DataFrame

:ビッグデータの処理に使用します.データは表形式で表示され、SQL文でデータを管理できます.

では、さまざまな方法でDataFrameを作成できます.ここでは、RDDを使用したデータフレーム

の作成について説明する.

from pyspark.sql import Row

l = [('Ankit', 25),('Jalfaizy',22),('saurabh',20),('Bala',26)]
rdd = sc.parallelize(l)
people = rdd.map(lambda x : Row(name=x[0], age=int(x[1])))
schemaPeople = sqlContext.createDataFrame(people)

# 만들어진 데이터 프레임을 확인할 수 있음

print(type(schemaPeople))
schemaPeople.printSchema()
print(schemaPeople.collect())
print(schemaPeople.count())

<class 'pyspark.sql.dataframe.DataFrame'>
root
 |-- age: long (nullable = true)
 |-- name: string (nullable = true)

[Row(age=25, name='Ankit'), Row(age=22, name='Jalfaizy'), Row(age=20, name='saurabh'), Row(age=26, name='Bala')]
4

# 파일로부터 데이터프레임을 만드는 방법
# Parquet 파일 또는 Json 파일로 부터 데이터프레임을 주로 생성한다.

peopleDF = spark.read.parquet("./people-10m.parquet")

# 데이터프레임은 스키마를 가지고 있어서 아래 명령어로 확인할 수 있음
peopleDF.printSchema()

root
 |-- id: integer (nullable = true)
 |-- firstName: string (nullable = true)
 |-- middleName: string (nullable = true)
 |-- lastName: string (nullable = true)
 |-- gender: string (nullable = true)
 |-- birthDate: timestamp (nullable = true)
 |-- ssn: string (nullable = true)
 |-- salary: integer (nullable = true)

# 테이블형태로 확인할 수 있음
# head 명령어를 사용하게 되면 row 데이터 스트럭쳐로 보여줘 보기 힘듬
peopleDF.show(5)

peopleDF.head(5)

+---+---------+----------+----------+------+-------------------+-----------+------+
| id|firstName|middleName|  lastName|gender|          birthDate|        ssn|salary|
+---+---------+----------+----------+------+-------------------+-----------+------+
|  1|   Pennie|     Carry|Hirschmann|     F|1955-07-02 13:30:00|981-43-9345| 56172|
|  2|       An|     Amira|    Cowper|     F|1992-02-08 14:00:00|978-97-8086| 40203|
|  3|    Quyen|    Marlen|      Dome|     F|1970-10-11 13:00:00|957-57-8246| 53417|
|  4|  Coralie|  Antonina|   Marshal|     F|1990-04-11 13:00:00|963-39-4885| 94727|
|  5|   Terrie|      Wava|     Bonar|     F|1980-01-16 14:00:00|964-49-8051| 79908|
+---+---------+----------+----------+------+-------------------+-----------+------+
only showing top 5 rows






[Row(id=1, firstName='Pennie', middleName='Carry', lastName='Hirschmann', gender='F', birthDate=datetime.datetime(1955, 7, 2, 13, 30), ssn='981-43-9345', salary=56172),
 Row(id=2, firstName='An', middleName='Amira', lastName='Cowper', gender='F', birthDate=datetime.datetime(1992, 2, 8, 14, 0), ssn='978-97-8086', salary=40203),
 Row(id=3, firstName='Quyen', middleName='Marlen', lastName='Dome', gender='F', birthDate=datetime.datetime(1970, 10, 11, 13, 0), ssn='957-57-8246', salary=53417),
 Row(id=4, firstName='Coralie', middleName='Antonina', lastName='Marshal', gender='F', birthDate=datetime.datetime(1990, 4, 11, 13, 0), ssn='963-39-4885', salary=94727),
 Row(id=5, firstName='Terrie', middleName='Wava', lastName='Bonar', gender='F', birthDate=datetime.datetime(1980, 1, 16, 14, 0), ssn='964-49-8051', salary=79908)]

print(peopleDF.count()) # 인스턴스의 개수를 알려준다.
print(peopleDF.columns) # 컬럼만을 보여준다.
print(peopleDF.describe().show(5)) # show(5)와 같은 결과

10000000
['id', 'firstName', 'middleName', 'lastName', 'gender', 'birthDate', 'ssn', 'salary']

#select 문을 통해 데이터를 확인할 수 있음
peopleDF.select('firstName', 'birthDate').show(5)

+---------+-------------------+
|firstName|          birthDate|
+---------+-------------------+
|   Pennie|1955-07-02 13:30:00|
|       An|1992-02-08 14:00:00|
|    Quyen|1970-10-11 13:00:00|
|  Coralie|1990-04-11 13:00:00|
|   Terrie|1980-01-16 14:00:00|
+---------+-------------------+
only showing top 5 rows

peopleDF.select('gender').distinct().count()

2

# filter 는 여러가지 형태로 사용가능
peopleDF.filter(peopleDF['gender']=="F").count()
peopleDF.filter(peopleDF.gender == 'F').count()
peopleDF.filter("gender == 'F'").count()

# filter 안에 펑션을 넣을 수도 있음
from pyspark.sql.functions import year
peopleDF.filter(year("birthDate") > "1995").count()

849353

# and 를 사용하여 조건을 추가할 수도 있음
test1 = peopleDF.select("firstName").filter("gender = 'M'" and year("birthDate")>"1960")
test1.show(5)

+---------+
|firstName|
+---------+
|       An|
|    Quyen|
|  Coralie|
|   Terrie|
| Chassidy|
+---------+
only showing top 5 rows

# select 문을 사용한 결과에 filter 펑션을 적용하는 것도 가능.
peopleDF.select("firstName", year("birthDate")).filter("gender='M'" and year("birthDate")>"1960").show(10)

+---------+---------------+
|firstName|year(birthDate)|
+---------+---------------+
|       An|           1992|
|    Quyen|           1970|
|  Coralie|           1990|
|   Terrie|           1980|
| Chassidy|           1990|
|     Geri|           1970|
|   Patria|           1985|
|   Terese|           1967|
|     Wava|           1963|
|   Sophie|           1979|
+---------+---------------+
only showing top 10 rows

#GroupBy를 이용해서 특정 컬럼을 그룹핑 할 수 있음
peopleDF.groupBy("gender").count().show()

+------+-------+
|gender|  count|
+------+-------+
|     F|5187302|
|     M|4812698|
+------+-------+

# firstName을 알파벳순으로 정렬한 후 groupBy
peopleDF.orderBy('firstName').groupBy("firstName").count().show(5)

+---------+-----+
|firstName|count|
+---------+-----+
|    Aaron| 4020|
|    Abbey| 1273|
|    Abbie| 1288|
|     Abby| 1235|
|    Abdul| 3966|
+---------+-----+
only showing top 5 rows

# firstName으로 그룹핑 한 후에, 가장 많이 나온 순으로 정렬해서 볼 수 있다.
peopleDF.groupBy("FirstName").count().orderBy("count", ascending=False).show(5)

+---------+-----+
|FirstName|count|
+---------+-----+
|   Sammie| 5462|
|    Daryl| 5442|
|   Bobbie| 5432|
|      Lee| 5411|
|     Troy| 5408|
+---------+-----+
only showing top 5 rows

# count() 을 사용하면 count 이름의 column이 생기게 된다.
# 특정 수치 이상의 count를 가진 값을 출력하는 명령어
# column에 접근하기 위해서 col을 import.

from pyspark.sql.functions import col
peopleDF.groupBy("firstName").count().filter(col("count") > "5430")\
        .orderBy("count", ascending=False).show(6)

+---------+-----+
|firstName|count|
+---------+-----+
|   Sammie| 5462|
|    Daryl| 5442|
|   Bobbie| 5432|
+---------+-----+

# Multiline command는 \을 이용해서 가능
peopleDF\
    .select("firstname", "lastName")\
    .filter("gender = 'M'")\
    .show(5)

+---------+--------+
|firstname|lastName|
+---------+--------+
|   Jasper|Neubigin|
|   Robert| Kipling|
|    Abdul|  Pachta|
|    Earle|Gieraths|
|   Mickey|  Wotton|
+---------+--------+
only showing top 5 rows

testDF1 = peopleDF\
    .select("firstname", "lastName", "birthDate")\
    .filter("gender = 'M'")
testDF1.show(5)

+---------+--------+-------------------+
|firstname|lastName|          birthDate|
+---------+--------+-------------------+
|   Jasper|Neubigin|1961-02-20 13:30:00|
|   Robert| Kipling|1986-01-02 14:00:00|
|    Abdul|  Pachta|1956-01-28 13:30:00|
|    Earle|Gieraths|1958-04-17 13:30:00|
|   Mickey|  Wotton|1976-08-29 13:00:00|
+---------+--------+-------------------+
only showing top 5 rows

# display를 사용하면 이차워 그래프와 같은 visualization 이 가능
# (jupyter notebook에서는 안되는 것 같음)
from pyspark.sql.functions import year
testDF2 = testDF1.\
            filter(year("birthDate")>"1990")
testDF2.show(5)

+---------+---------+-------------------+
|firstname| lastName|          birthDate|
+---------+---------+-------------------+
|   Theron|    Aplin|1995-03-16 14:00:00|
| Valentin|  Yakunin|1999-04-06 13:00:00|
|   Emmett|    Grgic|1991-04-22 13:00:00|
|   Nelson|     Kelf|1995-08-17 13:00:00|
|    Sonny|Kornousek|1994-11-04 14:00:00|
+---------+---------+-------------------+
only showing top 5 rows

display(testDF2)

DataFrame[firstname: string, lastName: string, birthDate: timestamp]

from pyspark.sql.functions import year
display(
    peopleDF
        .select("firstName", "middleName", "lastName", "birthDate", "gender")
        .filter("gender = 'F'")
        .filter(year("birthDate") > "1990")
)

DataFrame[firstName: string, middleName: string, lastName: string, birthDate: timestamp, gender: string]

marysDF = peopleDF\
            .select(year("birthDate").alias('birthYear'))\
            .filter("firstName = 'Mary'")\
            .filter("gender = 'F'")\
            .orderBy('birthYear')\
            .groupBy('birthYear')\
            .count()

display(marysDF)

DataFrame[birthYear: int, count: bigint]

from pyspark.sql.functions import col
dordonDF = peopleDF\
            .select(year("birthDate").alias("birthYear"), "firstName")\
            .filter((col("firstName") == 'Donna') | (col("firstName") == 'Dorothy'))\
            .orderBy('birthYear')\
            .groupBy('birthYear', 'firstName')\
            .count()\
            .show(5)

+---------+---------+-----+
|birthYear|firstName|count|
+---------+---------+-----+
|     1952|    Donna|   24|
|     1952|  Dorothy|   21|
|     1953|    Donna|   37|
|     1953|  Dorothy|   38|
|     1954|  Dorothy|   29|
+---------+---------+-----+
only showing top 5 rows

Temporary Views

データフレームを作成した後、sqlのフォーマットを使用できます.

spark sessionが終わってからなくなりました.