CODEWITHSUNDEEP
pythonpandasdataframe
Blue482
Blue482

Reputation: 3146

Removing rows following (and including) the first occurrence of a particular column value

I have a very large pd.Dataframe contains millions of records where PID and Ses_ID are both index columns, and Var_3 indicates the occurrence of some event.

PID Ses_ID Var_1 Var_2 Var_3
001 001
002
003		 0.7
0.8
0.9		 0.5
0.4
0.3		 0
1
0
002 004
005
006
007
008		 0.8
0.7
0.8
0.2
0.8		 0.2
0.1
0.7
0.2
0.2		 0
0
1
0
1

I want to remove/filter out sessions following and including the first occurrence of Var_3==1 from each person's (indexed by PID) records. Thus the provided example would result as:

PID Ses_ID Var_1 Var_2 Var_3
001 001 0.7 0.5 0
002 004
005		 0.8
0.7		 0.2
0.1		 0
0

I could iteratively add relevant sessions and corresponding PID to a new dataframe but that would be extremely time-consuming given the size of the current dataframe. What would be an efficient way of achieving this? Many thanks!

Updated situation: I have found many rows have the same Ses_ID. How do I remove sessions following (and including) the first occurrence of a particular column value? So for the example below, both rows for Ses_ID==005 would be removed because the event of Var_3==1 occurred in this session.

PID Ses_ID Var_1 Var_2 Var_3
001 001
002
003		 0.7
0.8
0.9		 0.5
0.4
0.3		 0
1
0
002 009
004
004
005
005
006
007		 0.1
0.8
0.8
0.7
0.8
0.2
0.8		 0.3
0.1
0.2
0.1
0.7
0.2
0.2		 0
0
0
0
1
0
1

should be transformed to:

PID Ses_ID Var_1 Var_2 Var_3
001 001 0.7 0.5 0
002 009
004
004		 0.1
0.8
0.8		 0.3
0.1
0.2		 0
0
0

Upvotes: 2

Views: 275

Answers (1)

Andrej Kesely
Andrej Kesely

Reputation: 195428

You can try to use boolean indexing:

# assuming PID, Ses_ID are indices:
mask = df.groupby(level=0)["Var_3"].cumsum().eq(0)
print(df[mask])

Prints:

            Var_1  Var_2  Var_3
PID Ses_ID                     
1   1         0.7    0.5      0
2   4         0.8    0.2      0
    5         0.7    0.1      0

EDIT:

g = df.groupby(level=0)
df["Var_3"] = g["Var_3"].transform(
    lambda x: x.groupby(level=1).transform(sorted, reverse=True)
)
mask = g["Var_3"].cumsum().eq(0)
print(df[mask])

Prints:

            Var_1  Var_2  Var_3
PID Ses_ID                     
1   1         0.7    0.5      0
2   4         0.8    0.2      0

Upvotes: 2

Related Questions