MultiIndex Group By in Pandas Data Frame
In this case, you don't actually need a groupby
. You also don't have a MultiIndex
. You can make one like this:
import pandas
from io import StringIO
datastring = StringIO("""\
Country Metric 2011 2012 2013 2014
USA GDP 7 4 0 2
USA Pop. 2 3 0 3
GB GDP 8 7 0 7
GB Pop. 2 6 0 0
FR GDP 5 0 0 1
FR Pop. 1 1 0 5
""")
data = pandas.read_table(datastring, sep='\s\s+')
data.set_index(['Country', 'Metric'], inplace=True)
Then data
looks like this:
2011 2012 2013 2014
Country Metric
USA GDP 7 4 0 2
Pop. 2 3 0 3
GB GDP 8 7 0 7
Pop. 2 6 0 0
FR GDP 5 0 0 1
Pop. 1 1 0 5
Now to get the GDPs, you can take a cross-section of the dataframe via the xs
method:
data.xs('GDP', level='Metric')
2011 2012 2013 2014
Country
USA 7 4 0 2
GB 8 7 0 7
FR 5 0 0 1
It's so easy because your data are already pivoted/unstacked. IF they weren't and looked like this:
data.columns.names = ['Year']
data = data.stack()
data
Country Metric Year
USA GDP 2011 7
2012 4
2013 0
2014 2
Pop. 2011 2
2012 3
2013 0
2014 3
GB GDP 2011 8
2012 7
2013 0
2014 7
Pop. 2011 2
2012 6
2013 0
2014 0
FR GDP 2011 5
2012 0
2013 0
2014 1
Pop. 2011 1
2012 1
2013 0
2014 5
You could then use groupby
to tell you something about the world as a whole:
data.groupby(level=['Metric', 'Year']).sum()
Metric Year
GDP 2011 20
2012 11
2013 0
2014 10
Pop. 2011 5
2012 10
2013 0
2014 8
Or get real fancy:
data.groupby(level=['Metric', 'Year']).sum().unstack(level='Metric')
Metric GDP Pop.
Year
2011 20 5
2012 11 10
2013 0 0
2014 10 8
Is this what you are looking for:
df = df.groupby(['Metric'])
df.get_group('GDP')
Country Metric 2011 2012 2013 2014
0 USA GDP 7 4 0 2
2 GB GDP 8 7 0 7
4 FR GDP 5 0 0 1