Just a quick post – recently was trying to see examples of both stacking and clustering matplotlib + pandas bar charts. The few examples that come up online do not use what I consider to be one of the simpler approaches. The idea is that clustering is the hard part, and you can replicate the stacked bars fairly easily by superimposing multiple bars on top of each other. For just a quick example, here is a set of similar data to a project I am working on, cases cleared by different units over several years:

import pandas as pd
import matplotlib.pyplot as plt

# data of cases and cleared over time
years = [20,21,22]
acases = [180,190,200]
aclosed = [90,90,90]
bcases = [220,200,220]
bclosed = [165,150,165]
ccases = [90,110,100]
cclosed = [70,100,90]

df = pd.DataFrame(zip(years,acases,aclosed,bcases,bclosed,ccases,cclosed),
                  columns=['Year','Cases_A','Closed_A','Cases_B','Closed_B',
                           'Cases_C','Closed_C'])
print(df)


# Transpose
dfT = df.set_index('Year').T
dfT.index = dfT.index.str.split("_",expand=True) # multi-index

The biggest pain in doing these bar charts is figuring out how to shape the data. My data initially came in the top format, where each row is a year. For this plot though, we want years to be the columns and rows to be the different units. So you can see the format after I transposed the data. And this is pretty much the first time in my life a multi-index in pandas was useful.

To superimpose multiple pandas plots, you can have the first (which is the full bar in the background) return its ax object. Then if you pass that to the subsequent bar calls in superimposes on the plot. Then to finish I need to redo the legend.

# Now can plot
ax = dfT.loc['Cases',:].plot.bar(legend=False,color='#a6611a',edgecolor='black')
dfT.loc['Closed',:].plot.bar(ax=ax,legend=False,color='#018571',edgecolor='black')

# Redo legend
handler, labeler = ax.get_legend_handles_labels()
hd = [handler[0],handler[-1]]
lab = ['Not Closed','Closed']
ax.legend(hd, lab)
ax.set_title('Cases Assigned by Unit 21-23')

plt.show()

To make this a percentage filled bar plot, you just need to change the input data. So something like:

den = dfT.loc['Cases',:]
ax = (den/den).plot.bar(legend=False,color='#a6611a',edgecolor='black')
(dfT.loc['Closed',:]/den).plot.bar(ax=ax,legend=False,color='#018571',edgecolor='black')

# Redo legend
handler, labeler = ax.get_legend_handles_labels()
hd = [handler[0],handler[-1]]
lab = ['Not Closed','Closed']
ax.legend(hd, lab,loc='lower left')
ax.set_title('Proportion Cases Closed by Unit 21-23')

plt.show()

This does not have any way to signal that each subsequent bar in the clustered subsequent is a new year. So I just put that in the title. But I think the part is pretty simple to interpret even absent that info.

Crime De-Coder Updates

For other Crime De-Coder updates. I was interviewed by Jason Elder on his law enforcement analysts podcast.

Also I am still doing the newsletter, most recent talks about my social media strategy (small, regular posts), an interesting research job analyzing administrative datasets at Chase, and some resources I have made for Quarto (which I much prefer over Jupyter notebooks).