Let's assume we have a table:
df = pd.DataFrame({'A' : ['foo', 'bar', 'foo', 'bar', 'foo', 'bar', 'foo', 'foo'],
                   'B' : ['one', 'one', 'two', 'three', 'two', 'two', 'one', 'three'],
                   'C' : np.random.randn(8), 'D' : np.random.randn(8)})
Output:
      A       B          C           D
0    foo     one    -1.304026    0.237045
1    bar     one     0.030488   -0.672931
2    foo     two     0.530976   -0.669559
3    bar     three  -0.004624   -1.604039
4    foo     two    -0.247809   -1.571291
5    bar     two    -0.570580    1.454514
6    foo     one     1.441081    0.096880
7    foo     three   0.296377    1.575791
I want to count how many positive and negative numbers in column C belong to each group in column A and in what proportion. There are much more groups in A than foo and bar, so group names shouldn't be in the code.
I was trying to groupby A and then filter, but didn't find the right way. Also tried to aggregate with some smart lambda, but didn't succeed.
You could do this as a one line apply (the first column being negative, the second positive):
In [11]: df.groupby('A').C.apply(lambda x: pd.Series([(x < 0).sum(), (x >= 0).sum()])).unstack()
Out[111]: 
     0  1
A        
bar  2  1
foo  2  3
[2 rows x 2 columns]
However, I think a neater way is to use a dummy column and use value_counts:
In [21]: df['C_sign'] = np.sign(df.C)
In [22]: df.groupby('A').C_sign.value_counts()
Out[22]: 
A      
bar  -1    2
      1    1
foo   1    3
     -1    2
dtype: int64
In [23]: df.groupby('A').C_sign.value_counts().unstack()
Out[23]: 
     -1   1
A          
bar   2   1
foo   2   3
[2 rows x 2 columns]
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With