按pandas / matplotlib条形图中的条形顺序排序

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()]

Answer 2

I know this response is late, but a simplistic solution to the two cases presented, without use of a dictionary/mappings would be something like I've posted below.

Setting 'day' as an index enables you to use .loc to select data in a specific order

1) For the two separate plots

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]})

order = ['Mon', 'Tues', 'Weds','Thurs','Fri','Sat','Sun']`
df.set_index('day').loc[order].groupby('group').plot(kind='bar')

2) For the pivot example with the dodged plot:

order = ['Mon', 'Tues', 'Weds','Thurs','Fri','Sat','Sun']
df.pivot('day','group','amount').loc[order].plot(kind='bar')

note that pivot results in day being in the index already so you can use .loc here again.

Edit: it is best practice to use .loc instead of .ix in these solutions, .ix will be deprecated and can have weird results when column names and indexes are numbers.

Answer 3

I will provide bellow code to extend Dan's answer to address the "FURTHER GENERALIZATION" section of the OP's question. First, a complete example for the simple case (just one variable) based in Dan's solution:

import pandas as pd

# Create dataframe 
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]
})


# Calculate the total amount for each day
df_grouped = df.groupby(['day']).sum().amount.reset_index()

# Use Dan's trick to order days names in the table created by groupby
weekdays = ['Mon', 'Tues', 'Weds', 'Thurs', 'Fri', 'Sat', 'Sun']
mapping = {day: i for i, day in enumerate(weekdays)}
key = df_grouped['day'].map(mapping)    
df_grouped = df_grouped.iloc[key.argsort()]

# Draw the bar chart
df_grouped.plot(kind='bar', x='day')

And now, we use the same ordering technique to order the rows of the pivot table (instead of the rows created by groupby).

import pandas as pd

# Create dataframe 
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]
})

# Get the amount for each day AND EACH GROUP
df_grouped = df.groupby(['group', 'day']).sum().amount.reset_index()

# Create pivot table to get the total amount for each day and each in the proper format to plot multiple series with pandas
df_pivot = df_grouped.pivot('day','group','amount').reset_index()

# Use Dan's trick to order days names in the table created by PIVOT (not the table created by groupby, in the previous example)
weekdays = ['Mon', 'Tues', 'Weds', 'Thurs', 'Fri', 'Sat', 'Sun']
mapping = {day: i for i, day in enumerate(weekdays)}
key = df_pivot['day'].map(mapping)    
df_pivot = df_pivot.iloc[key.argsort()]

# Draw the bar chart
df_pivot.plot(kind='bar', x='day')

The result is shown bellow: