按组滚动的独特分类特征

Question

I would like to get a rolling count of unique categoricals by group:

Group  Item
A      pen
A      pen
A      elbow
A      warthog
B      elbow
B      peach

Should result in:

Group  Item     Unique_item_count
A      pen      1
A      pen      1
A      elbow    2
A      warthog  3
B      elbow    1
B      peach    2

I feel like pd.rolling_count might have the answer, but I haven't figured it out. Thanks for your wisdom and wizardry!

Answer 1

We can GroupBy twice. First we get the nunique values back and second time we get cumsum to make our count go up for each unique value in Item :

Then we merge these results back to our original dataframe.

s = df.groupby(['Group', 'Item'], sort=False)['Item'].nunique().groupby(level=0).cumsum()

final = df.merge(s.reset_index(name='Unique_item_count'), on=['Group', 'Item'])

Output

  Group     Item  Unique_item_count
0     A      pen                  1
1     A      pen                  1
2     A    elbow                  2
3     A  warthog                  3
4     B    elbow                  1
5     B    peach                  2

Answer 2

Approach is same as suggested by Erfan. Just don't have to do merge

df.groupby(['Group', 'Item'], sort=False)['Item'].nunique().groupby(level=0).cumsum().reindex(df).reset_index(name='Unique_count')

Output

  Group     Item    Unique_count
0   A       pen         1
1   A       pen         1
2   A       elbow       2
3   A       warthog     3
4   B       elbow       1
5   B       peach       2