pandas normalization of groupby.sum

Question

I have a pandas dataframe that looks like this:

      **I     SI     weights**
        1     3      0.3  
        2     4      0.2
        1     3      0.5
        1     5      0.5

I need to do this: given a value of I, consider each value of SI and add the total weight. At the end, I should have, for each realization, something like that:

             I = 1     SI = 3      weight = 0.8
                       SI = 5      weight = 0.5

             I = 2     SI = 4      weight = 0.2

This is easily achievable by calling groupby and sum:

       name = ['I', 'SI','weight']
       Location = 'Simulationsdata/prova.csv'
       df = pd.read_csv(Location, names = name,sep='\t',encoding='latin1') 

       results = df.groupby(['I', 'real', 'SI']).weight.sum()

Now I want the weights to be normalized to one, so that it should be something like that:

             I = 1     SI = 3      weight = 0.615
                       SI = 5      weight = 0.385

             I = 2     SI = 4      weight = 1

I tried this:

        for idx2, j in enumerate(results.index.get_level_values(1).unique()):
            norm = [float(i)/sum(results.loc[j]) for i in results.loc[j]]

but when I try to plot, for each I, the distribution of the SI, I find that also the SI are normalized, and I do not want this to happen.

P.s. this question is related to this one but, since it involves another aspect of the problem, I tought that it would be better to ask it separately

Answer 1

You should be able to divide the weight column by its own sum:

# example data
df
   I  SI   weight
0  1   3      0.3
1  2   4      0.2
2  1   3      0.5
3  1   5      0.5

# two-level groupby, with the result as a DataFrame instead of Series:
# df['col'] gives a Series, df[['col']] gives a DF
res = df.groupby(['I', 'SI'])[['weight']].sum()
res
       weight
I SI         
1 3       0.8
  5       0.5
2 4       0.2

# Get the sum of weights for each value of I,
# which will serve as denominators in normalization
denom = res.groupby('I')['weight'].sum()
denom
I
1    1.3
2    0.2
Name: weight, dtype: float64

# Divide each result value by its index-matched
# denominator value
res.weight = res.weight / denom
res
        weight
I SI          
1 3   0.615385
  5   0.384615
2 4   1.000000