索引中按日期排列的多个箱线图

Question

My dataframe

index   Dates        Hours_played
0       2014-11-06   11
1       2014-12-06   4
2       2015-09-06   5
3       2015-97-06   5

Then, I set Dates as index:

             Hours_played
Dates        
2014-11-06   11
2014-12-06   4
2015-09-06   5
2015-97-06   5

The Problem: When I tried to create one box plot for each year found in index, I got both plots on the same grid.

df.loc['2014']['Hours_played'].plot.box(ylim=(0,200))
df.loc['2015']['Hours_played'].plot.box(ylim=(0,200))

I tried the following but the plot comes up empty:

data_2015 = df.loc['2015']['Hours_played']
data_2016 = df.loc['2016']['Hours_played']
data_to_plot = [data_2015, data_2016]

mpl_fig = plt.figure()
ax = mpl_fig.add_subplot(111)
ax.boxplot(data_to_plot)
ax.set_ylim(0,300)

Is it possible to have them in the same grid, one by the other?

Answer 1

A simple solution will be grouping by year first and then making boxplot:

import io

import matplotlib.pyplot as plt
import pandas as pd

# Re-create your sample data
s = """Dates,Hours_played
2014-11-06,11
2014-12-06,4
2015-09-06,5
2015-07-06,5"""
df = pd.read_table(io.StringIO(s), sep=',', index_col=0, parse_dates=True)

# The following codes are the answer relevant to your question.
df.groupby(df.index.year).boxplot()
plt.show()

Your second method ends up with an empty plot because matplotlib fail to recognize pandas.DataFrame correctly. Try use Numpy-array representation:

import io

import matplotlib.pyplot as plt
import pandas as pd

# Re-create your sample data
s = """Dates,Hours_played
2014-11-06,11
2014-12-06,4
2015-09-06,5
2015-07-06,5"""
df = pd.read_table(io.StringIO(s), sep=',', index_col=0, parse_dates=True)

# The following codes are the answer relevant to your question.    
data_2014 = df[df.index.year == 2014].as_matrix()
data_2015 = df[df.index.year == 2015].as_matrix()
data_to_plot = [data_2014, data_2015]

mpl_fig = plt.figure()
ax = mpl_fig.add_subplot(111)
ax.boxplot(data_to_plot)

plt.show()

To use subplots, you will need to plot them one by one:

import io

import matplotlib.pyplot as plt
import pandas as pd

# Re-create your sample data
s = """Dates,Hours_played
2014-11-06,11
2014-12-06,4
2015-09-06,5
2015-07-06,5"""
df = pd.read_table(io.StringIO(s), sep=',', parse_dates=[0])
df['Year'] = df.Dates.dt.year
df.set_index(['Year', 'Dates'], inplace=True)

# The following codes are the answer relevant to your question.
mpl_fig = plt.figure()
ax1 = mpl_fig.add_subplot(121)
ax1.boxplot(df.loc[2014]['Hours_played'], labels=[2014])
ax2 = mpl_fig.add_subplot(122)
ax2.boxplot(df.loc[2015]['Hours_played'], labels=[2015])

plt.show()

Answer 2

Let's reshape the data with Years in columns and boxplot :

df.set_index(['Dates',df.Dates.dt.year])['Hours_played'].unstack().boxplot()

Answer 3

If you want to put all the boxes in the same plot, you can do something like this:

import matplotlib.pyplot as plt

def setBoxColors(bp, num_plots):
    color = ['red', 'blue', 'green']
    for idx in range(num_plots):
        plt.setp(bp['boxes'][idx],        color=color[idx])
        plt.setp(bp['caps'][2*idx],       color=color[idx])
        plt.setp(bp['caps'][2*idx+1],     color=color[idx])
        plt.setp(bp['whiskers'][2*idx],   color=color[idx])
        plt.setp(bp['whiskers'][2*idx+1], color=color[idx])
        plt.setp(bp['fliers'][2*idx],     color=color[idx])
        plt.setp(bp['fliers'][2*idx+1],   color=color[idx])
        plt.setp(bp['medians'][idx],      color=color[idx])

# Some fake data to plot
A = [[1, 2, 5,]]
B = [[3, 4, 5]]
C = [[1, 7, 10]]

fig = plt.figure()
ax = plt.axes()
plt.hold(True)

bp = plt.boxplot(A, positions = [2], widths = 0.6, patch_artist=True)
setBoxColors(bp, 1)

bp = plt.boxplot(B, positions = [6], widths = 0.6, patch_artist=True)
setBoxColors(bp, 1)

bp = plt.boxplot(C, positions = [10], widths = 0.6, patch_artist=True)
setBoxColors(bp, 1)

# set axes limits and labels
plt.xlim(0,12)
plt.ylim(0,12)
ax.set_xticklabels(['A', 'B', 'C'])
ax.set_xticks([2, 6, 10])

# draw temporary legend
hB, = plt.plot([1,1],'r-')
plt.legend((hB, ),('Type1', ))
hB.set_visible(False)

plt.show()

Answer 4

With the help of Scott Boston, Y. Luo, and yuhow5566, I was able to devise an interesting answer. From Scott, I learned that it's better not to index the Dates (keep them a regular column) for this type of boxplot; and from Y. Luo, I learned how to create a new column, while isolating the year from a datetime value.

df['Year'] = s['Dates'].dt.year

df.boxplot(column='Hours_played', by='Year', figsize=(9,9))