Pandas Groupby Multiple function application with weighted average

Question

I have a DataFrame that look like this:

In [55]:df
Out [55]:

      WEIGHT  TR_YTD  TR_3Y_Ann  Categ1 Categ2 Number
0   0.131214   -0.28       8.49       1      a    0
1   0.052092    1.69      13.70       3      b    1
2   0.045993    2.50        NaN       1      a    2
3   0.041450   -4.57       5.07       1      c    3
4   0.040769    7.64      17.49       2      a    4
5   0.039791    0.07       0.21       1      a    5
6   0.039271   -6.14       8.88       3      a    6
7   0.038340   -8.13        NaN       1      c    7
8   0.038227    9.26      13.78       2      a    8
9   0.033878    0.02      11.45       1      a    9
10  0.029455    5.91      24.86       3      b   10

I have weights, performances and categories (in reality I have many more columns but this is a minimum working example). I would like to:

• group first by each categ alone and then by combinations of the first two categories
• aggregate several columns with different functions using a dictionary (as shown in Wes McKinney's book)
• calculate the weighted average performances using my user defined function
• take into into account NaNs in my data
• do some sorting in my output

The code:

def get_wavg(df, field):
    return np.average(df[field], weights=df['WEIGHT'])

groups = [df['Categ1'], df['Categ2'], [df['Categ1'],df['Categ2']]]   

funcdict = {'Number':'count',
            'WEIGHT': 'sum', \
            'TR_YTD': lambda x: get_wavg(x,'TR_YTD'), \
            'TR_3Y_Ann': lambda x: get_wavg(x,'TR_3Y_Ann')}


for group in groups:  

    # preparing list to sort dynamically
    # sorting by the first layers of groups (excluding last) and then weights
    groupnames = [x.name for x in group]
    sortinglist = groupnames[:-1]
    sortinglist.append('WEIGHTS')
    ascendinglist = [True]*(len(groupnames)-1)+[False]

    # apply agg functions
    grouped = df.groupby(group)
    grouped = grouped.agg(funcdict)

    # sorting
    grouped  = pd.DataFrame(grouped)
    grouped.reset_index(inplace=True)
    grouped.sort_values(by=sortinglist,ascending=ascendinglist,inplace=True)
    grouped.set_index(groupnames,inplace=True)

    print (grouped)

Problem is that I can't find the correct syntax to make it work with my user defined function. If I use np.nanmean instead it works but doesn't give me the result I am looking for.

What is the correct way to do this?