Verifying integrity

Joining Data with pandas

Aaren Stubberfield

Instructor

Let's check our data

Possible merging issue:

Two DataFrames put together horizontally with a one-to-many relationship

Unintentional one-to-many relationship
Unintentional many-to-many relationship

Possible concatenating issue:

Two DataFrames vertically joined with duplicate row in second DataFrames

Duplicate records possibly unintentionally introduced

Validating merges

.merge(validate=None):

Checks if merge is of specified type
'one_to_one'
'one_to_many'
'many_to_one'
'many_to_many'

Merge dataset for example

Table Name: tracks

  tid  name             aid  mtid  gid  u_price
0 2    Balls to the...  2    2     1    0.99   
1 3    Fast As a Shark  3    2     1    0.99   
2 4    Restless and...  3    2     1    0.99

Table Name: specs

  tid  milliseconds  bytes  
0 2    342562        5510424
1 3    230619        3990994
2 2    252051        4331779

Merge validate: one_to_one

tracks.merge(specs, on='tid', 
             validate='one_to_one')

Traceback (most recent call last):
MergeError: Merge keys are not unique in right dataset; not a one-to-one merge

Merge validate: one_to_many

albums.merge(tracks, on='aid', 
             validate='one_to_many')

  aid  title            artid  tid  name             mtid  gid  u_price
0 2    Balls to the...  2      2    Balls to the...  2     1    0.99   
1 3    Restless and...  2      3    Fast As a Shark  2     1    0.99   
2 3    Restless and...  2      4    Restless and...  2     1    0.99

Verifying concatenations

.concat(verify_integrity=False):

Check whether the new concatenated index contains duplicates
Default value is False

Dataset for .concat() example

Table Name: inv_feb

     cid  invoice_date  total
iid 
7    38   2009-02-01    1.98 
8    40   2009-02-01    1.98 
9    42   2009-02-02    3.96

Table Name: inv_mar

     cid  invoice_date  total
iid 
9    17   2009-03-04    1.98 
15   19   2009-03-04    1.98 
16   21   2009-03-05    3.96

Verifying concatenation: example

pd.concat([inv_feb, inv_mar], 
          verify_integrity=True)

Traceback (most recent call last):
ValueError: Indexes have overlapping 
values: Int64Index([9], dtype='int64', 
name='iid')

pd.concat([inv_feb, inv_mar], 
          verify_integrity=False)

     cid  invoice_date  total
iid 
7    38   2009-02-01    1.98 
8    40   2009-02-01    1.98 
9    42   2009-02-02    3.96 
9    17   2009-03-04    1.98 
15   19   2009-03-04    1.98 
16   21   2009-03-05    3.96

Why verify integrity and what to do

Why:

Real world data is often NOT clean

What to do:

Fix incorrect data
Drop duplicate rows

Let's practice!

Joining Data with pandas