Pandas - sample many groups with different proportions

Question

I need to sample a dataframe by group using different proportions for each group. I have more than 100 groups, but for sake of simplicity my example has just 3 groups. Let's suppose I have this dataframe:

df2 = pd.DataFrame({'group_id': np.repeat(['A', 'B', 'C'], (40, 60, 20)),
               'vals': np.random.randn(120)})
N = len(df2)
df2.groupby('group_id').count()

#           vals
#group_id   
#A         40
#B         60
#C         20

And I want to sample groups A, B and C using the dataframe below for proportion:

 prop = pd.DataFrame({'A': {0.45},
                      'B': {0.55},
                      'C': {0.62}})

When I try to sample, I get an error:

grouped = df2.groupby('group_id')
x = grouped.apply(lambda x: x.sample(frac=props))

error: NameError: global name 'props' is not defined

Any help is highly appreciated! Thanks

Answer 1

I think need DataFrame of scalar and then lookup by x.name:

prop = pd.DataFrame({'A': [0.45],
                      'B': [0.55],
                      'C': [0.62]})

grouped = df2.groupby('group_id')
x = grouped.apply(lambda x: x.sample(frac=prop[x.name]))
print (x.head(20))
            group_id      vals
group_id                      
A        19        A  1.157552
         37        A  0.086347
         0         A -0.668129
         8         A -0.345811
         27        A -0.301085
         14        A -0.325130
         6         A  0.301966
         15        A  1.944702
         4         A  1.350509
         1         A -0.498210
         2         A  0.618576
         31        A -0.274381
         16        A  1.915676
         25        A  0.136372
         32        A  0.864837
         9         A -0.315231
         20        A -0.106208
         34        A  1.324797
B        85        B -0.861647
         55        B -0.079275