如何使用 pandas groupby 函数根据 groupby 值应用公式

Question

My question may be a little confusing, so let me explain. I have a dataframe of information that I would like to group by the unique order id that will produce the following columns:

sum qty = the total amount that was executed per order id. csv = this is the sum of the csv column per order id divided by the sum of the executed amount of the order id.

The first column is easy to create with groupby, it's the second column that I am having issues with. Here is sample data that I am working with:

    qty     sym     price   ordrefno    ord_bidprice    ord_askprice    csv
0   -25000  TEST    0.044   984842      0.0435          0.044          12.5
1   100     TEST    0.0443  984702      0.0435          0.044          0.03
2   -10000  TEST    0.0405  983375      0.039           0.0405         15
3   -100    TEST    0.0443  984842      0.0435          0.044          0.03

This is my code:

cs1 = lambda x: np.sum(test.csv / test.qty)
f2 = {'qty' : ['sum'], 'csv' : {'es' : cs1}}

agg_td = trades.groupby('ordrefno').agg(f2)

Answer 1

Writing a named funtion and using apply works:

def func(group):
    sum_ = group.qty.sum()
    es = (group.csv / group.qty).sum()
    return pd.Series([sum_, es], index=['qty', 'es'])

trades.groupby('ordrefno').apply(func)

Result:

            qty     es
ordrefno               
983375   -10000 -0.0015
984702      100  0.0003
984842   -25100 -0.0008

Answer 2

Assuming you want the ratio of the sums rather than the sum of the ratios (the way the question is worded suggest this but the function in you code would give the sum of the ratios if applied to the df), I think the cleanest way to do this is in two steps. First just get the sum of the two columns and then divide:

agg_td = trades.groupby('ordrefno')[['qty', 'csv']].sum()
agg_td.eval('es = csv/qty')

You could also create a special function and pass it to the groupby apply method:

es = trades.groupby('ordrefno').apply(lambda df: df.csv.sum() / df.qty.sum()) 

But this will only get you the 'es' column. The problem with using agg is that the dict of functions are column-specific where here you need to combine two columns.