python - Sorting the order of bars in pandas/matplotlib bar plots

Question

What is the Pythonic/pandas way of sorting 'levels' within a column in pandas to give a specific ordering of bars in bar plot.

For example, given:

import pandas as pd
df = pd.DataFrame({
    'group': ['a', 'a', 'a', 'a', 'a', 'a', 'a', 
              'b', 'b', 'b', 'b', 'b', 'b', 'b'],
    'day': ['Mon', 'Tues', 'Fri', 'Thurs', 'Sat', 'Sun', 'Weds',
            'Fri', 'Sun', 'Thurs', 'Sat', 'Weds', 'Mon', 'Tues'],
    'amount': [1, 2, 4, 2, 1, 1, 2, 4, 5, 3, 4, 2, 1, 3]})
dfx = df.groupby(['group'])
dfx.plot(kind='bar', x='day')

I can generate the following pair of plots:

The order of the bars follows the row order.

What's the best way of reordering the data so that the bar charts have bars ordered Mon-Sun?

UPDATE: this rubbish solution works - but it's far from elegant in the way it uses an extra sorting column:

df2 = pd.DataFrame({
    'day': ['Mon', 'Tues', 'Weds', 'Thurs', 'Fri', 'Sat', 'Sun'],
    'num': [0, 1, 2, 3, 4, 5, 6]})
df = pd.merge(df, df2, on='day')
df = df.sort_values('num')
dfx = df.groupby(['group'])
dfx.plot(kind='bar', x='day')

FURTHER GENERALISATION:

Is there a solution that also fixes the order of bars in a 'dodged' bar plot:

df.pivot('day', 'group', 'amount').plot(kind='bar')

Answer 1

You'll have to provide a mapping to specify how to order the day names. (If they were stored as proper dates, there would be other ways to do this.)

Updated:

Build the key. You could write out a dictionary explicitly or use something clever like this dict comprehension.

weekdays = ['Mon', 'Tues', 'Weds', 'Thurs', 'Fri', 'Sat', 'Sun']
mapping = {day: i for i, day in enumerate(weekdays)}
key = df['day'].map(mapping)

And the sorting is simple:

df.iloc[key.argsort()]