GroupBy columns on column header prefix

user_7442626 0
Viewed 2316

I have a dataframe with column names that start with a set list of prefixes. I want to get the sum of the values in the dataframe grouped by columns that start with the same prefix.

df = pd.DataFrame([[1,2,3,4],[1,2,3,4],[1,2,3,4],[1,2,3,4]],
              columns=['abc', 'abd', 'wxy', 'wxz'])
prefixes = ['ab','wx']
df
    abc abd wxy wxz
0   1   2   3   4
1   1   2   3   4
2   1   2   3   4
3   1   2   3   4

The only way I could figure out how to do it was to loop through the prefix list, get the columns from the dataframe that startwith that string, and then sum the results.

results = []
for p in prefixes:
  results.append([p, df.loc[:, df.columns.str.startswith(p)].values.sum()])
results = pd.DataFrame(results,)
results.set_index(keys=[0], drop=True).T

    ab  wx
1   12  28

I hoped there was a more elegant way to do it, perhaps with groupby(), but I couldn't figure it out.

pandas-groupby pandas dataframe group-by python
4 Answers

First, it is necessary to determine what columns contain what prefix. We then use this to perform a groupby.

grouper = [next(p for p in prefixes if p in c) for c in df.columns]
u = df.groupby(grouper, axis=1).sum()

   ab  wx
0   3   7
1   3   7
2   3   7
3   3   7

Almost there, now,

u.sum().to_frame().T

   ab  wx
0  12  28

Another option is using np.char.startswith and argmax to vectorize:

idx = np.char.startswith(
    df.columns.values[:, None].astype(str), prefixes).argmax(1)

(pd.Series(df.groupby(idx, axis=1).sum().sum().values, index=prefixes)
   .to_frame()
   .transpose())

   ab  wx
0  12  28
user_49090870

Using groupby after slice your columns 

df.groupby(df.columns.str[:-1],axis=1).sum().sum().to_frame().T
Out[317]: 
   ab  wx
0  12  28

Update 

l=sum([[x]*df.columns.str.startswith(x).sum() for x in prefixes],[])
df.groupby(l,axis=1).sum().sum().to_frame().T
Out[329]: 
   ab  wx
0  12  28
user_79645270

I tried the grouper method suggested by @cs95 (on different data), but it missed 4 of the prefixes, where columns beginning with each of those 4 prefixes exist in the dataframe, so 4 columns were missing in the outputted dataframe.

I got different results when I then tried:

df.groupby(df.columns.str[:2], axis=1).sum().reset_index()

This second way includes the missing 4 prefixes (which are derived from df, here hardcoded as the first 2 characters of the column names), but also gives different summation results to the grouper method.

(This other method doesn't include the prefix list that OP specifically asked about)

user_52725910

Try this, this not use groupby:

pd.concat([df.iloc[:,df.columns.str.startswith(prefixes[i])].sum().to_frame().rename(columns = {0:prefixes[i]}).sum() for i in range(len(prefixes))])

@FullMetalScientist could you please kindly validate if this works in your new data frame?

user_140043950
Related