Dotplot bar chart?

[concimuscb]

Is there any way to create a chart similar to this?

Bar charts are one of the most used chart types and having alternatives for them is great. Currently Let's Plot has geom_lollipop but DotPlot bars are another great alternative, especially when the measured value is an integer. Counting number of records is probably the best use case for this.

I tried creating them using geom_dotplot and geom_ydotplot but was not able to do it.

If it is not currently possible I believe it would be a great addition.

[alshan]

I see, geom_dotplot() paints each observation as a point, and it doesn't use any kind of summary or stat.

As a workaround you can have a synthetic dataset containing "observations" according to the counts by category and then use geom_dotplot().

If we had "patterns" like in ggplot2 : Patterns and Gradients then "polka dot" pattern would also do I guess.

[concimuscb]

To be honest even after your pointers I was not able to create the chart that I wanted.

I got close but the dots were very, very small and using dotsize would just make them overlap instead of expand upwards.

[alshan]

You can fix the overlapping by adjusting the stackratio parameter. Examples: dot_plots.ipynb.

[concimuscb]

My best two attempts, still not quite what I want.

Prepare the data

mpg = (
    pl.read_csv("../../../data/raw/external/mpg.csv")
    .filter(pl.col("class").is_in(["pickup", "suv", "subcompact", "midsize"]))
    .select("class")
    .with_columns(pl.col("class").cum_count().over("class").alias("count"))
)

Attempt with geom_dotplot

dotplot = (
    ggplot(mpg, aes(x=as_discrete("class", order_by="..count..")))
    + scale_y_continuous()
    + geom_dotplot(binwidth=1 / 60, fill="orange", color="orange")
)

Issues:

• Axis is displaying density instead of count
• Dots are way too small

To fix the dot size issue I used dotsize but this eventually leads to the dots going out of chart bounds.

dotplot = (
    ggplot(mpg, aes(x=as_discrete("class", order_by="..count..")))
    + geom_dotplot(
        binwidth=1 / 60,
        fill="orange",
        color="orange",
        dotsize=4,
    )
)

Any attempts at resizing the y axis with scale_y_continuous fails with the plot being exactly the same but with different labels:

dotplot = (
    ggplot(mpg, aes(x=as_discrete("class", order_by="..count..")))
    + scale_y_continuous(limits=[0, 1000])
    + geom_dotplot(
        binwidth=1 / 60,
        fill="orange",
        color="orange",
        dotsize=4,
    )
)

Attempt with geom_ydotplot

ydotplot = (
    ggplot(mpg, aes(x="class", y="count"))
    + scale_y_continuous()
    + geom_ydotplot(binwidth=1, method="histodot")
)

Issues:

• Dots are way too small

To fix the dot size issue I used dotsize but this eventually leads to overlapping.

ydotplot = (
    ggplot(mpg, aes(x="class", y="count"))
    + scale_y_continuous()
    + geom_ydotplot(binwidth=1, method="histodot", dotsize=5)
)

Use stackratio to fix overlapping:

ydotplot = (
    ggplot(mpg, aes(x="class", y="count"))
    + scale_y_continuous()
    + geom_ydotplot(binwidth=1, method="histodot", dotsize=5, stackratio=20)
)

Chart seems unresponsive to it.

All of the above happens whether method is histodot or dotdensity.

I guess the addition of patterns and gradients would also help with #1199 but probably quite low on the priority list and understandably so.

[alshan]

I see, geom_dotplot and geom_ydotplot aren't really cut for what you want to achieve.

However, this can be done using just geom_point:

categories = ['a', 'b', 'c']
counts = [6, 5, 8] 

xs_arrays = [[cat] * count for cat, count in zip(categories, counts)]
ys_arrays = [list(range(1, count + 1)) for count in counts]

data = dict(
   xs = sum(xs_arrays, []), 
   ys = sum(ys_arrays, []) 
)

ggplot(data) + geom_point(aes('xs', 'ys', color='xs'),
                          size_unit='y',              # <-- relative size
                          size=0.5
                         )

The size_unit='y', here is our secret weapon that allows us to define size of points relative to 'data resolution' (i.e. 1 here) along the y-axis.

[alshan]

Or better 0-based maybe:

xs_arrays = [[cat] + [cat] * count for cat, count in zip(categories, counts)]
ys_arrays = [[0] + list(range(1, count + 1)) for count in counts]

data = dict(
   xs = sum(xs_arrays, []), 
   ys = sum(ys_arrays, []) 
)

(ggplot(data) + geom_point(aes('xs', 'ys', color='xs'),
                          size_unit='y',              # <-- relative size
                          size=0.5
                         )
 + coord_cartesian(ylim=[0, None])
)