Compare values with previous values and count the change : Pandas Groupby-CodePudding

I have a panadas dataframe shown below :

masterid    price
1001        12
1001        12
1001        11
1001        14
1002        15
1003        10
1003        10
1003        10
1004        16
1004        17

I want to count the times when the price changes for a particular master id

Example : For masterid 1001 the price changed from 12 - 11 and from 11 -14 ( 2 times )

Expected output :

masterid    price
1001        2
1002        0
1003        0
1004        1

I have tried using (df['price'].ne(df['price'].shift()) but not sure how to use group by in this case. Thanks in advance

CodePudding user response：

You are almost there. The next step is to use pd.groupby.apply with your existing function and sum:

df.groupby('masterid').apply(
    lambda df: (df['price'].ne(df['price'].shift())
).sum())

This stragegy should work pretty generally, though in some cases apply will be slow.

CodePudding user response：


#calculate the price difference using diff
# take absolute value and compare with zero
# go another group by and take the count

df.assign(price=
          df.groupby('masterid')['price']
          .diff()
          .abs()>0)
.groupby(['masterid'])['price']
.sum()
.reset_index()

masterid    price
0   1001    2
1   1002    0
2   1003    0
3   1004    1