The Yhat Blog


machine learning, data science, engineering


pandasql: Make python speak SQL

by Yhat |


Introduction

One of my favorite things about Python is that users get the benefit of observing the R community and then emulating the best parts of it. I'm a big believer that a language is only as helpful as its libraries and tools.

This post is about pandasql, a Python package we (Yhat) wrote that emulates the R package sqldf. It's a small but mighty library comprised of just 358 lines of code. The idea of pandasql is to make Python speak SQL. For those of you who come from a SQL-first background or still "think in SQL", pandasql is a nice way to take advantage of the strengths of both languages.

In this introduction, we'll show you to get up and running with pandasql inside of Rodeo, the integrated development environment (IDE) we built for data exploration and analysis. Rodeo is an open source and completely free tool. If you're an R user, its a comparable tool with a similar feel to RStudio. As of today, Rodeo can only run Python code, but last week we added syntax highlighting for a bunch of other languages to the editor (markdown, JSON, julia, SQL, markdown). As you may have read or guessed, we've got big plans for Rodeo, including adding SQL support so that you can run your SQL queries right inside of Rodeo, even without our handy little pandasql. More on that in the next week or two!

Downloading Rodeo

Start by downloading Rodeo for Mac, Windows or Linux from the Rodeo page on the Yhat website.

ps If you download Rodeo and encounter a problem or simply have a question, we monitor our discourse forum 24/7 (okay, almost).

A bit of background, if you're curious

Behind the scenes, pandasql uses the pandas.io.sql module to transfer data between DataFrame and SQLite databases. Operations are performed in SQL, the results returned, and the database is then torn down. The library makes heavy use of pandas write_frame and frame_query, two functions which let you read and write to/from pandas and (most) any SQL database.

Install pandasql

Install pandasql using the package manager pane in Rodeo. Simply search for pandasql and click Install Package.

You can also run ! pip install pandasql from the text editor if you prefer to install that way.

Check out the datasets

pandasql has two built-in datasets which we'll use for the examples below.

  • meat: Dataset from the U.S. Dept. of Agriculture containing metrics on livestock, dairy, and poultry outlook and production
  • births: Dataset from the United Nations Statistics Division containing demographic statistics on live births by month

Run the following code to check out the data sets.

#Checking out meat and birth data
from pandasql import sqldf
from pandasql import load_meat, load_births

meat = load_meat()
births = load_births()

#You can inspect the dataframes directly if you're using Rodeo
#These print statements are here just in case you want to check out your data in the editor, too
print meat.head()
print births.head()

Inside Rodeo, you really don't even need the print.variable.head() statements, since you can actually just examine the dataframes directly.

An odd graph

# Let's make a graph to visualize the data
# Bet you haven't had a title quite like this before
import matplotlib.pyplot as plt
from pandasql import *
import pandas as pd

pysqldf = lambda q: sqldf(q, globals())

q  = """
SELECT
  m.date
  , m.beef
  , b.births
FROM
  meat m
LEFT JOIN
  births b
    ON m.date = b.date
WHERE
    m.date > '1974-12-31';
"""

meat = load_meat()
births = load_births()

df = pysqldf(q)
df.births = df.births.fillna(method='backfill')

fig = plt.figure()
ax1 = fig.add_subplot(111)
ax1.plot(pd.rolling_mean(df['beef'], 12), color='b')
ax1.set_xlabel('months since 1975')
ax1.set_ylabel('cattle slaughtered', color='b')

ax2 = ax1.twinx()
ax2.plot(pd.rolling_mean(df['births'], 12), color='r')
ax2.set_ylabel('babies born', color='r')
plt.title("Beef Consumption and the Birth Rate")
plt.show()

Notice that the plot appears both in the console and the plot tab (bottom right tab).

Tip: You can "pop out" your plot by clicking the arrows at the top of the pane. This is handy if you're working on multiple monitors and want to dedicate one just to your data visualzations.

Usage

To keep this post concise and easy to read, we've just given the code snippets and a few lines of results for most of the queries below.

If you're following along in Rodeo, a few tips as you're getting started:

  • Run Script will indeed run everything you have written in the text editor
  • You can highlight a code chunk and run it by clicking Run Line or pressing Command + Enter
  • You can resize the panes (when I'm not making plots I shrink down the bottom right pane)

Basics

Write some SQL and execute it against your pandas DataFrame by substituting DataFrames for tables.

q = """
    SELECT
        *
    FROM
        meat
    LIMIT 10;"""

print sqldf(q, locals())

#                   date  beef  veal  pork  lamb_and_mutton broilers other_chicken turkey
# 0  1944-01-01 00:00:00   751    85  1280               89     None          None   None
# 1  1944-02-01 00:00:00   713    77  1169               72     None          None   None
# 2  1944-03-01 00:00:00   741    90  1128               75     None          None   None
# 3  1944-04-01 00:00:00   650    89   978               66     None          None   None

pandasql creates a DB, schema and all, loads your data, and runs your SQL.

Aggregation

pandasql supports aggregation. You can use aliased column names or column numbers in your group by clause.

# births per year
q = """
    SELECT
        strftime("%Y", date)
        , SUM(births)
    FROM births
    GROUP BY 1
    ORDER BY 1;
            """

print sqldf(q, locals())

#    strftime("%Y", date)  SUM(births)
# 0                  1975      3136965
# 1                  1976      6304156
# 2                  1979      3333279
# 3                  1982      3612258

locals() vs. globals()

pandasql needs to have access to other variables in your session/environment. You can pass locals() to pandasql when executing a SQL statement, but if you're running a lot of queries that might be a pain. To avoid passing locals all the time, you can add this helper function to your script to set globals() like so:

def pysqldf(q):
    return sqldf(q, globals())

q = """
    SELECT
        *
    FROM
        births
    LIMIT 10;"""

print pysqldf(q)
# 0  1975-01-01 00:00:00  265775
# 1  1975-02-01 00:00:00  241045
# 2  1975-03-01 00:00:00  268849

joins

You can join dataframes using normal SQL syntax.

# joining meats + births on date
q = """
    SELECT
        m.date
        , b.births
        , m.beef
    FROM
        meat m
    INNER JOIN
        births b
            on m.date = b.date
    ORDER BY
        m.date
    LIMIT 100;
    """

joined = pysqldf(q)
print joined.head()
#date  births    beef
#0  1975-01-01 00:00:00.000000  265775  2106.0
#1  1975-02-01 00:00:00.000000  241045  1845.0
#2  1975-03-01 00:00:00.000000  268849  1891.0

WHERE conditions

Here's a WHERE clause.

q = """
    SELECT
        date
        , beef
        , veal
        , pork
        , lamb_and_mutton
    FROM
        meat
    WHERE
        lamb_and_mutton >= veal
    ORDER BY date DESC
    LIMIT 10;
    """

print pysqldf(q)
#                   date    beef  veal    pork  lamb_and_mutton
# 0  2012-11-01 00:00:00  2206.6  10.1  2078.7             12.4
# 1  2012-10-01 00:00:00  2343.7  10.3  2210.4             14.2
# 2  2012-09-01 00:00:00  2016.0   8.8  1911.0             12.5
# 3  2012-08-01 00:00:00  2367.5  10.1  1997.9             14.2

It's just SQL

Since pandasql is powered by SQLite3, you can do most anything you can do in SQL. Here are some examples using common SQL features such as subqueries, order by, functions, and unions.

#################################################
# SQL FUNCTIONS
# e.g. `RANDOM()`
#################################################
q = """SELECT
    *
    FROM
        meat
    ORDER BY RANDOM()
    LIMIT 10;"""
print pysqldf(q)
#                   date  beef  veal  pork  lamb_and_mutton  broilers other_chicken  turkey
# 0  1967-03-01 00:00:00  1693    65  1136               61     472.0          None    26.5
# 1  1944-12-01 00:00:00   764   146  1013               91       NaN          None     NaN
# 2  1969-06-01 00:00:00  1666    50   964               42     573.9          None    85.4
# 3  1983-03-01 00:00:00  1892    37  1303               36    1106.2          None   182.7

#################################################
# UNION ALL
#################################################
q = """
        SELECT
            date
            , 'beef' AS meat_type
            , beef AS value
        FROM meat
        UNION ALL
        SELECT
            date
            , 'veal' AS meat_type
            , veal AS value
        FROM meat

        UNION ALL

        SELECT
            date
            , 'pork' AS meat_type
            , pork AS value
        FROM meat
        UNION ALL
        SELECT
            date
            , 'lamb_and_mutton' AS meat_type
            , lamb_and_mutton AS value
        FROM meat
        ORDER BY 1
    """
print pysqldf(q).head(20)
#                    date        meat_type  value
# 0   1944-01-01 00:00:00             beef    751
# 1   1944-01-01 00:00:00             veal     85
# 2   1944-01-01 00:00:00             pork   1280
# 3   1944-01-01 00:00:00  lamb_and_mutton     89


#################################################
# subqueries
# fancy!
#################################################
q = """
    SELECT
        m1.date
        , m1.beef
    FROM
        meat m1
    WHERE m1.date IN
        (SELECT
            date
        FROM meat
        WHERE
            beef >= broilers
        ORDER BY date)
"""

more_beef_than_broilers = pysqldf(q)
print more_beef_than_broilers.head(10)
#                   date  beef
# 0  1960-01-01 00:00:00  1196
# 1  1960-02-01 00:00:00  1089
# 2  1960-03-01 00:00:00  1201
# 3  1960-04-01 00:00:00  1066

Final thoughts

pandas is an incredible tool for data analysis in large part, we think, because it is extremely digestible, succinct, and expressive. Ultimately, there are tons of reasons to learn the nuances of merge, join, concatenate, melt and other native pandas features for slicing and dicing data. Check out the docs for some examples.

Our hope is that pandasql will be a helpful learning tool for folks new to Python and pandas. In my own personal experience learning R, sqldf was a familiar interface helping me become highly productive with a new tool as quickly as possible.

We hope you'll check out pandasql and Rodeo; if you do, please let us know what you think!



Our Products


Rodeo: a native Python editor built for doing data science on your desktop.

Download it now!

ScienceOps: deploy predictive models in production applications without IT.

Learn More

Yhat (pronounced Y-hat) provides data science solutions that let data scientists deploy and integrate predictive models into applications without IT or custom coding.