Using a figure layout with computational images with different dimensions for the subfigures

[andrewheiss]

Description

Currently it's possible to use a code chunk to generate a figure layout with the images created by the chunk, like so:

See @fig-example-same-size-1 and @fig-example-same-size-2

```{r}
#| echo: false
#| label: fig-example-same-size
#| fig-cap: "Charts"
#| fig-subcap: 
#|   - "Cars"
#|   - "Pressure"
#| layout-ncol: 2

plot(cars)
plot(pressure)
```

 

It's possible to control the widths of those subplots too with the layout option:

See @fig-example-different-sizes-1 and @fig-example-different-sizes-2

```{r}
#| echo: false
#| label: fig-example-different-sizes
#| fig-cap: "Charts"
#| fig-subcap: 
#|   - "Cars"
#|   - "Pressure"
#| layout: "[[ 25, 75 ]]"

plot(cars)
plot(pressure)
```

 

However, the dimensions are all squished because both images generated by the chunk use the same width and height.

It would be neat to be able to feed a list of dimensions to fig.width and/or fig.height so that each of the generated images use different dimensions, but I don't think that's possible (see this issue over at {knitr} yihui/knitr#1517)

As an alternative, it's possible to put the different figures in separate chunks and use a ::: container to build the layout:

See @fig-example-multiple-chunks-1 and @fig-example-multiple-chunks-2.

::: {#fig-example-multiple-chunks layout="[[ 25, 75 ]]"}

```{r fig1, echo=FALSE, fig.width=2, fig.height=5}
#| fig-subcap: "Cars"
plot(cars)
```

```{r fig2, echo=FALSE, fig.width=6, fig.height=5}
#| fig-subcap: "Pressure"
plot(pressure)
```

Charts

:::

This looks great!

 

But it no longer treats the plots as subfigures. The cross referencing is broken ("See ?@fig-example-multiple-chunks-1" appears in the text), and the captions don't have the "(a)" and "(b)" prefixes.

Is there a way to get differently-sized computational images into a custom figure layout like this? Either with a vector of dimensions (like fig.width = c(2, 6), fig.height = 5 or by somehow building it with a ::: block?

[mcanouil]

This should work fig.width = c(2, 6), fig.height = 5, as the two parameters are vectorised, or they were at some point.

Note that you should not mix "inline R option" and "yaml option" in the same code cell or expect unintended behaviour.

[cscheid]

Like this:

---
title: Test
---

See @fig-subfig-1 and @fig-subfig-2.

::: {#fig-example-multiple-chunks layout="[[ 25, 75 ]]"}

::: {#fig-subfig-1}
```{r}
#| echo: false
#| fig-width: 2
#| fig-height: 5
plot(cars)
```

Cars

:::

::: {#fig-subfig-2}
```{r}
#| echo: false
#| fig-width: 6
#| fig-height: 5
plot(pressure)
```

Pressure

:::

Charts

:::

[cscheid]

NB: We should make something like your previous syntax work. But in Quarto 1.4 prereleases, we're a lot more lax about what constitutes figures and subfigures, and so you can get away with stuff like what I just wrote.

(Note also that if you echo: true, then you might end up with code where you don't expect it to! So this isn't actually perfect. But it should be enough to get you going.)

[andrewheiss]

Ahhh that works! It's definitely more verbose than the static elephant syntax, but it works.

See @fig-subfig-1 and @fig-subfig-2.

:::: {#fig-example-multiple-chunks layout="[[ 25, 75 ]]"}

::: {#fig-subfig-1}
```{r}
#| echo: false
#| fig-width: 2
#| fig-height: 5
plot(cars)
```

Cars

:::

::: {#fig-subfig-2}
```{r}
#| echo: false
#| fig-width: 6
#| fig-height: 5
plot(pressure)
```

Pressure

:::

Charts

::::

[cscheid]

Btw, this also works (surprisingly to me!)

---
title: Test
---

See @fig-subfig-1 and @fig-subfig-2.

::: {#fig-example-multiple-chunks layout="[[ 25, 75 ]]"}

```{r}
#| label: fig-subfig-1
#| fig-cap: Cars
#| echo: false
#| fig-width: 2
#| fig-height: 5
plot(cars)
```

```{r}
#| label: fig-subfig-2
#| fig-cap: Pressure
#| echo: false
#| fig-width: 6
#| fig-height: 5
plot(pressure)
```

Charts

:::

[andrewheiss]

That's even better, and it's actually basically what I had in the original question, where I ended up with this output with the wrong crossreferences:

 

But that's because I was thinking it was like the output from a multi-plot chunk, so I was using fig-example-multiple-chunks-1 and fig-example-multiple-chunks-2 to reference items 1 and 2 inside the figure div, which didn't work. Also, I wasn't using fig-X for those interior chunks, which was also breaking stuff.

Referencing the labels inside directly works, though.

So in my original question, adding fig- to the interior chunks and then referencing them fixes everything:

See @fig-fig1 and @fig-fig2.

::: {#fig-example-multiple-chunks layout="[[ 25, 75 ]]"}

```{r fig-fig1, echo=FALSE, fig.width=2, fig.height=5}
#| fig-subcap: "Cars"
plot(cars)
```

```{r fig-fig2, echo=FALSE, fig.width=6, fig.height=5}
#| fig-subcap: "Pressure"
plot(pressure)
```

Charts

:::

[cscheid]

The general principle is that "anything goes" as the content of a layout block. But if the inner content has figure-like crossref information, then we should be able to parse it as a subfigure and make the layout work.

Hopefully that will help clean up a lot of previously-necessary-but-ugly quarto code out there. (and sorry again about the regressions..)

[cscheid]

@andrewheiss Just please please, can we convince you to get in the habit of using YAML instead of knitr chunk options? We can't autocomplete or validate knitr chunk options, so those are going to be even riskier going forward...

[andrewheiss]

Oh, is there a practical difference between using fig-subcap and fig-cap here? In your example you used fig-cap for the interior plots; I used fig-subcap. The output seems to be the same. Is that a fluke, or part of the "anything goes" approach?

[andrewheiss]

@andrewheiss Just please please, can we convince you to get in the habit of using YAML instead of knitr chunk options? We can't autocomplete or validate knitr chunk options, so those are going to be even riskier going forward...

@cscheid Ha yes, force of habit. But also I've been keeping only fig.width and fig.height up in the chunk options because it seems to be the only way RStudio can recognize the correct dimensions (via rstudio/rstudio#4521), so my normal approach has been to put all chunk options as YAML except for dimensions. But that's a whole separate issue :)

[andrewheiss]

Continued the issue over there: rstudio/rstudio#4521 (comment)