来自Pandas DataFrame的叠加多个直方图

Question

I've had the hardest time figuring this out. I have a dataframe with multiple categorical fields and I wish to plot them all as histograms with the target variable (Income) overlaid on each histogram. I had hoped to be able to use Pandas to do the histogram and just iterate over all the fields, but when I try just to plot Race and overlay Income the legend says None and I can't seem to get the Income to stack on one another.

Below is a sample dataframe similar to mine and the latest thing I have tried..

exampledf = {'Race': ['Black', 'White', 'Asian', 'White', 
                  'White', 'Asian', 'White', 'White', 
                  'White', 'Black', 'White', 'Asian'],
        'Income': ['>=50k', '>=50k', '>=50k', '>=50k',
                   '>=50k', '<50k', '<50k', '>=50k',
                   '>=50k', '>=50k', '<50k', '>=50k',],
        'Gender': ['M', 'F', 'F', 'F',
                   'M', 'M', 'M', 'M',
                   'M', 'M', 'M', 'M']}
exampledf =pd.DataFrame(exampledf)
exampledf.groupby(['Income','Race']).size().plot(x=exampledf['Race'], kind='bar', color=['r','b'], logy=False, legend=True)

Answer 1

The way you are calling plot is not correct. You don't pass an x variable for a bar plot using pandas. It will automatically use the index as for the x axis. However, because you have a multi-index, it is probably not going to give you the chart you want.

To create a bar chart of race vs income, you need to have race as the index (rows), income as the columns, and the count as the values. You don't want groupby , you want to pivot your data. In this case, you want to use .pivot_table .

This will create a new dataframe with the index as race (the x-values for pandas .plot ) and the different incomes as the the columns (the y-values for .plot ).

pt = exampledf[['Race','Income']].pivot_table(index='Race', columns='Income', 
                                              aggfunc=len, fill_value=0) 
# output of pt:
# Income  <50k  >=50k
# Race
# Asian      1      2
# Black      0      2
# White      2      5

# make the plot
pt.plot.bar()

Here is the image using IPython. The defaults using Jupyter Notebook look better.

Answer 2

The answer from James using pure pandas is likely what you're looking for, but I've been more and more turning to altair for visualization from DataFrames because of its amazing simplicity.

You can get something like what you want by just assigning your frame columns to dimensions in the chart:

from altair import Chart

Chart(exampledf).mark_bar(
).encode(
    y='Race',
    x='count(*)',
    color='Income'
)

or:

Chart(exampledf).mark_bar(
).encode(
    column='Race',
    y='count(*)',
    x='Income'
)