Comparison with SQL

Since many potential pandas users have some familiarity with SQL, this page is meant to provide some examples of how various SQL operations would be performed using pandas.

If you’re new to pandas, you might want to first read through 10 Minutes to pandas to familiarize yourself with the library.

As is customary, we import pandas and NumPy as follows:

In [1]: import pandas as pd

In [2]: import numpy as np

Most of the examples will utilize the tips dataset found within pandas tests. We’ll read the data into a DataFrame called tips and assume we have a database table of the same name and structure.

In [3]: url = ('https://raw.github.com/pandas-dev'
   ...:        '/pandas/master/pandas/tests/data/tips.csv')
   ...: 

In [4]: tips = pd.read_csv(url)
---------------------------------------------------------------------------
ConnectionRefusedError                    Traceback (most recent call last)
/usr/lib/python3.8/urllib/request.py in do_open(self, http_class, req, **http_conn_args)
   1349             try:
-> 1350                 h.request(req.get_method(), req.selector, req.data, headers,
   1351                           encode_chunked=req.has_header('Transfer-encoding'))

/usr/lib/python3.8/http/client.py in request(self, method, url, body, headers, encode_chunked)
   1239         """Send a complete request to the server."""
-> 1240         self._send_request(method, url, body, headers, encode_chunked)
   1241 

/usr/lib/python3.8/http/client.py in _send_request(self, method, url, body, headers, encode_chunked)
   1285             body = _encode(body, 'body')
-> 1286         self.endheaders(body, encode_chunked=encode_chunked)
   1287 

/usr/lib/python3.8/http/client.py in endheaders(self, message_body, encode_chunked)
   1234             raise CannotSendHeader()
-> 1235         self._send_output(message_body, encode_chunked=encode_chunked)
   1236 

/usr/lib/python3.8/http/client.py in _send_output(self, message_body, encode_chunked)
   1005         del self._buffer[:]
-> 1006         self.send(msg)
   1007 

/usr/lib/python3.8/http/client.py in send(self, data)
    945             if self.auto_open:
--> 946                 self.connect()
    947             else:

/usr/lib/python3.8/http/client.py in connect(self)
   1401 
-> 1402             super().connect()
   1403 

/usr/lib/python3.8/http/client.py in connect(self)
    916         """Connect to the host and port specified in __init__."""
--> 917         self.sock = self._create_connection(
    918             (self.host,self.port), self.timeout, self.source_address)

/usr/lib/python3.8/socket.py in create_connection(address, timeout, source_address)
    807         try:
--> 808             raise err
    809         finally:

/usr/lib/python3.8/socket.py in create_connection(address, timeout, source_address)
    795                 sock.bind(source_address)
--> 796             sock.connect(sa)
    797             # Break explicitly a reference cycle

ConnectionRefusedError: [Errno 111] Connection refused

During handling of the above exception, another exception occurred:

URLError                                  Traceback (most recent call last)
<ipython-input-4-8ab2297b7141> in <module>
----> 1 tips = pd.read_csv(url)

/usr/lib/python3/dist-packages/pandas/io/parsers.py in parser_f(filepath_or_buffer, sep, delimiter, header, names, index_col, usecols, squeeze, prefix, mangle_dupe_cols, dtype, engine, converters, true_values, false_values, skipinitialspace, skiprows, skipfooter, nrows, na_values, keep_default_na, na_filter, verbose, skip_blank_lines, parse_dates, infer_datetime_format, keep_date_col, date_parser, dayfirst, cache_dates, iterator, chunksize, compression, thousands, decimal, lineterminator, quotechar, quoting, doublequote, escapechar, comment, encoding, dialect, error_bad_lines, warn_bad_lines, delim_whitespace, low_memory, memory_map, float_precision)
    683         )
    684 
--> 685         return _read(filepath_or_buffer, kwds)
    686 
    687     parser_f.__name__ = name

/usr/lib/python3/dist-packages/pandas/io/parsers.py in _read(filepath_or_buffer, kwds)
    437     # though mypy handling of conditional imports is difficult.
    438     # See https://github.com/python/mypy/issues/1297
--> 439     fp_or_buf, _, compression, should_close = get_filepath_or_buffer(
    440         filepath_or_buffer, encoding, compression
    441     )

/usr/lib/python3/dist-packages/pandas/io/common.py in get_filepath_or_buffer(filepath_or_buffer, encoding, compression, mode)
    194 
    195     if _is_url(filepath_or_buffer):
--> 196         req = urlopen(filepath_or_buffer)
    197         content_encoding = req.headers.get("Content-Encoding", None)
    198         if content_encoding == "gzip":

/usr/lib/python3.8/urllib/request.py in urlopen(url, data, timeout, cafile, capath, cadefault, context)
    220     else:
    221         opener = _opener
--> 222     return opener.open(url, data, timeout)
    223 
    224 def install_opener(opener):

/usr/lib/python3.8/urllib/request.py in open(self, fullurl, data, timeout)
    523 
    524         sys.audit('urllib.Request', req.full_url, req.data, req.headers, req.get_method())
--> 525         response = self._open(req, data)
    526 
    527         # post-process response

/usr/lib/python3.8/urllib/request.py in _open(self, req, data)
    540 
    541         protocol = req.type
--> 542         result = self._call_chain(self.handle_open, protocol, protocol +
    543                                   '_open', req)
    544         if result:

/usr/lib/python3.8/urllib/request.py in _call_chain(self, chain, kind, meth_name, *args)
    500         for handler in handlers:
    501             func = getattr(handler, meth_name)
--> 502             result = func(*args)
    503             if result is not None:
    504                 return result

/usr/lib/python3.8/urllib/request.py in https_open(self, req)
   1391 
   1392         def https_open(self, req):
-> 1393             return self.do_open(http.client.HTTPSConnection, req,
   1394                 context=self._context, check_hostname=self._check_hostname)
   1395 

/usr/lib/python3.8/urllib/request.py in do_open(self, http_class, req, **http_conn_args)
   1351                           encode_chunked=req.has_header('Transfer-encoding'))
   1352             except OSError as err: # timeout error
-> 1353                 raise URLError(err)
   1354             r = h.getresponse()
   1355         except:

URLError: <urlopen error [Errno 111] Connection refused>

In [5]: tips.head()
---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-5-53a0cf752a4f> in <module>
----> 1 tips.head()

NameError: name 'tips' is not defined

SELECT

In SQL, selection is done using a comma-separated list of columns you’d like to select (or a * to select all columns):

SELECT total_bill, tip, smoker, time
FROM tips
LIMIT 5;

With pandas, column selection is done by passing a list of column names to your DataFrame:

In [6]: tips[['total_bill', 'tip', 'smoker', 'time']].head(5)
---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-6-0de747740afe> in <module>
----> 1 tips[['total_bill', 'tip', 'smoker', 'time']].head(5)

NameError: name 'tips' is not defined

Calling the DataFrame without the list of column names would display all columns (akin to SQL’s *).

WHERE

Filtering in SQL is done via a WHERE clause.

SELECT *
FROM tips
WHERE time = 'Dinner'
LIMIT 5;

DataFrames can be filtered in multiple ways; the most intuitive of which is using boolean indexing.

In [7]: tips[tips['time'] == 'Dinner'].head(5)
---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-7-fc48c260e052> in <module>
----> 1 tips[tips['time'] == 'Dinner'].head(5)

NameError: name 'tips' is not defined

The above statement is simply passing a Series of True/False objects to the DataFrame, returning all rows with True.

In [8]: is_dinner = tips['time'] == 'Dinner'
---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-8-dfe63ed92808> in <module>
----> 1 is_dinner = tips['time'] == 'Dinner'

NameError: name 'tips' is not defined

In [9]: is_dinner.value_counts()
---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-9-900d99c3802f> in <module>
----> 1 is_dinner.value_counts()

NameError: name 'is_dinner' is not defined

In [10]: tips[is_dinner].head(5)
---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-10-7cfab4b2f27f> in <module>
----> 1 tips[is_dinner].head(5)

NameError: name 'tips' is not defined

Just like SQL’s OR and AND, multiple conditions can be passed to a DataFrame using | (OR) and & (AND).

-- tips of more than $5.00 at Dinner meals
SELECT *
FROM tips
WHERE time = 'Dinner' AND tip > 5.00;
# tips of more than $5.00 at Dinner meals
In [11]: tips[(tips['time'] == 'Dinner') & (tips['tip'] > 5.00)]
---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-11-4ef9dbe7035d> in <module>
----> 1 tips[(tips['time'] == 'Dinner') & (tips['tip'] > 5.00)]

NameError: name 'tips' is not defined
-- tips by parties of at least 5 diners OR bill total was more than $45
SELECT *
FROM tips
WHERE size >= 5 OR total_bill > 45;
# tips by parties of at least 5 diners OR bill total was more than $45
In [12]: tips[(tips['size'] >= 5) | (tips['total_bill'] > 45)]
---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-12-4338232c82b5> in <module>
----> 1 tips[(tips['size'] >= 5) | (tips['total_bill'] > 45)]

NameError: name 'tips' is not defined

NULL checking is done using the notna() and isna() methods.

In [13]: frame = pd.DataFrame({'col1': ['A', 'B', np.NaN, 'C', 'D'],
   ....:                       'col2': ['F', np.NaN, 'G', 'H', 'I']})
   ....: 

In [14]: frame
Out[14]: 
  col1 col2
0    A    F
1    B  NaN
2  NaN    G
3    C    H
4    D    I

Assume we have a table of the same structure as our DataFrame above. We can see only the records where col2 IS NULL with the following query:

SELECT *
FROM frame
WHERE col2 IS NULL;
In [15]: frame[frame['col2'].isna()]
Out[15]: 
  col1 col2
1    B  NaN

Getting items where col1 IS NOT NULL can be done with notna().

SELECT *
FROM frame
WHERE col1 IS NOT NULL;
In [16]: frame[frame['col1'].notna()]
Out[16]: 
  col1 col2
0    A    F
1    B  NaN
3    C    H
4    D    I

GROUP BY

In pandas, SQL’s GROUP BY operations are performed using the similarly named groupby() method. groupby() typically refers to a process where we’d like to split a dataset into groups, apply some function (typically aggregation) , and then combine the groups together.

A common SQL operation would be getting the count of records in each group throughout a dataset. For instance, a query getting us the number of tips left by sex:

SELECT sex, count(*)
FROM tips
GROUP BY sex;
/*
Female     87
Male      157
*/

The pandas equivalent would be:

In [17]: tips.groupby('sex').size()
---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-17-02c58371e730> in <module>
----> 1 tips.groupby('sex').size()

NameError: name 'tips' is not defined

Notice that in the pandas code we used size() and not count(). This is because count() applies the function to each column, returning the number of not null records within each.

In [18]: tips.groupby('sex').count()
---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-18-a9454ebb58fc> in <module>
----> 1 tips.groupby('sex').count()

NameError: name 'tips' is not defined

Alternatively, we could have applied the count() method to an individual column:

In [19]: tips.groupby('sex')['total_bill'].count()
---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-19-adc5242bcc56> in <module>
----> 1 tips.groupby('sex')['total_bill'].count()

NameError: name 'tips' is not defined

Multiple functions can also be applied at once. For instance, say we’d like to see how tip amount differs by day of the week - agg() allows you to pass a dictionary to your grouped DataFrame, indicating which functions to apply to specific columns.

SELECT day, AVG(tip), COUNT(*)
FROM tips
GROUP BY day;
/*
Fri   2.734737   19
Sat   2.993103   87
Sun   3.255132   76
Thur  2.771452   62
*/
In [20]: tips.groupby('day').agg({'tip': np.mean, 'day': np.size})
---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-20-3f94d9cd254c> in <module>
----> 1 tips.groupby('day').agg({'tip': np.mean, 'day': np.size})

NameError: name 'tips' is not defined

Grouping by more than one column is done by passing a list of columns to the groupby() method.

SELECT smoker, day, COUNT(*), AVG(tip)
FROM tips
GROUP BY smoker, day;
/*
smoker day
No     Fri      4  2.812500
       Sat     45  3.102889
       Sun     57  3.167895
       Thur    45  2.673778
Yes    Fri     15  2.714000
       Sat     42  2.875476
       Sun     19  3.516842
       Thur    17  3.030000
*/
In [21]: tips.groupby(['smoker', 'day']).agg({'tip': [np.size, np.mean]})
---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-21-e5dc526454fc> in <module>
----> 1 tips.groupby(['smoker', 'day']).agg({'tip': [np.size, np.mean]})

NameError: name 'tips' is not defined

JOIN

JOINs can be performed with join() or merge(). By default, join() will join the DataFrames on their indices. Each method has parameters allowing you to specify the type of join to perform (LEFT, RIGHT, INNER, FULL) or the columns to join on (column names or indices).

In [22]: df1 = pd.DataFrame({'key': ['A', 'B', 'C', 'D'],
   ....:                     'value': np.random.randn(4)})
   ....: 

In [23]: df2 = pd.DataFrame({'key': ['B', 'D', 'D', 'E'],
   ....:                     'value': np.random.randn(4)})
   ....: 

Assume we have two database tables of the same name and structure as our DataFrames.

Now let’s go over the various types of JOINs.

INNER JOIN

SELECT *
FROM df1
INNER JOIN df2
  ON df1.key = df2.key;
# merge performs an INNER JOIN by default
In [24]: pd.merge(df1, df2, on='key')
Out[24]: 
  key   value_x   value_y
0   B -0.282863  1.212112
1   D -1.135632 -0.173215
2   D -1.135632  0.119209

merge() also offers parameters for cases when you’d like to join one DataFrame’s column with another DataFrame’s index.

In [25]: indexed_df2 = df2.set_index('key')

In [26]: pd.merge(df1, indexed_df2, left_on='key', right_index=True)
Out[26]: 
  key   value_x   value_y
1   B -0.282863  1.212112
3   D -1.135632 -0.173215
3   D -1.135632  0.119209

LEFT OUTER JOIN

-- show all records from df1
SELECT *
FROM df1
LEFT OUTER JOIN df2
  ON df1.key = df2.key;
# show all records from df1
In [27]: pd.merge(df1, df2, on='key', how='left')
Out[27]: 
  key   value_x   value_y
0   A  0.469112       NaN
1   B -0.282863  1.212112
2   C -1.509059       NaN
3   D -1.135632 -0.173215
4   D -1.135632  0.119209

RIGHT JOIN

-- show all records from df2
SELECT *
FROM df1
RIGHT OUTER JOIN df2
  ON df1.key = df2.key;
# show all records from df2
In [28]: pd.merge(df1, df2, on='key', how='right')
Out[28]: 
  key   value_x   value_y
0   B -0.282863  1.212112
1   D -1.135632 -0.173215
2   D -1.135632  0.119209
3   E       NaN -1.044236

FULL JOIN

pandas also allows for FULL JOINs, which display both sides of the dataset, whether or not the joined columns find a match. As of writing, FULL JOINs are not supported in all RDBMS (MySQL).

-- show all records from both tables
SELECT *
FROM df1
FULL OUTER JOIN df2
  ON df1.key = df2.key;
# show all records from both frames
In [29]: pd.merge(df1, df2, on='key', how='outer')
Out[29]: 
  key   value_x   value_y
0   A  0.469112       NaN
1   B -0.282863  1.212112
2   C -1.509059       NaN
3   D -1.135632 -0.173215
4   D -1.135632  0.119209
5   E       NaN -1.044236

UNION

UNION ALL can be performed using concat().

In [30]: df1 = pd.DataFrame({'city': ['Chicago', 'San Francisco', 'New York City'],
   ....:                     'rank': range(1, 4)})
   ....: 

In [31]: df2 = pd.DataFrame({'city': ['Chicago', 'Boston', 'Los Angeles'],
   ....:                     'rank': [1, 4, 5]})
   ....: 
SELECT city, rank
FROM df1
UNION ALL
SELECT city, rank
FROM df2;
/*
         city  rank
      Chicago     1
San Francisco     2
New York City     3
      Chicago     1
       Boston     4
  Los Angeles     5
*/
In [32]: pd.concat([df1, df2])
Out[32]: 
            city  rank
0        Chicago     1
1  San Francisco     2
2  New York City     3
0        Chicago     1
1         Boston     4
2    Los Angeles     5

SQL’s UNION is similar to UNION ALL, however UNION will remove duplicate rows.

SELECT city, rank
FROM df1
UNION
SELECT city, rank
FROM df2;
-- notice that there is only one Chicago record this time
/*
         city  rank
      Chicago     1
San Francisco     2
New York City     3
       Boston     4
  Los Angeles     5
*/

In pandas, you can use concat() in conjunction with drop_duplicates().

In [33]: pd.concat([df1, df2]).drop_duplicates()
Out[33]: 
            city  rank
0        Chicago     1
1  San Francisco     2
2  New York City     3
1         Boston     4
2    Los Angeles     5

Pandas equivalents for some SQL analytic and aggregate functions

Top N rows with offset

-- MySQL
SELECT * FROM tips
ORDER BY tip DESC
LIMIT 10 OFFSET 5;
In [34]: tips.nlargest(10 + 5, columns='tip').tail(10)
---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-34-470f6bc86f8f> in <module>
----> 1 tips.nlargest(10 + 5, columns='tip').tail(10)

NameError: name 'tips' is not defined

Top N rows per group

-- Oracle's ROW_NUMBER() analytic function
SELECT * FROM (
  SELECT
    t.*,
    ROW_NUMBER() OVER(PARTITION BY day ORDER BY total_bill DESC) AS rn
  FROM tips t
)
WHERE rn < 3
ORDER BY day, rn;
In [35]: (tips.assign(rn=tips.sort_values(['total_bill'], ascending=False)
   ....:                     .groupby(['day'])
   ....:                     .cumcount() + 1)
   ....:      .query('rn < 3')
   ....:      .sort_values(['day', 'rn']))
   ....: 
---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-35-8f10c7fd9b13> in <module>
----> 1 (tips.assign(rn=tips.sort_values(['total_bill'], ascending=False)
      2                     .groupby(['day'])
      3                     .cumcount() + 1)
      4      .query('rn < 3')
      5      .sort_values(['day', 'rn']))

NameError: name 'tips' is not defined

the same using rank(method=’first’) function

In [36]: (tips.assign(rnk=tips.groupby(['day'])['total_bill']
   ....:                      .rank(method='first', ascending=False))
   ....:      .query('rnk < 3')
   ....:      .sort_values(['day', 'rnk']))
   ....: 
---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-36-59cdc5622ab0> in <module>
----> 1 (tips.assign(rnk=tips.groupby(['day'])['total_bill']
      2                      .rank(method='first', ascending=False))
      3      .query('rnk < 3')
      4      .sort_values(['day', 'rnk']))

NameError: name 'tips' is not defined
-- Oracle's RANK() analytic function
SELECT * FROM (
  SELECT
    t.*,
    RANK() OVER(PARTITION BY sex ORDER BY tip) AS rnk
  FROM tips t
  WHERE tip < 2
)
WHERE rnk < 3
ORDER BY sex, rnk;

Let’s find tips with (rank < 3) per gender group for (tips < 2). Notice that when using rank(method='min') function rnk_min remains the same for the same tip (as Oracle’s RANK() function)

In [37]: (tips[tips['tip'] < 2]
   ....:     .assign(rnk_min=tips.groupby(['sex'])['tip']
   ....:                         .rank(method='min'))
   ....:     .query('rnk_min < 3')
   ....:     .sort_values(['sex', 'rnk_min']))
   ....: 
---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-37-c3c42fb11f53> in <module>
----> 1 (tips[tips['tip'] < 2]
      2     .assign(rnk_min=tips.groupby(['sex'])['tip']
      3                         .rank(method='min'))
      4     .query('rnk_min < 3')
      5     .sort_values(['sex', 'rnk_min']))

NameError: name 'tips' is not defined

UPDATE

UPDATE tips
SET tip = tip*2
WHERE tip < 2;
In [38]: tips.loc[tips['tip'] < 2, 'tip'] *= 2
---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-38-f7e81bb0cba5> in <module>
----> 1 tips.loc[tips['tip'] < 2, 'tip'] *= 2

NameError: name 'tips' is not defined

DELETE

DELETE FROM tips
WHERE tip > 9;

In pandas we select the rows that should remain, instead of deleting them

In [39]: tips = tips.loc[tips['tip'] <= 9]
---------------------------------------------------------------------------
NameError                                 Traceback (most recent call last)
<ipython-input-39-e3d668c4397d> in <module>
----> 1 tips = tips.loc[tips['tip'] <= 9]

NameError: name 'tips' is not defined
Scroll To Top