Create sections through a loop with knitr

Question

See this reproducible example :

---
title: "test"
output: html_document
---

## foo

```{r}
plot(1:3)
```

## bar

```{r}
plot(4:7)
```

## baz

```{r}
plot(8:12)
```

I want to be able to automate the creation of these sections as I can't know how many they will be before going further in my analysis.

My input to get this would be :

my_list   <- list(foo = 1:3, bar = 4:7, baz = 8:12)
my_fun    <- plot
my_depth  <- 2

And the ideal answer (though I'm welcoming any improvement) would help me build a mdapply function so that I could just run:

```{r}
mdapply(X = my_list, FUN = my_fun, title_depth = my_depth)
```

And get the same output.

Answer 1

Taking a similar approach to @Georgery... but in a somewhat over-engineered fashion (also somewhat more general?). Anyway, here it goes.

make_template <- function(my_list, my_fun, my_depth, my_title, my_output_type, my_template_file){

  require(glue)

  n <- length(my_list)


  # --- Rmd header ---
  make_header <- function(my_title, my_output_type){
    #
    my_header <- glue(
                      "---", "\n",
                      "title: ", deparse({my_title}), "\n",
                      "output: ", deparse({my_output_type}), "\n",
                      "---", "\n",
                      "\n",
                      "\n"
    )
    return(my_header)
  }

  # --- one section only ---
  make_section <- function(i){
    one_section <- glue(
                        "\n",
                        "\n",
                        paste0(rep("#", times = {my_depth}), collapse = ""), " ", names({my_list})[[i]], "\n",
                        "\n",
                        "```{{r}}", "\n",
                        paste0({my_fun}, "(", deparse({my_list}[[i]]), ")"), "\n",
                        "```", "\n",
                        "\n",
                        "\n"
    )
    return(one_section)
  }


  # --- produce whole template ---

  my_header <- make_header(my_title, my_output_type)

  all_my_sections <- ""
  for (i in seq_along(my_list)) {
    all_my_sections <- paste0(all_my_sections, make_section(i))
  }

  my_template <- paste0(my_header, "\n", "\n", all_my_sections)

  # --- write out
  cat(my_template, file = my_template_file)
}




# --- try it

make_template(my_list = list(foo = 1:3, bar = 4:7, baz = 8:12, glop = 1:7),
              my_fun = "plot",
              my_depth = 4,
              my_title = "super cool title",
              my_output_type = "html_document",
              my_template_file = "my_template_file.Rmd"
)

Answer 2

I would actually suggest a solution that works a little bit different, i.e. create the R-Markdown file from an R-script and then render it from the same R-script:

# function that creates the markdown header
rmd_header <- function(title){
paste0(
"---
title: \"", title, "\"
output: html_document
---
"
)
}

# function that creates the Rmd code for the plots
rmd_plot <- function(my_list, my_fun){
paste0(
"
## ", names(my_list), "

```{r}
", deparse(substitute(my_fun)), "(", deparse(substitute(my_list)), "[[", seq_along(my_list), "]])
```
"
)
}

# your objects
my_list   <- list(foo = 1:3, bar = 4:7, baz = 8:12)
my_fun    <- plot
my_depth  <- 2 # I actually don't get what this is for

# now write everything into an rmd file
cat(rmd_header("Your Title")
    , rmd_plot(my_list, plot)
    , file = "test.rmd")

# and then create the html from that
rmarkdown::render("test.rmd", output_file = "test.html")

One thing to mention here: the indentation in the Rmd file does matter and when you copy the code here, make sure that R-Studio inserts it in the R-script as intended (because often it doesn't).

Answer 3

R package pander can generate Pandoc's markdown on the fly.

The key is to use the chunk option results='asis' to tell R Markdown to render pander's output as Markdown. You just need to be careful to generate valid Markdown!

Try this:

---
title: "Test sections"
output: html_document
---

## A function that generates sections

```{r}
library(pander)

create_section <- function() {

   # Inserts "## Title (auto)"
   pander::pandoc.header('Title (auto)', level = 2)

   # Section contents
   # e.g. a random plot
   plot(sample(1000, 10))

   # a list, formatted as Markdown
   # adding also empty lines, to be sure that this is valid Markdown
   pander::pandoc.p('')
   pander::pandoc.list(letters[1:3])
   pander::pandoc.p('')
}
```

## Generate sections

```{r, results='asis'}
n_sections <- 3

for (i in seq(n_sections)) {
   create_section()
}
```

It still looks hackish, but Markdown has its limits...

Answer 4

It seems like I found a way!

The whole idea is to pass what would be typed by hand as a string inside of knit(text=the_string) used in inline code.

So the function basically pastes a bunch of strings together, with a bit of substitute magic to have a function that feels like it's part of the apply family.

• Parameter depth decides how many # you want.

• Parameter options contains the chunk options, as a vector.

A vector shouldn't be able to contain logical and characters together but here it doesn't matter as it will all be coerced to character anyway, so c(echo= FALSE, results="hide") is fine.

I expect that it's easy to break but seems to work fine when treated gently.

---
title: "test"
output: html_document
---

```{r setup, include = FALSE}
library(knitr)    
mdapply <- function(X, FUN, depth, options=""){
  FUN       <- as.character(substitute(FUN))
  list_name <- as.character(substitute(X))
  if(options != "")
    options <- paste(",",names(options),"=",options,collapse="")
  build_chunk <- function(nm)
  {
    paste0(
      paste0(rep("#",depth), collapse=""),
      " ",
      nm,
      "\n\n```{r", options, "}\n",
      FUN,
      "(", list_name, "[['", nm, "']])\n```")
  }      
  parts <- sapply(names(X), build_chunk)
  whole <- paste(parts, collapse="\n\n")
  knit(text=whole)
  }
```

```{r code}
my_list   <- list(foo = 1:3, bar = 4:7, baz = 8:12)
```

`r mdapply(my_list, plot, 2, c(echo=FALSE))`