使用Pandas数据框绘制Seaborn中的时间序列

Question

I am trying to explore some data that I have imported from a .csv. Within the file, there are around 100 'Companies', each of which has a 'SG' and a 'RG' metric. The data has several years of data for these metrics across the columns.

I am trying to build some seaborn time series charts to overlay 'SG' and 'RG' lines on the same chart for a given 'Company'. This image should explain what I mean:

Summary of the lux dataframe

Could anyone give me some guidance on how to create such a plot? For example, plotting the 'Barbour' Company (from image above) with two lines, one for 'SG' and one for 'RG'.

(Note, all the data types are floats, but with some NaNs in there, and I have included the usual stuff up front, such as:

%matplotlib inline
import numpy as np
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns

Answer 1

It is typically better to carry your data in long-form when using seaborn (well, in general). An exception to this is when you'd like to use heatmaps, for which you'll need to pivot into a clean 2-variable table or matrix.

Anyways, in long-form, each row has exactly one observation, and supporting features are just additional columns on these rows. In your particular case, each row has several observations (1 for each year, given a company and metric), so we'd like to convert this to long-form, in which the year is merely another feature of your observation. Fortunately, pnd.melt can help you with that, as it is intended for that exact purpose.

Let's start with a generic pnd.DataFrame , based on yours:

In [1]: import pandas as pnd
In [2]: import seaborn as sns

In [3]: df = DataFrame.from_dict({
     ...:      'company': ['A', 'B', 'A', 'B', 'C'],
     ...:      'metric': ['SG', 'SG', 'RG', 'RG', 'SG'], 
     ...:      '2005': [1, 2, 3, 4, 5],
     ...:      '2006': [4, 5, 6, 7, 8]})    
In [4]: df
Out[4]: 
   2005  2006 company metric
0     1     4       A     SG
1     2     5       B     SG
2     3     6       A     RG
3     4     7       B     RG
4     5     8       C     SG

Converting to long-form using pnd.melt :

In [5]: df_melt = pnd.melt(df, 
                           id_vars=['company', 'metric'], 
                           value_vars=['2005', '2006'], 
                           var_name='year', 
                           value_name='value')
In [6]: df_melt
Out[6]: 
  company metric  year  value
0       A     SG  2005      1
1       B     SG  2005      2
2       A     RG  2005      3
3       B     RG  2005      4
4       C     SG  2005      5
5       A     SG  2006      4
6       B     SG  2006      5
7       A     RG  2006      6
8       B     RG  2006      7
9       C     SG  2006      8

And finally with sns.factorplot you can make use of parameters like, x, hue, row, col to visualize the data by doing factor-breakdowns:

In [7]: sns.factorplot(data=df_melt, 
                       x='year', 
                       y='value', 
                       hue='metric', 
                       col='company')

Out[7]: <seaborn.axisgrid.FacetGrid at 0x7f6286fee890>

In [8]: from matplotlib import pyplot as plt
In [9]: plt.show()