr/tree/master/inst/bin but I saw you said texi2pdf() did not work for you, so probably this script will not work for you either.

Originally posted on 2012-10-01 20:07:13

Perhaps these two examples can help: https://gist.github.com/2760395

Originally posted on 2012-10-04 16:36:40

Guest *David Orme* @ 2012-10-04 17:14:17 originally posted:

That's very helpful, thanks, and thanks for such a quick response. Sorry I didn't spot the examples. 

That provides a simple way of cutting down the space around the verbatim and before and after the whole knitrout chunk. There is still some vertical space between input and output lines that remains that doesn't seem to be affected by those lengths. It has to be something that flushleft triggers - if I replace everything in the flushleft environment with simple text, the space around it remains. 

The flushleft environment should not come unless you are using an old version of knitr, or have explicitly set opts_knit$set(use.highlight = TRUE).

Originally posted on 2012-10-04 18:11:25

1. I cannot reproduce it; do you have a space after ###?
2. That is because the number was represented in scientific notation; I think you can use r I(knitr:::format_sci(msft.VaRq01, format="latex")) if you want scientific notation, otherwise just write r I(as.character(msft.VaRq01))

Originally posted on 2012-10-20 19:52:59

Guest *Chuck Up* @ 2012-10-21 13:10:27 originally posted:

Thanks for the quick reply.

For 1, putting a space corrected the issue, but this has not been required for other Markdown documents.

For 2, your suggestion worked.

1. I see this is required according to the markdown documentation: http://daringfireball.net/projects/markdown/syntax#header

Originally posted on 2012-10-21 17:34:39

Guest *Chuck Up* @ 2012-10-21 18:15:12 originally posted:

Technically, it is not specified. Markdown's # headers are based on atx, which is defined here: http://www.aaronsw.com/2002/atx/intro

Well, first you should look at Markdown's syntax instead of Atx, and Markdown has showed examples where you can see the spaces after the hashes; secondly, even if you insist on looking at Atx, see line 47 of its source code http://www.aaronsw.com/2002/atx/atx.py and it will be clear that a space is required. BTW, I'm not the right person to argue with in terms of the Markdown syntax; that is not something I invented.

Originally posted on 2012-10-21 19:16:34

Duplicate with yihui/knitr#413

Originally posted on 2012-10-29 20:21:13

Thanks for the report. I kind of expected that. In this case, recordPlot() can no longer record the plot as displaylist because it is too big. Before I start thinking of a solution, may I ask if you really need such a huge plot? I mean scatter plots of a huge number of points are often not useful because you cannot really see anything in the plots.

Originally posted on 2012-11-02 16:16:34

Guest *Jinsong Zheng* @ 2012-11-05 13:04:34 originally posted:

Many thanks for your reply.  I use pair plot to check the correlation structure with low panel of correlation value  and upper panel of contour plots.  It occurs the same problem. I use  scatter plots  here as an illustration to show that there may be a problem by using pairs().  Anyway, I can run this part in sweave or R-console.  Thanks again.

I just wrote an example for this case which works fine for me: https://github.com/yihui/knitr-examples/blob/master/056-huge-plot.Rmd

Originally posted on 2012-11-05 19:31:01

Does figure/r-logo.png exist at all? I guess not.

Originally posted on 2012-11-15 14:57:19

1. Yes, they do need LaTeX, but I guess installing MikTeX is easy, and it would be much more painful if they do not know LaTeX at all;
2. You can use R Markdown (.Rmd) to generate a Markdown document (.md), then use pandoc to convert it to many other formats including Word;

Originally posted on 2012-11-15 15:00:54

Guest *Mons Magnusson* @ 2012-11-30 09:27:16 originally posted:

Hi Yihui!

I use knitr in R-Studio and pandoc. I use pandor to produce pandoc markdown tables in R-studio but it does only seems to work with simple pandoc tables and not "grid" tables for example, is there something I've done wrong or is it simply not implemented in knitr/Rmarkdown yet?

Regards
Måns

It is a matter of syntax. The biggest pain of markdown is that there are tons of variants of markdown. For now, you have to stick to one of them, either pandoc's markdown, or R's markdown.

Originally posted on 2012-11-30 16:29:37

Yes, please update R; knitr requires R >= 2.14.1. Note the current version of R is 2.15.2.

https://github.com/yihui/knitr/blob/master/FAQ.md

Originally posted on 2012-11-15 17:32:43

Guest *Aureincm* @ 2012-11-15 23:12:10 originally posted:

Silly me.....
Thanks for you QUICK answer!!!  Everything seems to be working fine now!!
Can't wait to use your package!!! And as a knitter.. I  have to say, I LOVE its name ;-)
Thanks 
 - Aure

You are not the only real knitter who uses knitr, and I'm always happy to know that this is a neat name for real knitters. Perhaps someone can draw knitting patterns using R some day :)

Originally posted on 2012-11-16 01:29:32

It is really a challenging problem but not impossible. Here is my example:  https://github.com/yihui/knitr-examples/blob/master/064-bib-plot.Rnw

Note you need to install the tikzDevice package from R-Forge.

Originally posted on 2012-11-20 20:18:10

Guest *ADPowers* @ 2012-11-21 00:56:06 originally posted:

Fantastic. That works perfectly, also needed to install the R filehash package in order to get it to work.

I'm getting less issues with image scaling too now (better consistency), since the dev='tikz' package is being used.

Is there a way to set the fig.height and fig.width values as default values, once at the beginning of the document, so that I can keep the same across multiple plots on different pages without having to specify on a individual plot basis?

This package is fantastic, it is going to save me so much time on the whole.....

Cheers.

See http://yihui.name/knitr/options; use opts_chunk$set()

Originally posted on 2012-11-21 02:27:45

Guest *ADPowers* @ 2012-11-21 02:39:25 originally posted:

'cite{xyz}' works perfectly but if I use 'citep{xyz}' or 'citet{xyz}' as per natbib, it throws error. Any ideas?

You need to add natbib to tikz metric packages; see ?tikzDevice

For example,

library(tikzDevice)
options(tikzMetricPackages = c(getOption('tikzMetricPackages'),
'usepackage[authoryear]{natbib}'))

Originally posted on 2012-11-21 04:17:36

What does that mean?

Originally posted on 2012-11-21 02:29:28

Guest *ADPowers* @ 2012-11-21 04:23:45 originally posted:

I accidentally posted a new post, when what I wanted to do was reply to an old one. There didn't seem to be a button to delete the error/mistake post, only edit what was already submitted provided one doesn't simply delete the entire paragraph leaving an empty post, so I had to type something trivial and re-submit where it was supposed to go.

use out.width and out.height instead; see http://yihui.name/knitr/options again

Originally posted on 2012-11-21 05:24:49

Guest *ADPowers* @ 2012-11-28 23:26:42 originally posted:

Hi Yihui,

out.width and out.height aren't doing what is expected. I have read the link above, but cant seem to get it to work.

defmywidth{500px}
defmyheight{400px}
<<defaultfiguresizes,eval=true,echo=false>>=
W <- 'mywidth'
H <- 'myheight'
opts_chunk$set(out.width=W, out.height=H)
@

It i use fig.width or fig.height, with explicit numerical values INSIDE the block, works fine, but the above has no effect. Can you see anything obvious that I am doing wrong?

Cheers.

It does work for me. Can you take a look at the tex file? If you can see width=mywidth,height=myheight after includegraphics, it is your LaTeX problem instead of knitr's. And you need to keep in mind that the unit px may not be supported by LaTeX; Google search gives me this http://tex.stackexchange.com/a/41371/9128

Originally posted on 2012-11-29 03:18:17

Guest *ADPowers* @ 2012-11-29 03:39:05 originally posted:

There is no includegraphics in the 'tex' file. Below is the relevant part. This TEX file comes from the relevant .Rnw file, which contains the actual code, being a call to R to plot something. The object DOES plot, but the out.width and out.height have apparent bearing on the size as it appears in the final typset .pdf. If I use fig.width and fig.height the plot gets inserted at the right size.

begin{knitrout}
definecolor{shadecolor}{rgb}{0.969, 0.969, 0.969}input{figure/MYPLOT.tikz}
end{knitrout}

Guest *ADPowers* @ 2012-11-29 03:40:03 originally posted:

*In the below post meant to say *NO apparent bearing....

So your example code did not tell me you were using the tikz device, which makes this a completely different story. Tikz graphics cannot be rescaled in the usual way. Search for resize.width in http://yihui.name/knitr/options

Originally posted on 2012-11-29 20:22:18

Guest *ADPowers* @ 2012-11-29 23:18:13 originally posted:

Apologies, should have made made that clear. Based on my other post, where you suggested to use library(tikzDevice) to get citations/references into the plot area, I assumed that was understood.

resize.width and resize.height screw up the scaling.

out.width and out.height do a perfect job, I just cant use declared variables outside of the scope of a chunk, declared elsewhere in the LaTeX document, as per the above post.

Why use the equation environment? I mean you can change the legend labels (write $$ there); this trick was extensively used in the book "The Analysis of Data", e.g. http://theanalysisofdata.com/probability/2_2.html

Originally posted on 2012-11-21 17:15:43

Guest *ADPowers* @ 2012-11-22 00:51:50 originally posted:

Thanks. Feel a bit dumb but Works perfectly.

You are right. This is a long history (WEB, Noweb, weave, tangle, literate programming, Sweave = S + weave). As time goes by, people will never understand knitr is neater than "what" and there will be packages neater than knitr :)

Originally posted on 2012-11-21 17:00:26

Guest *ADPowers* @ 2012-11-21 17:04:47 originally posted:

Is it appropriate to make some reference to string theory?

Guest *ADPowers* @ 2012-11-21 17:06:06 originally posted:

People should just 'cotton-on'