1
$\begingroup$

i have a pandas dataframe df with one column z filled with set values

i want to drop duplicated rows where 2 rows are considered duplicated version of one another when they have same column z values ( which are sets ).

import pandas as pd

lnks = [ ( 'a' , 'b' , { 'a' , 'b' } ) , ( 'b' , 'c' , { 'b' , 'c' } ) , ( 'b' , 'a' , { 'a' , 'b' } ) ]
lbls = [ 'x' , 'y' , 'z' ]
df = pd.DataFrame.from_records( lnks , columns = lbls )

Trying to drop duplicated rows based on column z values :

df.drop_duplicates( subset = 'z' , keep='first')

And i get the error message :

TypeError: unhashable type: 'set'

Is there a way to drop duplicated rows based on a unhashable typed column ?

$\endgroup$
2
  • $\begingroup$ I assume it is a typo - but there isn't actually a duplicate in row z anyway because one b also has a space: 'b '. $\endgroup$
    – n1k31t4
    Mar 2, 2019 at 20:04
  • $\begingroup$ right. I've made a correction. thx. $\endgroup$ Mar 2, 2019 at 20:41

2 Answers 2

4
$\begingroup$

It is true that a set is not hashable (it cannot be used as a key in a hashmap a.k.a a dictionary). So what you can do is to just convert the column to a type that is hashable - I would go for a tuple.

I made a new column that is just the "z" column you had, converted to tuples. Then you can use the same method you tried to, on the new column:

In [1] : import pandas as pd 
    ...:  
    ...: lnks = [ ( 'a' , 'b' , { 'a' , 'b' } ) , ( 'b' , 'c' , { 'b' , 'c' } ) 
    ...: , ( 'b' , 'a' , { 'a' , 'b' } ) ] 
    ...: lbls = [ 'x' , 'y' , 'z' ] 
    ...: df = pd.DataFrame.from_records( lnks , columns = lbls)                 

In [2]: df["z_tuple"] = df.z.apply(lambda x: tuple(x))                         

In [3]: df.drop_duplicates(subset="z_tuple", keep="first")                     
Out[3]: 
   x  y       z z_tuple
0  a  b  {b, a}  (b, a)
1  b  c  {c, b}  (c, b)

The apply method lets you apply a function to each item in a column, and then returns the values as a new column (a Pandas Series object). This lets you assign it back to the original DataFrame as a new column, as I did.

You can also remove the "z_tuple" column then if you no longer want it:

In [4] : df.drop("z_tuple", axis=1, inplace=True)                               

In [5] : df                                                                     
Out[5] : 
   x  y       z
0  a  b  {b, a}
1  b  c  {c, b}
2  b  a  {b, a}
$\endgroup$
0
$\begingroup$

I have to admit I did not mention the reason why I was trying to drop duplicated rows based on a column containing set values. The reason is that the set { 'a' , 'b' } is the same as { 'b' , 'a' } so 2 apparently different rows are considered the same regarding the set column and are then deduplicated... but this is not possible because sets are unhashable ( like list )

Tuples are hashable but order of their elements matters... so when I build the tuples for each row i sort them :

import pandas as pd

lnks = [ ( 'a' , 'b' ) , ( 'b' , 'c' ) , ( 'b' , 'a' ) , ( 'a' , 'd' ) , ( 'd' , 'e' ) ]
lbls = [ 'x' , 'y' ]
df = pd.DataFrame.from_records( lnks , columns = lbls )

Building the tuple column (each tuple are sorted) :

df[ 'z' ] = df.apply( lambda d : tuple( sorted( [ d[ 'x' ]  , d[ 'y' ] ] ) ) , axis = 1 )

Droping duplicated rows (keeping first occurence) using the new tuple column :

df.drop_duplicates(subset="z", keep="first" , inplace = True ) 
$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.