在 pandas dataframe 中使用 groupby 计算组的百分比

Question

In the following dataframe:

country admin_1 admin_2 year    season_name production  area
A1  B1  C1  1991    Primary 43170   25980
A1  B1  C1  1990    Primary 45624   29820
A1  B1  C1  1989    Primary 56310   31284
A1  B1  C1  1988    Primary 33523   24832
A1  B1  C1  1987    Primary 49388   33479
A1  B1  C1  1986    Primary 36475   27425
A1  B1  C1  1985    Primary 32278   32046
A1  B1  C1  1984    Primary 52073   28929
A1  B1  C1  1983    Primary 51746   32855
A1  B1  C2  1991    Primary 32010   20010
A1  B1  C2  1990    Primary 52704   19520
A1  B1  C2  1989    Primary 65240   18640
A1  B1  C2  1988    Primary 11570   17800
A1  B1  C2  1987    Primary 51282   20350
A1  B1  C2  1986    Primary 25808   19816
A1  B1  C2  1985    Primary 16935   18817
A1  B2  C3  1987    Primary 51282   20350
A1  B2  C3  1986    Primary 25808   19816
A1  B2  C3  1985    Primary 16935   18817

I want to determine the percentage of area for each admin_2 by averaging the area across all years for each admin_2 and them computing the percentage. This is what I tried:

df['area_percentage'] = df.groupby(['country', 'admin_2'])['area'].apply(lambda x: x / x.sum())

Answer 1

Try:

df['area_percentage'] = df['area'] /  df.groupby(['country', 'admin_2'])['area'].transform('sum') * 100 

Answer 2

Well, since the question is somewhat vague, I'll do a sort of mental exercise.

Let's see how we can interpret "the percentage of area for each admin_2 by averaging the area across all years for each admin_2" .

Looking at the first attempt, which is considered wrong, I could come up with something like this:

s = df.groupby(['country', 'admin_2'])['area'].mean()
s /= s.sum() 

Output:

country  admin_2
A1       C1         0.432095
         C2         0.281167
         C3         0.286738

But it looks weird. What could be the benefit of normalizing means? I don't know. So let's skip it and do something different.

What if we sum up the area for all the years grouped by admini_2 and divide the result by the total area ?

s = df.groupby(['country', 'admin_2'])['area'].sum()
s /= s.sum()

Output:

country  admin_2
A1       C1         0.578936
         C2         0.293003
         C3         0.128061

Well, it might work if the nature of area implies the accumulation. In this case, we can talk about the overall percentage of area for each admin_2 .

But what if there's no "accumulation" on the table? Let's say the area means the amount of work that must be done when necessary during the year. In this case we could interpret the request this way: compare area for each admin_2 by year . From the data presented I might conclude that country and admin_1 do not really matter. They are the same for each admin_2 , so I guess they can be ignored (otherwise, we simply add them to the pivot table index). In this case, I'd do something like this:

_df = df.pivot('admin_2','year','area')
_df = (_df / _df.sum()).T

display(_df.fillna(0).style.format('{0:.0%}'.format))
_df.plot(kind='bar'))

Output: