Plot multiIndex and multicolumn dataframe

Python Help
1

Hello,

assume the following minimal example

from pandas import DataFrame
import pandas as pd
import numpy as np
import matplotlib.pyplot as plt

samples = [100, 200]
epochs = [3, 4, 5]
kpi = ['ACC', 'PRC', 'REC', 'MM']
models = ['M0', 'M1', 'M2']
data = np.random.rand(12, 6)
df = pd.DataFrame(data)
df.columns = pd.MultiIndex.from_product([kpi, models], names=['kpi','models'])
df.index = pd.MultiIndex.from_product([samples, epochs], names=['samples', 'epochs'])
print(df)

which outputs the following dataframe with multiple indices both for the rows and for the columns

kpi                  ACC                           PRC                           REC                            MM                    
models                M0        M1        M2        M0        M1        M2        M0        M1        M2        M0        M1        M2
samples epochs                                                                                                                        
100     3       0.906675  0.319360  0.675432  0.253069  0.584511  0.303165  0.248698  0.071041  0.454044  0.969305  0.922541  0.671037
        4       0.142995  0.048634  0.315539  0.407142  0.345572  0.909697  0.757394  0.131977  0.595874  0.995964  0.516483  0.068203
        5       0.905476  0.216253  0.252296  0.296729  0.128222  0.885560  0.891368  0.766562  0.086110  0.919574  0.543337  0.839931
200     3       0.776404  0.677118  0.190398  0.362749  0.740333  0.678845  0.761190  0.164483  0.369186  0.656634  0.481825  0.026312
        4       0.395710  0.751088  0.016821  0.077905  0.916027  0.470270  0.349636  0.903484  0.851275  0.709731  0.004874  0.984908
        5       0.564700  0.744458  0.011534  0.905957  0.135282  0.637977  0.633657  0.192773  0.175391  0.163563  0.679527  0.334195

I want to plot these results in the following configuration:

  • One bar chart per each of the kpi index values ([ACC, PRC, REC, MM])
    • In each of these charts I will plot bars for each of the models ([M0, M1, M2])
    • I want to be able to select which values will be selected based on the samples or the epochs or for all of their combinations. For example, I might want to compare the models for epochs=3 and for all values of samples, or for samples=100 and for all values of epochs and somehow show this configuration on the plot (as a title maybe)?

Any ideas?