python - Single legend at changing categories (!) in subplots from pandas df -

roughly speaking need create 1 legend several subplots, have changing number of categories = legend entries. let me clarify bit deeper:

i have figure 20 subplots, 1 each country within spatial scope:

fig, ax = plt.subplots(nrows=4, ncols=5, sharex=true, sharey=false, figsize = (32,18)) 

within loop, logic group data need normal 2-dimensional pandas dataframe stats , plot each of these 20 axes:

colors = stats.index.to_series().map(type_to_color()).tolist() stats.t.plot.bar(ax=ax[i,j], stacked=true, legend=false, color=colors) 

however, stats dataframe changing size loop loop, since not every category applies each of these countries (i.e. in 1 country there can 2 types, in there more 10). reason pre-defined specific color each type. far, creating 1 legend every subplot within loop:

ax[i,j].legend(fontsize=9, loc='upper right') 

this works, blows subplots unnecessarily. how can plot 1 big legend above/below/beside these plots, since have defined according color. given approach here fig.legend(handles, labels, ...)does not work since line handles not available pandas plot. plotting legend directly with

plt.legend(loc = 'lower center',bbox_to_anchor = (0,-0.3,1,1),     bbox_transform = plt.gcf().transfigure) 

shows entries last subplot, not sufficient.

any appreciated! thank much!

edit example dataframe stats in 1 country this:

              2015       2020       2025       2030       2035       2040                                                                           hydro        29.229082  28.964424  28.528139  27.120194  25.932098  24.675778    natural gas   0.926800   0.926800   0.926800   0.926800   0.003600        nan    wind         25.799950  25.797550   0.776400   0.520800   0.234400        nan    

whereas in country might this:

              2015        2020        2025        2030        2035                                                                  bioenergy     0.033690    0.033690    0.030000         nan         nan    hard coal     5.307300    0.065100    0.021000         nan         nan    hydro        22.834454   23.930642   23.169014   21.639914   19.623791    natural gas   8.378116    8.674121    8.013598    6.755498    5.255450    solar         5.100403    5.100403    5.100403    5.100403    5.093403    wind          8.983560    8.974740    8.967240    8.378300    0.195800  

here's how works legend alphabetical order without messing colors up:

import matplotlib.patches mpatches import collections  fig, ax = plt.subplots(nrows=4, ncols=5, sharex=true, sharey=false, figsize = (32,18)) labels_mpatches = collections.ordereddict()  a, b in enumerate(countries())     # data logic here     colors = stats.index.to_series().map(type_to_color()).tolist()     stats.t.plot.bar(ax=ax[i,j],stacked=true,legend=false,color=colors)     # pass legend information ordereddict     stats_handle, stats_labels = ax[i,j].get_legend_handles_labels()         u, v in enumerate(stats_labels):             if v not in labels_mpatches:                 labels_mpatches[v] = mpatches.patch(color=colors[u], label=v) # after loop, legend layouting. labels_mpatches = collections.ordereddict(sorted(labels_mpatches.items())) fig.legend(handles=labels_mpatches.values(),labels=labels_mpatches.keys())