Incorrect errobar for drawstyle='steps-post'

[MLopez-Ibanez]

Plotting an errorbar for several step functions draws the error bars ignoring the steps:

import pandas as pd
import seaborn as sns
import matplotlib.pyplot as plt
data = pd.DataFrame({
    "iteration" : [ 1, 3, 8,  1, 3, 8,  1, 3, 8],
    "value"     : [10, 9, 8, 11, 6, 5, 20, 5, 4],
    "run"       : [1,  1, 1,  2, 2, 2,  3, 3, 3],
})
sns.lineplot(data=data, x='iteration', y='value', errorbar='sd',drawstyle='steps-post')
plt.show()

It should be something like this:

using R:

library(ggplot2)
library(pammtools)
data <- read.table(header=TRUE, text='
   iteration  value  run
          1     10    1
          3      9    1
          8      8    1
          1     11    2
          3      6    2
          8      5    2
          1     20    3
          3      5    3
          8      4    3
')
ggplot(data, aes(x=iteration, y=value)) +
  geom_step(stat = "summary", fun = mean) +
  stat_summary(geom='stepribbon', 
    fun.data = mean_se, alpha = 0.5)  # added transparency

Step functions arise in many real-world scenarios. See a more complicated example here: https://adibender.github.io/pammtools/reference/geom_stepribbon.html#examples

[mwaskom]

I don't think this is a bug really, just a consequence of lineplot accepting all of the parameters for the underlying matplotlib plot function, which has a vast functionality. So this is a feature request for lineplot, but probably not one that would get added any time soon.

[MLopez-Ibanez]

I don't think this is a bug really, just a consequence of lineplot accepting all of the parameters for the underlying matplotlib plot function, which has a vast functionality. So this is a feature request for lineplot, but probably not one that would get added any time soon.

I can edit the title to mention Feature Request if you wish.

Step functions are so common (timeseries, ECDFs, survival analysis, inventory levels, etc) that they may warrant a dedicated function stairplot() function. Perhaps this feature request may inspire someone not willing to use R to implement it (or may tell people looking for this feature how to do it in R).