Matplotlib PyPlot Stacked histograms - stacking different attributes in each bar

Question

I have the following code to draw some histograms about subjects in a database:

import matplotlib.pyplot as plt

attr_info = {
    'Gender': ['m', 'f', 'm', 'm', 'f', 'm', 'm', 'f', 'm', 'f'],
    'Age': [9, 43, 234, 23, 2, 95, 32, 63, 58, 42],
    'Smoker': ['y', 'n', 'y', 'y', 'n', 'n', 'n', 'n', 'y', 'y']
}
bin_info = {key: None for key in attr_info}
bin_info['Age'] = 10

for name, a_info in attr_info.items():
    plt.figure(num=name)
    counts, bins, _ = plt.hist(a_info, bins=bin_info[name], color='blue', edgecolor='black')

    plt.margins(0)
    plt.title(name)
    plt.xlabel(name)
    plt.ylabel("# Subjects")
    plt.yticks(range(0, 11, 2))
    plt.grid(axis='y')
    plt.tight_layout(pad=0)

    plt.show()

This code works but it draws each attribute's distribution in a separate histogram. What I'd like to achieve is something like this:

I'm aware plt.hist has a stacked parameter, but that seems to be intended for a slightly different use, where you're stacking the same attributes on each other at different subject types. You could for example draw a histogram where each whole bar would represent some age range and the bar itself would be a stack of smokers in one colour and non-smokers in another.

I haven't been able to figure out how to use it to stack (and properly label as in the image) different attributes on top of each other in each bar.

Answer 1

How about trying out pandas:

import pandas as pd

attr_info = {
    'Gender': ['m', 'f', 'm', 'm', 'f', 'm', 'm', 'f', 'm', 'f'],
    'Age': [9, 43, 234, 23, 2, 95, 32, 63, 58, 42],
    'Smoker': ['y', 'n', 'y', 'y', 'n', 'n', 'n', 'n', 'y', 'y']
}

df =  pd.DataFrame(attr_info)

bins = [0,32,45,300] #bins can be adjusted to your liking

#deselect "Age" and select all remaining columns
counts = df.filter(regex="[^Age]").apply(pd.Series.value_counts) 
#bin age data and count
age_data = df.groupby(pd.cut(df['Age'], bins=bins))["Age"].count()

fig, ax = plt.subplots()
pd.concat([counts,age_data]).rename(columns={0:"Age"}).T.plot(kind="bar", stacked=True, ax=ax)
ax.legend(loc='center left', bbox_to_anchor=(1, 0.5))

Output:

The advantage of this approach is its generality, no matter how many columns you want to plot.

Answer 2

You need to play around with your data a bit, but this can be done without pandas. Also, what you want are stacked bar plots, not histograms:

import matplotlib.pyplot as plt

attr_info = {
'Gender': ['m', 'f', 'm', 'm', 'f', 'm', 'm', 'f', 'm', 'f'],
'Age': [9, 43, 234, 23, 2, 95, 32, 63, 58, 42],
'Smoker': ['y', 'n', 'y', 'y', 'n', 'n', 'n', 'n', 'y', 'y']
}

# Filter your data for each bar section that you want
ages_0_10 = [x for x in attr_info['Age'] if x < 10]
ages_10_40 = [x for x in attr_info['Age'] if x >= 10 and x < 40]
ages_40p = [x for x in attr_info['Age'] if x > 40]

gender_m = [x for x in attr_info['Gender'] if 'm' in x]
gender_f = [x for x in attr_info['Gender'] if 'f' in x]

smoker_y = [x for x in attr_info['Smoker'] if 'y' in x]
smoker_n = [x for x in attr_info['Smoker'] if 'n' in x]

# Locations for each bin (you can move them around)
locs = [0, 1, 2]

# I'm going to plot the Ages bin separate than the Smokers and Gender ones, 
# since Age has 3 stacked bars and the other have just 2 each
plt.bar(locs[0], len(ages_0_10), width=0.5)  # This is the bottom bar

# Second stacked bar, note the bottom variable assigned to the previous bar
plt.bar(locs[0], len(ages_10_40), bottom=len(ages_0_10), width=0.5) 

# Same as before but now bottom is the 2 previous bars    
plt.bar(locs[0], len(ages_40p), bottom=len(ages_0_10) + len(ages_10_40), width=0.5)

# Add labels, play around with the locations
#plt.text(x, y, text)
plt.text(locs[0], len(ages_0_10) / 2, r'$<10$')
plt.text(locs[0], len(ages_0_10) + 1, r'$[10, 40]$')
plt.text(locs[0], len(ages_0_10) + 5, r'$>40$')


# Define the top bars and bottom bars for the Gender and Smokers stack
# In both cases is just 2 stacked bars,
# so we can use a list for this instead of doing it separate as for Age
tops = [len(gender_m), len(smoker_y)]
bottoms = [len(gender_f), len(smoker_n)]

plt.bar(locs[1:], bottoms, width=0.5)
plt.bar(locs[1:], tops, bottom=bottoms, width=0.5)

# Labels again
# Gender
plt.text(locs[1], len(gender_m) / 2, 'm')
plt.text(locs[1], len(gender_m) + 2, 'f')

# Smokers
plt.text(locs[2], len(smoker_y) / 2, 'y')
plt.text(locs[2], len(smoker_n) + 2, 'n')

# Set tick labels
plt.xticks(locs, ('Age', 'Gender', 'Smoker'))
plt.show()

Result:

Check the documentation for pyplot.bar and this example.