Skip to Main Menu

Plotting football pitches

Posted on 9 mins read

I’ve recently been playing with some football data from Stratagem1 with locations of shots taken. And of course, the first thing I wanted to do was to plot it. This is different to my normal plots because I needed to give the plots context. So I learnt how to draw football pitches with ggplot2. On the way, I learnt about ggforce, a little about functional programming ggplot2 and a few other things. The following post documents a bit about the function and what I’ve learnt.

Goal

So this is what the final code and plot looks like:

plot_chances_football_pitch(
  ggplot() +
    # geom_point_football_pitch(data = chances, colour = "red", alpha = 0.3) +
    geom_hex_football_pitch(data = chances) +
    geom_football_pitch() +
    theme_football_pitch() +
    labs(title = "Probability of scoring by chance location",
       subtitle = str_c("The closer the chance is to the goal the higher the probability",
                        "\n",
                        "of scoring"),
         caption = str_c("as at 2018-04-11",
                         "\n",
                         "@tojyouso"),
       x = NULL, y = NULL)
)

This is all chances (excluding penalties) in the highest division of the top 5 European football leagues. There were 79127 chances which doesn’t give a meaningful plot so I have summarised them into small hexagons where the lighter the hexagon the higher the proportion of goals scored from that hexagon. So if you ignore the sparse data and focus on where there is volume, it shows that location plays a huge part in the success of a chance.

I love how simple the code looks and how easy it is to change certain items to get a completely different looking plot:

plot_chances_football_pitch(
  ggplot() +
    # geom_point_football_pitch(data = chances, colour = "red", alpha = 0.3) +
    geom_hex_football_pitch(data = chances,
                            prop = F,
                            vir_dir = -1) +
    geom_football_pitch(line_colour = "black", background_colour = "white") +
    theme_football_pitch(line_colour = "black", background_colour = "white") +
    labs(x = NULL, y = NULL,
       title = "Open play shots from the top five European leagues",
       subtitle = "Most shots are taken within the 18 yard box",
       caption = str_c("as at 2018-04-11",
                         "\n",
                         "@tojyouso")),
  logo_colour = "white"
)

This plot shows the number of chances within each hexagon and the darker areas are areas where more chances are taken. Apart from a different theme, the plot is showing a count (look at the legend) instead of a proportion like the first one.

Not sure which one looks better but at least I can switch it up as I like.

Final example: not only can the function handle hexagonal summaries, it can also plot individual chances. Here is a plot of Romelu Lukaku’s efforts on goal in the dataset:

plot_chances_football_pitch(
  ggplot() +
    geom_point_football_pitch(data = chances %>% filter(player == "R. Lukaku"),
                              colour = "red", alpha = 0.6) +
    geom_football_pitch() +
    theme_football_pitch() +
    labs(title = "Romelu Lukaku",
       subtitle = str_c("Manchester United | 2016/2017 - 2017/2018"),
         caption = str_c("as at 2018-04-11",
                         "\n",
                         "@tojyouso"),
       x = NULL, y = NULL))

Functional programming in ggplot2

The plot is made up of a few things as you can see from the code:

  1. geom_hex_football_pitch
  2. geom_point_football_pitch
  3. geom_football_pitch
  4. theme_football_pitch
  5. Stratabet logo

Each of these things relies on function I created. The full code behind each function can be found here. I’ll probably be updating it over time so if it’s possible the code examples used in this post might break in the future!

When I was investigating how to write the code for the plots as a function I came across this stackoverflow answer which led me to this tutorial by Hadley Wickham on functional programming in ggplot2. The main discovery here was how one can specify a block of ggplot2 calls as a list and then insert that into a function. Whenever the function is called, the block of ggplot2 code is used.

I also finally learnt how ... works. I call it the pass through. It allows you to specify a lower level function argument in a higher level function call. I think that makes sense. It’s best with an example, so here’s the code for the geom_hex_football_pitch:

# group individual chances into hexagons based on count or other proportional metric
geom_hex_football_pitch <- function(prop = T,
                              hex_num = 100,
                              x_loc = "chance_x",
                              y_loc = "chance_y",
                              metric = "goal",
                              vir_dir = 1,
                              ...) {
  if (prop) {
    
    # list of ggplot2 code
    list(
      stat_summary_hex(bins = hex_num, aes_string(y = y_loc, x = x_loc, z = metric), ...),
      scale_fill_viridis_c(option = "A", label = scales::percent, direction = vir_dir)
    )
  } else {
    
    # another list of ggplot2 code
    list(
      stat_bin_hex(bins = hex_num, aes_string(y = y_loc, x = x_loc), ...),
      scale_fill_viridis_c(option = "A", label = scales::comma, direction = vir_dir)
    )
  }
}

Because I’ve specified the pass through as a function argument and also in the stat_summary_hex, I can use any of the arguments in stat_summary_hex when I call geom_hex_football_pitch. That’s what I did in the earlier plots when I used the data argument:

geom_hex_football_pitch(data = chances)

The geom_point_football_pitch works in a very similar way. I was able to use quite a arguments from the geom_point that that function is wrapped around:

geom_point_football_pitch(data = chances %>% filter(player == "R. Lukaku"),
                              colour = "red", alpha = 0.6)

Drawing the pitch

The pitch plot is quite simple. It’s made up of straight lines and a couple of curved lines.

The straight lines were the easiest part - just a simple geom_segment which only needs a start coordinate, an end coordinate and a colour.

geom_segment(aes(x = 15, xend = 15, y = -10, yend = 0), colour = line_colour)

The circles were a bit more difficult. Actually I cheated here a little, they are ellipses but I think they look good enough for what I’m doing. I used yet another package from Thomas Pedersen and the function ggforce::geom_arc. I needed to specify a radius and then how much of the ellipse I wanted. For example, for a complete ellipse, I needed to specify a start and end that spanned 2 * pi. Here’s what the code looks like for the corner flag on the right:

ggforce::geom_arc(aes(x0 = 136, y0 = 0, start = 1.5 * pi, end = 2 * pi, r = 4), colour = line_colour)

Conclusion and next steps

So that’s it. I finally finished the code to create the plots that I’m satisfied with. Like I mentioned earlier, I’ll be updating this over the next few months as I explore the data further. There are quite a few other plots I’d like to do with the data so maybe this will give me a chance to try out creating a package.

Anyway, all the code for the function is here

  1. This article was written with the aid of StrataData, which is property of Stratagem Technologies. StrataData powers the StrataBet Sports Trading Platform, in addition to StrataBet Premium Recommendations.↩

  2. There’s an awesome vignette↩

comments powered by Disqus