Jev Jev - 2 months ago 15
Python Question

Pandas : Delete rows based on other rows

I have a pandas dataframe which looks like that :

qseqid sseqid qstart qend
2 1 125 345
4 1 150 320
3 2 150 450
6 2 25 300
8 2 50 500


I would like to remove rows based on other rows values with these criterias : A row (r1) must be removed if another row (r2) exist with the same
sseqid
and
r1[qstart] > r2[qstart]
and
r1[qend] < r2[qend]
.

Is this possible with pandas ?

Answer
df  = pd.DataFrame({'qend': [345, 320, 450, 300, 500],
 'qseqid': [2, 4, 3, 6, 8],
 'qstart': [125, 150, 150, 25, 50],
 'sseqid': [1, 1, 2, 2, 2]})

def remove_rows(df):
    merged = pd.merge(df.reset_index(), df, on='sseqid')
    mask = ((merged['qstart_x'] > merged['qstart_y']) 
            & (merged['qend_x'] < merged['qend_y']))
    df_mask = ~df.index.isin(merged.loc[mask, 'index'].values)
    result = df.loc[df_mask]
    return result

result = remove_rows(df)
print(result)

yields

   qend  qseqid  qstart  sseqid
0   345       2     125       1
3   300       6      25       2
4   500       8      50       2

The idea is to use pd.merge to form a DataFrame with every pairing of rows with the same sseqid:

In [78]: pd.merge(df.reset_index(), df, on='sseqid')
Out[78]: 
    index  qend_x  qseqid_x  qstart_x  sseqid  qend_y  qseqid_y  qstart_y
0       0     345         2       125       1     345         2       125
1       0     345         2       125       1     320         4       150
2       1     320         4       150       1     345         2       125
3       1     320         4       150       1     320         4       150
4       2     450         3       150       2     450         3       150
5       2     450         3       150       2     300         6        25
6       2     450         3       150       2     500         8        50
7       3     300         6        25       2     450         3       150
8       3     300         6        25       2     300         6        25
9       3     300         6        25       2     500         8        50
10      4     500         8        50       2     450         3       150
11      4     500         8        50       2     300         6        25
12      4     500         8        50       2     500         8        50

Each row of merged contains data from two rows of df. You can then compare every two rows using

mask = ((merged['qstart_x'] > merged['qstart_y']) 
        & (merged['qend_x'] < merged['qend_y']))

and find the labels in df.index that do not match this condition:

df_mask = ~df.index.isin(merged.loc[mask, 'index'].values)

and select those rows:

result = df.loc[df_mask]

Note that this assumes df has a unique index.