Basic introduction

into PySpark
B U I L D I N G D ATA E N G I N E E R I N G P I P E L I N E S I N P Y T H O N

Oliver Willekens
Data Engineer at Data Minded
What is Spark?
A fast and general engine for large-scale data processing

4 libraries built on top of Spark core:

API in several languages

Java, Scala, Python (“PySpark”), R

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

When to use Spark
Spark is used for:

Data processing at scale

Interactive analytics

Machine learning

Spark is not used for:

When you have only little data

When you have only simple operations

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Business case: nding the perfect diaper
Find the perfect diaper based on:

qualitative attributes e.g. comfort

quantitative attributes e.g. price

Scraped data available:

prices.csv: pricing details per model per store

ratings.csv: user ratings per model

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Starting the Spark analytics engine
from pyspark.sql import SparkSession

spark = SparkSession.builder.getOrCreate()

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Reading a CSV le
prices = spark.read.csv("mnt/data_lake/landing/prices.csv")

prices.show()

+---------+-----------+------------+-----+--------+--------+----------+
| _c0| _c1| _c2| _c4| _c5| _c6| _c7|
+---------+-----------+------------+-----+--------+--------+----------+
| store|countrycode| brand|price|currency|quantity| date|
| Aldi| BE|Diapers-R-Us| 6.8| EUR| 40|2019-02-03|
|Carrefour| FR| Nappy-k| 5.7| EUR| 30|2019-02-06|
| Tesco| IRL| Pampers| 6.3| EUR| 35|2019-02-07|
| DM| DE| Huggies| 6.8| EUR| 40|2019-02-01|
+---------+-----------+------------+-----+--------+--------+----------+

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Reading a CSV le with headers
prices = spark.read.options(header="true").csv("mnt/data_lake/landing/prices.csv")
prices.show()

+---------+-----------+------------+-----+--------+--------+----------+
| store|countrycode| brand|price|currency|quantity| date|
+---------+-----------+------------+-----+--------+--------+----------+
| Aldi| BE|Diapers-R-Us| 6.8| EUR| 40|2019-02-03|
|Carrefour| FR| Nappy-k| 5.7| EUR| 30|2019-02-06|
| Tesco| IRL| Pampers| 6.3| EUR| 35|2019-02-07|
| DM| DE| Huggies| 6.8| EUR| 40|2019-02-01|
+---------+-----------+------------+-----+--------+--------+----------+

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Automatically inferred data types
from pprint import pprint
pprint(prices.dtypes)

[('store', 'string'),
('countrycode', 'string'),
('brand', 'string'),
('price', 'string'),
('currency', 'string'),
('quantity', 'string'),
('date', 'string')]

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Enforcing a schema
schema = StructType([StructField("store", StringType(), nullable=False),
StructField("countrycode", StringType(), nullable=False),
StructField("brand", StringType(), nullable=False),
StructField("price", FloatType(), nullable=False),
StructField("currency", StringType(), nullable=True),
StructField("quantity", IntegerType(), nullable=True),
StructField("date", DateType(), nullable=False)])

prices = spark.read.options(header="true").schema(schema).csv("mnt/data_lake/landing/prices.csv")

print(prices.dtypes)

[('store', 'string'), ('countrycode', 'string'), ('brand', 'string'),

('price', 'float'), ('currency', 'string'), ('quantity', 'int'), ('date', 'date')]

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

 Let’s practice!
B U I L D I N G D ATA E N G I N E E R I N G P I P E L I N E S I N P Y T H O N
 Cleaning data
B U I L D I N G D ATA E N G I N E E R I N G P I P E L I N E S I N P Y T H O N

Oliver Willekens
Data Engineer at Data Minded
Reasons to clean data
Most data sources are not ready for analytics. This could be due to:

Incorrect data types

Invalid rows

Incomplete rows

Badly chosen placeholders

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Can we automate data cleaning?
Data cleaning depends on the context

Can our system cope with data that is 95% clean and 95% complete?

What are the implicit standards in the company?

regional datetimes vs. UTC

column naming conventions

What are the low-level details of the systems?

representation of unknown / incomplete data

ranges for numerical values

meaning of elds

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Selecting data types
Data type Value type in Python

ByteType Good for numbers that are within the range of -128 to 127.

ShortType Good for numbers that are within the range of -32768 to 32767.

IntegerType Good for numbers that are within the range of-2147483648 to 2147483647.

FloatType oat

StringType string

BooleanType bool

DateType datetime.date

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Badly formatted source data
cat bad_data.csv # prints the entire file on stdout

store,countrycode,brand,price,currency,quantity,date
Aldi,BE,Diapers-R-Us,6.8,EUR,40,2019-02-03
-----------------------------------
Kruidvat,NL,Nappy-k,5.6,EUR,40,2019-02-15
DM,AT,Huggies,7.2,EUR,40,2019-02-01

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Spark’s default handling of bad source data
prices = spark.read.options(header="true").csv('landing/prices.csv')
prices.show()

+--------------------+-----------+------------+-----+--------+--------+----------+
| store|countrycode| brand|price|currency|quantity| date|
+--------------------+-----------+------------+-----+--------+--------+----------+
| Aldi| BE|Diapers-R-Us| 6.8| EUR| 40|2019-02-03|
|-----------------...| null| null| null| null| null| null|
| Kruidvat| NL| Nappy-k| 5.6| EUR| 40|2019-02-15|
| DM| AT| Huggies| 7.2| EUR| 40|2019-02-01|
+--------------------+-----------+------------+-----+--------+--------+----------+

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Handle invalid rows
prices = (spark
.read
.options(header="true", mode="DROPMALFORMED")
.csv('landing/prices.csv'))

+--------+-----------+------------+-----+--------+--------+----------+
| store|countrycode| brand|price|currency|quantity| date|
+--------+-----------+------------+-----+--------+--------+----------+
| Aldi| BE|Diapers-R-Us| 6.8| EUR| 40|2019-02-03|
|Kruidvat| NL| Nappy-k| 5.6| EUR| 40|2019-02-15|
| DM| AT| Huggies| 7.2| EUR| 40|2019-02-01|
+--------+-----------+------------+-----+--------+--------+----------+

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

The signi cance of null
store,countrycode,brand,price,currency,quantity,date
Aldi,BE,Diapers-R-Us,6.8,EUR,40,2019-02-03
Kruidvat,,Nappy-k,5.6,EUR,,2019-02-15

prices = (spark.read.options(header="true")
.schema(schema)
.csv('/landing/prices_with_incomplete_rows.csv'))
prices.show()

+--------+-----------+------------+-----+--------+--------+----------+
| store|countrycode| brand|price|currency|quantity| date|
+--------+-----------+------------+-----+--------+--------+----------+
| Aldi| BE|Diapers-R-Us| 6.8| EUR| 40|2019-02-03|
|Kruidvat| null| Nappy-k| 5.6| EUR| null|2019-02-15|
+--------+-----------+------------+-----+--------+--------+----------+

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Supplying default values for missing data
prices.fillna(25, subset=['quantity']).show()

+--------+-----------+------------+-----+--------+--------+----------+
| store|countrycode| brand|price|currency|quantity| date|
+--------+-----------+------------+-----+--------+--------+----------+
| Aldi| BE|Diapers-R-Us| 6.8| EUR| 40|2019-02-03|
|Kruidvat| null| Nappy-k| 5.6| EUR| 25|2019-02-15|
+--------+-----------+------------+-----+--------+--------+----------+

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Badly chosen placeholders
Example: contracts of employees

employees = spark.read.options(header="true").schema(schema).csv('employees.csv')

+-------------+----------+----------+----------+
|employee_name|department|start_date| end_date|
+-------------+----------+----------+----------+
| Bob| marketing|2012-06-01|2016-05-02|
| Alice| IT|2018-04-03|9999-12-31|
+-------------+----------+----------+----------+

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Conditionally replace values
from pyspark.sql.functions import col, when
from datetime import date, timedelta
one_year_from_now = date.today().replace(year=date.today().year + 1)

better_frame = employees.withColumn("end_date",
when(col("end_date") > one_year_from_now, None).otherwise(col("end_date")))

better_frame.show()

+-------------+----------+----------+----------+
|employee_name|department|start_date| end_date|
+-------------+----------+----------+----------+
| Bob| marketing|2012-06-01|2016-05-02|
| Alice| IT|2018-04-03| null|
+-------------+----------+----------+----------+

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

 Let’s practice!
B U I L D I N G D ATA E N G I N E E R I N G P I P E L I N E S I N P Y T H O N
 Transforming data
with Spark
B U I L D I N G D ATA E N G I N E E R I N G P I P E L I N E S I N P Y T H O N

Oliver Willekens
Data Engineer at Data Minded
Why do we need to transform data?
Process:

1. Collect data

2. “Massage” data: involves cleaning and business logic

3. Derive insights

Example:

1. Collect data from booking.com and hotels.com.

2. Standardize hotel names, normalizing review scores.

3. Join datasets, lter on location and rank results.

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Common data transformations
1. Filtering rows

country | purchase_order
________|_______________
India | 87254800912
Ukraine | 32498562223

European purchases?

country | purchase_order
________|_______________
Ukraine | 32498562223

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Common data transformations
1. Filtering rows

2. Selecting and renaming columns

country | purchase_order | store_keep

________|________________|___________
Ukraine | 32498562223 | Oksana D.
Spain | 74398221190 | Pedro R.

->

country_of_purchase | purchase_order
____________________|________________
Ukraine | 32498562223
Spain | 74398221190

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Common data transformations
1. Filtering rows

2. Selecting and renaming columns

3. Grouping and aggregation

country | purchase_order | price country | total_revenue

________|________________|______ ________|______________
Ukraine | 32498562223 | $12 => Ukraine | $12
Spain | 74398221190 | $54 Spain | $80
Spain | 49876776100 | $26

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Common data transformations
1. Filtering rows

2. Selecting and renaming columns

3. Grouping and aggregation

4. Joining multiple datasets

country | purchase_order | price purchase_order | category

________|________________|______ _______________|_________
Ukraine | 32498562223 | $12 + 32498562223 | food
Spain | 74398221190 | $54 49876776100 | electronics
Spain | 49876776100 | $26 74398221190 | clothing

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Common data transformations
1. Filtering rows

2. Selecting and renaming columns

3. Grouping and aggregation

4. Joining multiple datasets

5. Ordering results

country | purchase_order | price country | purchase_order | price

________|________________|______ ________|________________|______
Spain | 74398221190 | $26 => Ukraine | 32498562223 | $12
Ukraine | 32498562223 | $12 Spain | 74398221190 | $26
Spain | 49876776100 | $54 Spain | 49876776100 | $54

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Recall the prices dataset
prices = spark.read.options(header="true").schema(schema).csv('landing/prices.csv')

+---------+-----------+------------+-----+--------+--------+----------+
| store|countrycode| brand|price|currency|quantity| date|
+---------+-----------+------------+-----+--------+--------+----------+
| Aldi| BE|Diapers-R-Us| 6.8| EUR| 40|2019-02-03|
| Kruidvat| BE| Nappy-k| 4.8| EUR| 30|2019-01-28|
|Carrefour| FR| Nappy-k| 5.7| EUR| 30|2019-02-06|
| Tesco| IRL| Pampers| 6.3| EUR| 35|2019-02-07|
| DM| DE| Huggies| 6.8| EUR| 40|2019-02-01|
+---------+-----------+------------+-----+--------+--------+----------+

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Filtering and ordering rows
prices_in_belgium = prices.filter(col('countrycode') == 'BE').orderBy(col('date'))

+--------+-----------+------------+-----+--------+--------+----------+
| store|countrycode| brand|price|currency|quantity| date|
+--------+-----------+------------+-----+--------+--------+----------+
|Kruidvat| BE| Nappy-k| 4.8| EUR| 30|2019-01-28|
| Aldi| BE|Diapers-R-Us| 6.8| EUR| 40|2019-02-03|
+--------+-----------+------------+-----+--------+--------+----------+

Function col creates Column objects

Method orderBy sorts values by a certain column.

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Selecting and renaming columns
prices.select(

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Selecting and renaming columns
prices.select(
col("store"),
col("brand")
)

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Selecting and renaming columns
prices.select( +---------+------------+
col("store"), | store| brandname|
col("brand").alias("brandname") +---------+------------+
) | Aldi|Diapers-R-Us|
| Kruidvat| Nappy-k|
|Carrefour| Nappy-k|
| Kruidvat| Nappy-k|
| Tesco| Pampers|
| DM| Huggies|
| DM| Huggies|
+---------+------------+

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Reducing duplicate values
prices.select( +---------+------------+
col("store"), | store| brandname|
col("brand").alias("brandname") +---------+------------+
).distinct() | DM| Huggies|
| Kruidvat| Nappy-k|
|Carrefour| Nappy-k|
| Aldi|Diapers-R-Us|
| Tesco| Pampers|
+---------+------------+

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Grouping and aggregating with mean()
(prices

.groupBy(col('brand'))

.mean('price')
).show()

+------------+------------------+
| brand| avg(price)|
+------------+------------------+
|Diapers-R-Us| 6.800000190734863|
| Pampers| 6.300000190734863|
| Huggies| 7.0|
| Nappy-k|5.3666666348775225|
+------------+------------------+

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Grouping and aggregating with agg()
(prices
.groupBy(col('brand'))

.agg(

avg('price').alias('average_price'),

count('brand').alias('number_of_items')

)
).show()

+------------+------------------+---------------+
| brand| average_price|number_of_items|
+------------+------------------+---------------+
|Diapers-R-Us| 6.800000190734863| 1|
| Pampers| 6.300000190734863| 1|
| Huggies| 7.0| 2|
| Nappy-k|5.3666666348775225| 3|

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Joining related data
+---------+-----------+------------+-------+-----+--------+--------+----------+
| store|countrycode| brand| model|price|currency|quantity| date|
+---------+-----------+------------+-------+-----+--------+--------+----------+
| Aldi| BE|Diapers-R-Us|6months| 6.8| EUR| 40|2019-02-03|
| Kruidvat| BE| Nappy-k|2months| 4.8| EUR| 30|2019-01-28|
|Carrefour| FR| Nappy-k|2months| 5.7| EUR| 30|2019-02-06|
| Tesco| IRL| Pampers|3months| 6.3| EUR| 35|2019-02-07|
| DM| DE| Huggies|newborn| 6.8| EUR| 40|2019-02-01|
+---------+-----------+------------+-------+-----+--------+--------+----------+
+------------+-------+---------------+-------+
| brand| model|absorption_rate|comfort|
+------------+-------+---------------+-------+
|Diapers-R-Us|6months| 2| 3|
| Nappy-k|2months| 3| 4|
| Pampers|3months| 4| 4|
| Huggies|newborn| 3| 5|
+------------+-------+---------------+-------+

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Executing a join with 2 foreign keys
ratings_with_prices = ratings.join(prices, ["brand", "model"])

+------------+-------+---------------+-------+---------+-----------+-----+--------+--------+---------
| brand| model|absorption_rate|comfort| store|countrycode|price|currency|quantity| dat
+------------+-------+---------------+-------+---------+-----------+-----+--------+--------+---------
|Diapers-R-Us|6months| 2| 3| Aldi| BE| 6.8| EUR| 40|2019-02-0
| Nappy-k|2months| 3| 4| Kruidvat| BE| 4.8| EUR| 30|2019-01-2
| Nappy-k|2months| 3| 4|Carrefour| FR| 5.7| EUR| 30|2019-02-0
| Pampers|3months| 4| 4| Tesco| IRL| 6.3| EUR| 35|2019-02-0
| Huggies|newborn| 3| 5| DM| DE| 6.8| EUR| 40|2019-02-0
+------------+-------+---------------+-------+---------+-----------+-----+--------+--------+---------

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

 Let’s practice!
B U I L D I N G D ATA E N G I N E E R I N G P I P E L I N E S I N P Y T H O N
 Packaging your
application
B U I L D I N G D ATA E N G I N E E R I N G P I P E L I N E S I N P Y T H O N

Oliver Willekens
Data Engineer at Data Minded
Running your pipeline locally
Running a Python program:

python hello_world.py # script does something

Running a PySpark program locally is no different:

python my_pyspark_data_pipeline.py # script starts at least a SparkSession

Conditions:

local installation of Spark

access to referenced resources

classpath is properly con gured

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Using the “spark-submit” helper program
spark-submit comes with any Spark installation

1. sets up launch environment for use with the

cluster manager and the selected deploy mode

2. invokes main class/app/module/function

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Basic arguments of “spark-submit”
spark-submit \ On your path, if Spark is installed
--master "local[*]" \ URL of the cluster manager
--py-files PY_FILES \ Comma-separated list of zip, egg or py
MAIN_PYTHON_FILE \ Path to the module to be run
app_arguments Optional arguments parsed by main script

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

Collecting all dependencies in one archive
zip \
--recurse-paths \
dependencies.zip \
pydiaper

spark-submit \
--py-files dependencies.zip \
pydiaper/cleaning/clean_prices.py

BUILDING DATA ENGINEERING PIPELINES IN PYTHON

 Let’s practice!
B U I L D I N G D ATA E N G I N E E R I N G P I P E L I N E S I N P Y T H O N