使用geom_tile时，如何控制ggplot2中分类数据的y轴刻度？

Question

I am trying to make a heatmap using ggplot in an RMarkdown document. The intended output is an html page. My input dataframe (npBasicLong)looks like this:

location    hh    medianEWS
--------    --    ---------
Alpha       1     2
Alpha       2     1
Alpha       3     3
Alpha       4     1
...(and so on)
Alpha       23    3
Beta        1     4
Beta        2     4
Beta        3     1

Using the following code...

ggplot(data = npBasicLong) +
  aes(x = hh) +
  aes(y = fct_rev(location)) + #fct_rev is used as a hack to reverse the ordering of the wards
  aes(fill = medianEWS) +
  geom_tile(colour = "white") +
  scale_fill_gradient2(name = 'Median\nNEWS\nScore', low = "#7cbc12", mid = "gold", high = "red", midpoint = 3) +
  geom_text (aes
             (label = medianEWS),
             size = 1.5) +
  scale_x_continuous('Hour',
                     expand = c(0, 0),
                     breaks = seq(0,24,1),
                     position = "top") +
  coord_fixed(ratio = 1) +
  theme(panel.grid.minor=element_blank(),
        panel.grid.major=element_blank(),
        axis.ticks = element_blank(),
        axis.title.y = element_blank(),
        text = element_text (size=8)
        )

... I get this...

The problem is that the text is tiny because the only way I've found to stop the labels overlapping or looking crowded I've had to reduce the text size.

If I change the coord_fixed ratio to 2 then rather than elongating the y axis ggplot just squashes the x-axis like so...

I've tried using theme(aspect_ratio) instead of coord_fixed and that seems to make no difference. I've also tried setting the chunk output parameters (out.height and fig.height) but they don't seem to make any difference either.

How can I end up with a plot that has a sufficient vertical height that the labels can be readable on a web page without making the x axis squashed?

Answer 1

tl;dr: The answer is to use fig.height and fig.width and ensure that you set them both.

---

More detail

Spurred on by @Jon Spring's comments above I discovered the answer. Posting it here in case it helps people in the future. In my previous explorations of the chunk options described in the zevross.com article I had made two mistakes.

Firstly I was using out.height/out.width to specify the output in px. This doesn't behave in a way that I could predict - changing the height from 800px to 1600px did not double the size of the image displayed in the knitted HTML page. Once I reached a certain threshold (possibly determined by the width of the container div within which all the notebook content sits) then increasing the values of out.height/width made no difference.

My (possibly mistaken) reading of the zevross article was that out.height/width were like a more flexible version of fig.height/width that applied to both figures and images. This is sort of true but I think it basically acts as a scaling factor. If the problem is, as in my case, that the canvas aspect ratio is wrong then once it hits the max width of the container div then nothing useful happens.

When I switched to using fig.height and fig.width then the plotting canvas size started to change, I could eliminate the pointless whitespace on either side of my plot and as a consequence the plot started to display bigger in the generated HTML document.

The other mistake I had made in my previous experimentation was to only set one parameter. My thought had been that I only cared about one dimension (the height in this case) so the content could determine the width. Setting only the height meant that nothing changed. You have to change both height and width.