Pandas to_sql make index unique

Ana picture Ana · Sep 2, 2017 · Viewed 7k times · Source

I have been readin about pandas to_sql solutions to not add duplicate records to a database. I am working with csv files of logs, each time i upload a new log file i then read the data and make some changes with pandas creating a new dataframe. Then i execute to_sql('Logs',con = db.engine, if_exists = 'append', index=True). With the if_exists arg i make sure each time the new created dataframe from the new file is appended to the existing database. The problem is it keeps adding duplicating values. I want to make sure that if a file which has already been uploaded is by mistake uploaded again it won't be appended to the database. I want to try do this directly when creating the database withouth finding a workaround like just checking if the filename has been used before.

I am working with flask-sqlalchemy.

Thank you.

Answer

andrew_reece picture andrew_reece · Sep 2, 2017

Your best bet is to catch duplicates by setting up your index as a primary key, and then using try/except to catch uniqueness violations. You mentioned another post that suggested watching for IntegrityError exceptions, and I agree that's the best approach. You can combine that with a de-deuplication function to make sure your table updates run smoothly.

Demonstrating the problem

Here's a toy example:

from sqlalchemy import *
import sqlite3

# make a database, 'test', and a table, 'foo'.
conn = sqlite3.connect("test.db")
c = conn.cursor()
# id is a primary key.  this will be the index column imported from to_sql().
c.execute('CREATE TABLE foo (id integer PRIMARY KEY, foo integer NOT NULL);')
# use the sqlalchemy engine.
engine = create_engine('sqlite:///test.db')

pd.read_sql("pragma table_info(foo)", con=engine)

   cid name     type  notnull dflt_value  pk
0    0   id  integer        0       None   1
1    1  foo  integer        1       None   0

Now, two example data frames, df and df2:

data = {'foo':[1,2,3]}
df = pd.DataFrame(data)
df
   foo
0    1
1    2
2    3

data2 = {'foo':[3,4,5]}
df2 = pd.DataFrame(data2, index=[2,3,4])
df2
   foo
2    3       # this row is a duplicate of df.iloc[2,:]
3    4
4    5

Move df into table foo:

df.to_sql('foo', con=engine, index=True, index_label='id', if_exists='append')

pd.read_sql('foo', con=engine)
   id  foo
0   0    1
1   1    2
2   2    3

Now, when we try to append df2, we catch the IntegrityError:

try:
    df2.to_sql('foo', con=engine, index=True, index_label='id', if_exists='append')
# use the generic Exception, both IntegrityError and sqlite3.IntegrityError caused trouble.
except Exception as e: 
    print("FAILURE TO APPEND: {}".format(e))

Output:

FAILURE TO APPEND: (sqlite3.IntegrityError) UNIQUE constraint failed: foo.id [SQL: 'INSERT INTO foo (id, foo) VALUES (?, ?)'] [parameters: ((2, 3), (3, 4), (4, 5))]

Suggested Solution

On IntegrityError, you can pull the existing table data, remove the duplicate entries of your new data, and then retry the append statement. Use apply() for this:

def append_db(data):
    try:
        data.to_sql('foo', con=engine, index=True, index_label='id', if_exists='append')
        return 'Success'
    except Exception as e:
        print("Initial failure to append: {}\n".format(e))
        print("Attempting to rectify...")
        existing = pd.read_sql('foo', con=engine)
        to_insert = data.reset_index().rename(columns={'index':'id'})
        mask = ~to_insert.id.isin(existing.id)
        try:
            to_insert.loc[mask].to_sql('foo', con=engine, index=False, if_exists='append')
            print("Successful deduplication.")
        except Exception as e2:
            "Could not rectify duplicate entries. \n{}".format(e2)
        return 'Success after dedupe'

df2.apply(append_db)

Output:

Initial failure to append: (sqlite3.IntegrityError) UNIQUE constraint failed: foo.id [SQL: 'INSERT INTO foo (id, foo) VALUES (?, ?)'] [parameters: ((2, 3), (3, 4), (4, 5))]

Attempting to rectify...
Successful deduplication.

foo    Success after dedupe
dtype: object