Help getting two R plots into a Beamer slide

Question

I am trying to develop slides for a course I am teaching in Org mode compiling to a latex beamer presentation. Overall most things are working fine. But for some reason I am having trouble getting two R plots aligned. Here is how the relevant slide looks currently. The problem should be obvious.

Here is the code for the presentation:

#+TITLE:     my class
#+AUTHOR:    me
#+EMAIL:     [email protected]
#+BEAMER_HEADER: \institute[]{Penn State University} 
#+DATE:      
#+DESCRIPTION: 
#+KEYWORDS: 
#+LANGUAGE:  en
#+BEAMER_THEME: Boadilla
#+BEAMER_HEADER: 
#+latex_class: beamer
#+latex_compiler: pdflatex
#+LaTeX_HEADER: \usepackage{pdfpages}
#+OPTIONS: H:2 toc:t
#+BEAMER_HEADER: \AtBeginSection[]{\begin{frame}<beamer>\vfill\centering\usebeamerfont{title}\insertsectionhead\vfill\end{frame}}
#+BEAMER_HEADER: \usepackage{minted}
#+BEAMER_HEADER: \usemintedstyle{emacs}
#+STARTUP: beamer


#+begin_src R :exports none 
  library(tidyverse) 
  set.seed(123)
#+end_src

* Bayesian Statistics

** Metropolis Algorithm Example
\begin{align*}
y_i &\sim N(\beta x_i, \sigma^2) \\
\beta &\sim N(\mu, \tau^2)
\end{align*}
*** XXX                                                                                    :BMCOL:
:PROPERTIES:
:BEAMER_col: .55
:END:
#+begin_src R :exports code :session "*R*"
# Simulate some data
n <- 100 
beta <- 4.1
sigma2 <- 1.3
x <- rnorm(n)
y <- rnorm(n, mean=beta*x, 
              sd=sqrt(sigma2))
#+end_src

*** XXX                                                                                    :BMCOL:
:PROPERTIES:
:BEAMER_col: .45
:END:
#+begin_src R :exports results :session "*R*" :results graphics file :file figures/foo.pdf
plot(x,y)
#+end_src

#+RESULTS:
**  Metropolis Algorithm Example
\begin{align*}
y_i &\sim N(\beta x_i, \sigma^2) \\
\beta &\sim N(\mu, \tau^2)
\end{align*}

\tiny
#+begin_src R :exports code :session "*R*"
# For the moment we deal only with 
# simple regression (x is univariate)
log_p_tilde <- function(beta, y, x, sigma2, mu, tau2){
  loglik <- dnorm(beta, mean=mu, 
                        sd=sqrt(tau2), 
                        log = TRUE)
  loglik <- loglik + 
            sum(dnorm(y, mean=beta*x, 
                         sd=sqrt(sigma2), 
                         log=TRUE))
  return(loglik)
}

# Now the proposal function J(theta | theta^(s))
proposal <- function(beta_s, delta2){
  return( rnorm(1, mean=beta_s, sd=delta2) )
}
#+end_src
\normalsize

** Metropolis Algorithm Example
*** XXX                                                                                    :BMCOL:
:PROPERTIES:
:BEAMER_col: .5
:END:
\tiny
#+begin_src R :exports code :session "*R*" :results silent
n_samples <- 4000
beta <- rep(NA, n_samples)
# Initialize chain
beta[1] <- 0
# Create partial function to make code easier to read
lpt <- function(beta) log_p_tilde(beta, y, x, 1.3, 
                                  # Choose priors
                                  mu = 0, tau2 = 1)
# Run MCMC
for (s in 2:(n_samples+1)){
  # Propose
  beta_prop <- proposal(beta[s-1], delta2 = .1)
  
  # Compute Acceptance Ratio (log-scale)
  r <- lpt(beta_prop) - lpt(beta[s-1])
  
  # Sample uniform
  u <- runif(1, min=0, max=1)
  
  # Accept/Reject
  if (u <= exp(r)){
    beta[s] <- beta_prop
  } else {
    beta[s] <- beta[s-1]
  }
}
# Drop "Burn-in" rid of initial sample which we chose
beta <- beta[-(1:n_samples/2)]
#+end_src
\normalsize
*** XXX                                                                                    :BMCOL:
:PROPERTIES:
:BEAMER_col: .5
:END:
#+begin_src R :exports results :session "*R*" :results graphics file :file figures/beta-posterior-trace.pdf :width 3 :height 3
plot(beta)
#+end_src

#+begin_src R :exports results :session "*R*" :results graphics file :file figures/beta-posterior-density.pdf :width 3 :height 3
plot(density(beta))
#+end_src

I would greatly appreciate any help in debugging this issue!

EDIT: After reading this post Change emacs generated image size in org-mode export, I came up with the following solution (which does not work):

#+name: beta-posterior-trace
#+begin_src R :session "*R*" :file figures/beta-posterior-trace.pdf 
plot(beta)
#+end_src

#+ATTR_LATEX: :width 0.3\linewidth
#+results: beta-posterior-trace
[[file:figures/beta-posterior-trace]]

#+name: beta-posterior-density
#+begin_src R :session "*R*" :file figures/beta-posterior-density.pdf
plot(density(beta))
#+end_src

#+ATTR_LATEX: :width 0.3\linewidth
#+results: beta-posterior-density
[[file:figures/beta-posterior-density]]

Here is how that ends up looking:

Answer 1

Checking the produced .tex file shows that the two plots on the last page are produced by this code:

\begin{column}{.5\columnwidth}
\begin{center}
\includegraphics[width=0.9\linewidth]{figures7/beta-posterior-trace.pdf}
\end{center}


\begin{center}
\includegraphics[width=0.9\linewidth]{figures7/beta-posterior-density.pdf}
\end{center}
\end{column}

A little trial-and-error tweaking of the .tex file shows that width=0.6\linewidth would make both figures small enough to fit in the RH column.

There is a setting, org-latex-image-default-width, whose default value is 0.9\textwidth so you could change it. The trouble with that is that all figures would be squeezed a bit.

The standard way of changing the width of one figure for LaTeX export is to add a #+ATTR_LATEX: :width ... directive just before the image link. So if we had an image that pre-existed, we could do

#+ATTR_LATEX: :width 0.6\linewidth
[[file:figures7/beta-posterior-density.pdf]]

and that would propagate the value into the resulting \includegraphics. However, here we have plots that are dynamically produced, so we don't have the link to the figure to attach the #+ATTR_LATEX line to.

The solution is in three parts:

• Give a name to each of the two code blocks with #+name.
• Run each code block by hand with C-c C-c to produce a labeled #+RESULTS line (and a link to the image file but that is not the main point here).
• Attach the #+ATTR_LATEX line to the #+RESULTS line and delete the links.

It will then look like this:

#+name: beta-plot
#+begin_src R :exports results :session "*R*" :results graphics file :file figures7/beta-posterior-trace.pdf :width 3 :height 3
plot(beta)
#+end_src

#+ATTR_LATEX: :width 0.6\linewidth
#+RESULTS: beta-plot


#+name: beta-density-plot
#+begin_src R :exports results :session "*R*" :results graphics file :file figures7/beta-posterior-density.pdf :width 3 :height 3
plot(density(beta))
#+end_src

#+ATTR_LATEX: :width 0.6\linewidth
#+RESULTS: beta-density-plot

When you export, the export happens in a temporary buffer containing a copy of the Org mode file. At that time, the code blocks will be executed and they will produce the image files (if they don't exist already), and also add the figure links after the corresponding #+RESULTS lines. That's why we had to name the code blocks: the labeled #+RESULTS lines act as place holders so that the links to the images would be put right after them (remember that's done in the temprorary copy), and the #+ATTR_LATEX lines are attached to the links (think of them as diffusing through the #+RESULTS lines). Without the names, this would not work. The export happens, the temporary copy disappears and your file looks exactly the same as before: you only have to do the manual stuff once.

Answer 2

Editing the code on the relevant slide to the following seems to have worked.

#+begin_src R :exports results :session "*R*" :file figures/beta-posterior-trace.pdf 
plot(beta)
#+end_src

#+ATTR_LATEX: :width 0.65\linewidth
[[file:figures/beta-posterior-trace.pdf]]

#+begin_src R :exports results :session "*R*" :file figures/beta-posterior-density.pdf
plot(density(beta))
#+end_src

#+ATTR_LATEX: :width 0.65\linewidth
[[file:figures/beta-posterior-density.pdf]]

That said, I am not exactly sure why this particular "incantation" worked but just adding #+ATTR_LATEX: :width 0.65\linewidth above the src block directly didnt...

I am also not sure why I needed the :exports results directive but I didnt need a :results none or :results file directive