Is it possible to do a groupby transform with custom functions?
data = {
'a':['a1','a2','a3','a4','a5'],
'b':['b1','b1','b2','b2','b1'],
'c':[55,44.2,33.3,-66.5,0],
'd':[10,100,1000,10000,100000],
}
import pandas as pd
df = pd.DataFrame.from_dict(data)
df['e'] = df.groupby(['b'])['c'].transform(sum) #this works as expected
print (df)
# a b c d e
#0 a1 b1 55.0 10 99.2
#1 a2 b1 44.2 100 99.2
#2 a3 b2 33.3 1000 -33.2
#3 a4 b2 -66.5 10000 -33.2
#4 a5 b1 0.0 100000 99.2
def custom_calc(x, y):
return (x * y)
#obviously wrong code here
df['e'] = df.groupby(['b'])['c'].transform(custom_calc(df['c'], df['d']))
As we can see from the above example, what I want is to explore the possibility of being able to pass in a custom function into .transform()
.
I am aware that .apply()
exists, but I want to find out if it is possible to use .transform()
exclusively.
More importantly, I want to understand how to formulate a proper function that can be passed into .transform()
for it to apply correctly.
P.S. Currently, I know default functions like 'count'
, sum
, 'sum'
, etc works.
One way I like to see what is happening is by creating a small custom function and printing out what is passed and its type. Then, you can see you have to work with.
def f(x):
print(type(x))
print('\n')
print(x)
print(x.index)
return df.loc[x.index,'d']*x
df['f'] = df.groupby('b')['c'].transform(f)
print(df)
#Output from print statements in function
<class 'pandas.core.series.Series'>
0 55.0
1 44.2
4 0.0
Name: b1, dtype: float64
Int64Index([0, 1, 4], dtype='int64')
<class 'pandas.core.series.Series'>
2 33.3
3 -66.5
Name: b2, dtype: float64
Int64Index([2, 3], dtype='int64')
#End output from print statements in custom function
a b c d e f
0 a1 b1 55.0 10 99.2 550.0
1 a2 b1 44.2 100 99.2 4420.0
2 a3 b2 33.3 1000 -33.2 33300.0
3 a4 b2 -66.5 10000 -33.2 -665000.0
4 a5 b1 0.0 100000 99.2 0.0
Here, I am transforming on column 'c' but I make an "extranal" call to the dataframe object in my custom function to get 'd'.
You can also pass the "external" to be used as an argument like this:
def f(x, col):
return df.loc[x.index, col]*x
df['g'] = df.groupby('b')['c'].transform(f, col='d')
print(df)
Output:
a b c d e f g
0 a1 b1 55.0 10 99.2 550.0 550.0
1 a2 b1 44.2 100 99.2 4420.0 4420.0
2 a3 b2 33.3 1000 -33.2 33300.0 33300.0
3 a4 b2 -66.5 10000 -33.2 -665000.0 -665000.0
4 a5 b1 0.0 100000 99.2 0.0 0.0
If you love us? You can donate to us via Paypal or buy me a coffee so we can maintain and grow! Thank you!
Donate Us With