import pandas as pd
import seaborn as sns

Wide data

Wide data is a dataframe where your columns measure the same things across different categories. In the example below, we have the number of accidents (our measurement) across different months (our categories).

wide_df = pd.DataFrame({
    'category': ['car', 'bus', 'plane', 'horse', 'submarine', 'train', 'subway', 'spaceship'],
    'JAN': [1, 4, 3, 2, 6, 2, 4, 5],
    'FEB': [3, 5, 2, 2, 6, 3, 2, 4],
    'MAR': [5, 3, 3, 5, 2, 5, 6, 4]
})
wide_df

# I can only plot one month at a time!
sns.catplot(data=wide_df, y='JAN', x='category')

<seaborn.axisgrid.FacetGrid at 0x117fb5240>

Wide data is great for stacked bar graphs, but for everything else it can be a real pain (especially with seaborn). Most of the software you’ll use to graph loves long data instead.

Long data

Long data is similar to the idea of tidy data which is very very popular in current-day R programming. Basically speaking, each row is a measurement.

To convert from wide data to long data, you use .melt. I think you can figure out what the columns mean based on what’s down below!

long_df = wide_df.melt(id_vars=['category'],
                       value_vars=['JAN', 'FEB', 'MAR'],
                       var_name='month',
                       value_name='accidents')
long_df

# Now I can plot whatever I want!
sns.catplot(data=long_df, y='month', x='accidents', hue='category')

<seaborn.axisgrid.FacetGrid at 0x1180bd1d0>

Transposing my data

What if I wanted every one of my categories to be a column?

wide_df

You can transpose with .T and it’s close, but you don’t end up with any column names.

wide_df.T

But notice how the column names are the index from the original dataframe? Turns out you just need to tell it what the index should be, then transpose.

wide_df.set_index('category').T

And then you’re all set to do a nice stacked bar (actually a horrible stacked bar).

wide_df.set_index('category').T.plot(kind='barh', stacked=True)

<matplotlib.axes._subplots.AxesSubplot at 0x1193bb4a8>