Getting idxmax values in pandas groupby in case of equal values or ties

MTALY 0
Viewed 25

I am trying to identify the most probable event (transportation) of customers. This is an example df:

df = pd.DataFrame({'customer id':[1,1,1,2,2,2,2,2,2,2,3,3],
                   'Trans':['Car','Car','Bus','Bus','Bus','Bus','Car','Car','Car','Plane','Car','Bus']})

gives:

    customer id  Trans
0             1    Car
1             1    Car
2             1    Bus
3             2    Bus
4             2    Bus
5             2    Bus
6             2    Car
7             2    Car
8             2    Car
9             2  Plane
10            3    Car
11            3    Bus

I did the following steps to get the largest probable values for each customer id:

#Get the count & percent of each customer id transportation
df2 = df.groupby(['customer id','Trans'])['Trans'].size().reset_index(name='Trans Counts')
df2['percent'] = df2.groupby("customer id")['Trans Counts'].transform(lambda x: (x / x.sum()).round(2))

#Get records > 1
df3 = df2[df2['Trans Counts'] > 1]

So I got this df:

  customer id Trans  Trans Counts  percent
1            1   Car             2     0.67
2            2   Bus             3     0.43
3            2   Car             3     0.43

There is a tie for customer id 2, so when I use idxmax() :

df3.loc[df3.groupby('customer id')['Trans Counts'].idxmax()]

It only shows the first row:

   customer id Trans  Trans Counts  percent
1            1   Car             2     0.67
2            2   Bus             3     0.43

How can I get the top 2 records within the same group in case of a tie? I also used try nlargest(2) but I got the results for the whole df not within the grouping & if I used it within the aggregate function, it does not show the expected output!

Expected output in case of a tie:

   customer id Trans            Trans Counts  percent
1            1   Car                      2     0.67
2            2   **Bus, Car**             3     0.43

Thanks

max pandas numpy python-3.x python
2 Answers

You could create a "rank" column:

df3["rank"] = df3.groupby("customer id")["Trans Counts"].rank(method="min")
df3
# Out[83]: 
#    customer id Trans  Trans Counts  percent  rank
# 1            1   Car             2     0.67   1.0
# 2            2   Bus             3     0.43   1.0
# 3            2   Car             3     0.43   1.0

Using method="min", the lowest rank is used when values are the same (in this case 1, rather than 2 (max) or 1.5 (average)). Then you could specify df3[df3["rank"] == 1] for those that are most frequent.

You could most likely combine several all these to not use groupby several times within the code to improve efficiency.

Rawson0

This is how you can assign the new column based on the maximum transportation type

In [144]: df2 = pd.get_dummies(df).groupby("customer id").sum()
#             Trans_Bus  Trans_Car  Trans_Plane
#customer id                                   
#1                    1          2            0
#2                    3          3            1
#3                    1          1            0


In [145]: df2['max_cols'] = [ ",".join([col for col in df2 if df2.lo
     ...: c[i,col]==mx]) for i,mx in df2.max(axis=1).items()]

In [146]: df2
Out[146]: 
             Trans_Bus  Trans_Car  Trans_Plane             max_cols
customer id                                                        
1                    1          2            0            Trans_Car
2                    3          3            1  Trans_Bus,Trans_Car
3                    1          1            0  Trans_Bus,Trans_Car

More or less the same approach, but adding percentage and max count as columns:

In [4]: df2 = pd.get_dummies(df).groupby("customer id").sum()

In [5]: trans_count = df2.max(axis=1)

In [6]: trans_perc = trans_count / df2.sum(axis=1)

In [7]: trans_type = [ ",".join([col for col in df2 if df2.loc[i,col]=
   ...: =mx]) for i,mx in trans_count.items()]

In [8]: pd.DataFrame({"count":trans_count, "perc":trans_perc, "kind":t
   ...: rans_type}, index=df2.index)
Out[8]: 
             count      perc                 kind
customer id                                      
1                2  0.666667            Trans_Car
2                3  0.428571  Trans_Bus,Trans_Car
3                1  0.500000  Trans_Bus,Trans_Car

Note, you can call reset_index() on the resulting dataframe to move customer id from index to column

dermen0
Related