CODEWITHSUNDEEP
pythonpandasdataframeboxplot
Trond Kristiansen
Trond Kristiansen

Reputation: 2446

Seaborn boxplot

I have a multi-index Pandas dataframe that I want to plot as a boxplot. This should be easy to do yet I find myself unable to get exactly what I want. The data looks like this:

                       hedges  mask model_name  hedges_std  hedges_min  \
    period    season                                                      
    2021-2025 winter  0.864328   1.0   ensemble    0.301748    0.124708   
          spring  0.740410   1.0   ensemble    0.202963    0.049319   
          summer  0.526264   1.0   ensemble    0.105750    0.162856   
          fall    0.531141   1.0   ensemble    0.046278    0.388827   
2025-2050 winter  1.715075   1.0   ensemble    0.373866    0.582819   
          spring  1.252963   1.0   ensemble    0.370402    0.408695   
          summer  0.854958   1.0   ensemble    0.076193    0.528038   
          fall    0.759645   1.0   ensemble    0.068928    0.498271   
2050-2075 winter  2.981373   1.0   ensemble    0.928940    1.139801   
          spring  2.042320   1.0   ensemble    0.748642    0.716289   
          summer  1.299277   1.0   ensemble    0.092611    0.812979   
          fall    1.108852   1.0   ensemble    0.109014    0.653199   
2021-2025 winter  0.864328   1.0   ensemble    0.301748    0.124708   
          spring  0.740410   1.0   ensemble    0.202963    0.049319   
          summer  0.526264   1.0   ensemble    0.105750    0.162856   
          fall    0.531141   1.0   ensemble    0.046278    0.388827   
2025-2050 winter  1.715075   1.0   ensemble    0.373866    0.582819   
          spring  1.252963   1.0   ensemble    0.370402    0.408695   
          summer  0.854958   1.0   ensemble    0.076193    0.528038   
          fall    0.759645   1.0   ensemble    0.068928    0.498271   
2050-2075 winter  2.981373   1.0   ensemble    0.928940    1.139801   
          spring  2.042320   1.0   ensemble    0.748642    0.716289   
          summer  1.299277   1.0   ensemble    0.092611    0.812979   
          fall    1.108852   1.0   ensemble    0.109014    0.653199   

                  hedges_max model_scenario  
period    season                             
2021-2025 winter    1.760912         ssp245  
          spring    1.189956         ssp245  
          summer    0.662142         ssp245  
          fall      0.687793         ssp245  
2025-2050 winter    2.423660         ssp245  
          spring    2.040903         ssp245  
          summer    1.055890         ssp245  
          fall      0.965831         ssp245  
2050-2075 winter    5.179203         ssp245  
          spring    3.898118         ssp245  
          summer    1.536149         ssp245  
          fall      1.435503         ssp245  
2021-2025 winter    1.760912         ssp585  
          spring    1.189956         ssp585  
          summer    0.662142         ssp585  
          fall      0.687793         ssp585  
2025-2050 winter    2.423660         ssp585  
          spring    2.040903         ssp585  
          summer    1.055890         ssp585  
          fall      0.965831         ssp585  
2050-2075 winter    5.179203         ssp585  
          spring    3.898118         ssp585  
          summer    1.536149         ssp585  
          fall      1.435503         ssp585

I want to plot the data showing one box for each period and season separated in color by scenario. Each box would be defined by its mean (hedges), standard deviation (std), and potentially min and max range. The idea is to show how the future periods will change the estimated hedges distributions. I have tried a variety of combinations around:

sns.boxplot(data=df, x="season", y="hedges", hue="model_scenario")

My error Could not interpret input 'season' is related to the multi-index which I clearly have to group or split somehow but that's where I keep failing. Suggestions for how to plot these data are appreciated.

Upvotes: 0

Views: 1136

Answers (1)

mosc9575
mosc9575

Reputation: 6337

I assume your goal is to generate a figure like this:

Boxplot generated by bxp()

Since you have the boxplot-statistics of your boxes already calculated, the function sns.boxplot() and also matplotlib.axes.Axes.boxplot() from matplotlib (which is the seaborn backend and called inside sns.boxplot()) aren't the functions you can use anymore. The ax.boxplot() trys to calculate the statistics by itself, therefor this is not the way to go.

After calculating the boxplot-statistics matplotlib.axes.Axes.boxplot() calls [matplotlib.axes.Axes.bxp()](https://matplotlib.org/stable/api/_as_gen/matplotlib.axes.Axes.bxp.html) and this is a function you can use, too.

The function matplotlib.axes.Axes.boxplot() takes a dict with this name convention:

  • med: The median (scalar float),
  • q1: The first quartile (25th percentile) (scalar float),
  • q3: The third quartile (75th percentile) (scalar float),
  • whislo: Lower bound of the lower whisker (scalar float),
  • whishi: Upper bound of the upper whisker (scalar float),

and with only small modifications we can rename or genreate the needed columns of you DataFrame. But first reset you multiindex.

# df is defined and the multiinde
df = df.rename({'hedges':'med', 'hedges_min':'whislo', 'hedges_max':'whishi'}, axis=1)
df['q1'] = df['med'] - df['hedges_std']
df['q3'] = df['med'] + df['hedges_std']
df['label'] = df.apply(lambda x: '('+ x['period'] +' , '+ x['season'] + ')', axis=1)
df = df[['med', 'whislo','whishi','q1','q3', 'label']] # this are the columns we need

>>> df.head(5)
        med    whislo    whishi        q1        q3                 label
0  0.864328  0.124708  1.760912  0.562580  1.166076  (2021-2025 , winter)
1  0.740410  0.049319  1.189956  0.537447  0.943373  (2021-2025 , spring)
2  0.526264  0.162856  0.662142  0.420514  0.632014  (2021-2025 , summer)
3  0.531141  0.388827  0.687793  0.484863  0.577419    (2021-2025 , fall)
4  1.715075  0.582819  2.423660  1.341209  2.088941  (2025-2050 , winter)

I decided to create a label combining period and season. Every label appears twice, for each model_scenario exactly one time.

Here is the code how I created the figure above. It is not perfect, but it shows, how it works. Some of the sections a realated to the code of sns.boxplot().

from matplotlib import rcParams
import matplotlib.pyplot as plt

colors = ['lightblue', 'olive']
model_scenario = ["ssp245", "ssp585"]
fig, ax = plt.subplots(figsize=(9, 4))
ax.set_title('box plot')

x_tick_label = []
x_tick_position = []
for i, group in enumerate(data_to_plt.groupby('label')):
    for j in range(group[1].shape[0]):
        x_tick_label.append(group[0])
        x_tick_position.append(i)
        if j ==0:
            p = i - 0.15
        else:
            p = i + 0.15
        artist_dict  = ax.bxp(
            bxpstats=[group[1].drop('label', axis=1).iloc[j].to_dict()], 
            showfliers=False, 
            patch_artist=True,
            positions=[p]
        )
        for box in artist_dict["boxes"]:
            box.update(dict(facecolor=colors[j],
                            zorder=.9,
                            edgecolor='gray',
                            linewidth=rcParams["lines.linewidth"])
            )
        if i == 0:
            rect = plt.Rectangle([0,0], 0, 0,
                                 linewidth=0,
                                 edgecolor='gray',
                                 facecolor=colors[j],
                                 label=model_scenario[j])
            ax.add_patch(rect)
            

ax.legend(loc='center left', bbox_to_anchor=(1, 0.5))
plt.xticks(x_tick_position, x_tick_label, rotation = 90)

To summerize what I am doing with matplotlib:

  1. I group by model_scenario aka labels
  2. I generate the labels for the legend
  3. I draw the boxes using bxp()
  4. I rewrite the x-ticks

Upvotes: 1

Related Questions