pyspark是spark為python開發(fā)者專門提供的api,他可以使用python來調(diào)用spark的計算引擎用于進行數(shù)據(jù)分析。學習pyspark的第一步就是pyspark環(huán)境配置和基本操作,接下來小編就來介紹一下這兩點內(nèi)容。
下載依賴
首先需要下載hadoop和spark,解壓,然后設(shè)置環(huán)境變量。
hadoop清華源下載
spark清華源下載
HADOOP_HOME => /path/hadoop SPARK_HOME => /path/spark
安裝pyspark。
pip install pyspark
基本使用
可以在shell終端,輸入pyspark,有如下回顯:
輸入以下指令進行測試,并創(chuàng)建SparkContext,SparkContext是任何spark功能的入口點。
>>> from pyspark import SparkContext
>>> sc = SparkContext("local", "First App")
如果以上不會報錯,恭喜可以開始使用pyspark編寫代碼了。
不過,我這里使用IDE來編寫代碼,首先我們先在終端執(zhí)行以下代碼關(guān)閉SparkContext。
>>> sc.stop()
下面使用pycharm編寫代碼,如果修改了環(huán)境變量需要先重啟pycharm。
在pycharm運行如下程序,程序會起本地模式的spark計算引擎,通過spark統(tǒng)計abc.txt文件中a和b出現(xiàn)行的數(shù)量,文件路徑需要自己指定。
from pyspark import SparkContext
sc = SparkContext("local", "First App")
logFile = "abc.txt"
logData = sc.textFile(logFile).cache()
numAs = logData.filter(lambda s: 'a' in s).count()
numBs = logData.filter(lambda s: 'b' in s).count()
print("Line with a:%i,line with b:%i" % (numAs, numBs))
運行結(jié)果如下:
20/03/11 16:15:57 WARN NativeCodeLoader: Unable to load native-hadoop library for your platform... using builtin-java classes where applicable
Using Spark's default log4j profile: org/apache/spark/log4j-defaults.properties
Setting default log level to "WARN".
To adjust logging level use sc.setLogLevel(newLevel). For SparkR, use setLogLevel(newLevel).
20/03/11 16:15:58 WARN Utils: Service 'SparkUI' could not bind on port 4040. Attempting port 4041.
Line with a:3,line with b:1
這里說一下,同樣的工作使用python可以做,spark也可以做,使用spark主要是為了高效的進行分布式計算。
戳pyspark教程
戳spark教程
RDD
RDD代表Resilient Distributed Dataset,它們是在多個節(jié)點上運行和操作以在集群上進行并行處理的元素,RDD是spark計算的操作對象。
一般,我們先使用數(shù)據(jù)創(chuàng)建RDD,然后對RDD進行操作。
對RDD操作有兩種方法:
Transformation(轉(zhuǎn)換) - 這些操作應用于RDD以創(chuàng)建新的RDD。例如filter,groupBy和map。
Action(操作) - 這些是應用于RDD的操作,它指示Spark執(zhí)行計算并將結(jié)果發(fā)送回驅(qū)動程序,例如count,collect等。
創(chuàng)建RDD
parallelize是從列表創(chuàng)建RDD,先看一個例子:
from pyspark import SparkContext
sc = SparkContext("local", "count app")
words = sc.parallelize(
["scala",
"java",
"hadoop",
"spark",
"akka",
"spark vs hadoop",
"pyspark",
"pyspark and spark"
])
print(words)
結(jié)果中我們得到一個對象,就是我們列表數(shù)據(jù)的RDD對象,spark之后可以對他進行操作。
ParallelCollectionRDD[0] at parallelize at PythonRDD.scala:195
Count
count方法返回RDD中的元素個數(shù)。
from pyspark import SparkContext
sc = SparkContext("local", "count app")
words = sc.parallelize(
["scala",
"java",
"hadoop",
"spark",
"akka",
"spark vs hadoop",
"pyspark",
"pyspark and spark"
])
print(words)
counts = words.count()
print("Number of elements in RDD -> %i" % counts)
返回結(jié)果:
Number of elements in RDD -> 8
Collect
collect返回RDD中的所有元素。
from pyspark import SparkContext
sc = SparkContext("local", "collect app")
words = sc.parallelize(
["scala",
"java",
"hadoop",
"spark",
"akka",
"spark vs hadoop",
"pyspark",
"pyspark and spark"
])
coll = words.collect()
print("Elements in RDD -> %s" % coll)
返回結(jié)果:
Elements in RDD -> ['scala', 'java', 'hadoop', 'spark', 'akka', 'spark vs hadoop', 'pyspark', 'pyspark and spark']
foreach
每個元素會使用foreach內(nèi)的函數(shù)進行處理,但是不會返回任何對象。
下面的程序中,我們定義的一個累加器accumulator,用于儲存在foreach執(zhí)行過程中的值。
from pyspark import SparkContext
sc = SparkContext("local", "ForEach app")
accum = sc.accumulator(0)
data = [1, 2, 3, 4, 5]
rdd = sc.parallelize(data)
def increment_counter(x):
print(x)
accum.add(x)
return 0
s = rdd.foreach(increment_counter)
print(s) # None
print("Counter value: ", accum)
返回結(jié)果:
None
Counter value: 15
filter
返回一個包含元素的新RDD,滿足過濾器的條件。
from pyspark import SparkContext
sc = SparkContext("local", "Filter app")
words = sc.parallelize(
["scala",
"java",
"hadoop",
"spark",
"akka",
"spark vs hadoop",
"pyspark",
"pyspark and spark"]
)
words_filter = words.filter(lambda x: 'spark' in x)
filtered = words_filter.collect()
print("Fitered RDD -> %s" % (filtered))
Fitered RDD -> ['spark', 'spark vs hadoop', 'pyspark', 'pyspark and spark']
也可以改寫成這樣:
from pyspark import SparkContext
sc = SparkContext("local", "Filter app")
words = sc.parallelize(
["scala",
"java",
"hadoop",
"spark",
"akka",
"spark vs hadoop",
"pyspark",
"pyspark and spark"]
)
def g(x):
for i in x:
if "spark" in x:
return i
words_filter = words.filter(g)
filtered = words_filter.collect()
print("Fitered RDD -> %s" % (filtered))
map
將函數(shù)應用于RDD中的每個元素并返回新的RDD。
from pyspark import SparkContext
sc = SparkContext("local", "Map app")
words = sc.parallelize(
["scala",
"java",
"hadoop",
"spark",
"akka",
"spark vs hadoop",
"pyspark",
"pyspark and spark"]
)
words_map = words.map(lambda x: (x, 1, "_{}".format(x)))
mapping = words_map.collect()
print("Key value pair -> %s" % (mapping))
返回結(jié)果:
Key value pair -> [('scala', 1, '_scala'), ('java', 1, '_java'), ('hadoop', 1, '_hadoop'), ('spark', 1, '_spark'), ('akka', 1, '_akka'), ('spark vs hadoop', 1, '_spark vs hadoop'), ('pyspark', 1, '_pyspark'), ('pyspark and spark', 1, '_pyspark and spark')]
Reduce
執(zhí)行指定的可交換和關(guān)聯(lián)二元操作后,然后返回RDD中的元素。
from pyspark import SparkContext
from operator import add
sc = SparkContext("local", "Reduce app")
nums = sc.parallelize([1, 2, 3, 4, 5])
adding = nums.reduce(add)
print("Adding all the elements -> %i" % (adding))
這里的add是python內(nèi)置的函數(shù),可以使用ide查看:
def add(a, b):
"Same as a + b."
return a + b
reduce會依次對元素相加,相加后的結(jié)果加上其他元素,最后返回結(jié)果(RDD中的元素)。
Adding all the elements -> 15
Join
返回RDD,包含兩者同時匹配的鍵,鍵包含對應的所有元素。
from pyspark import SparkContext
sc = SparkContext("local", "Join app")
x = sc.parallelize([("spark", 1), ("hadoop", 4), ("python", 4)])
y = sc.parallelize([("spark", 2), ("hadoop", 5)])
print("x =>", x.collect())
print("y =>", y.collect())
joined = x.join(y)
final = joined.collect()
print( "Join RDD -> %s" % (final))
返回結(jié)果:
x => [('spark', 1), ('hadoop', 4), ('python', 4)]
y => [('spark', 2), ('hadoop', 5)]
Join RDD -> [('hadoop', (4, 5)), ('spark', (1, 2))]
到這里pyspark環(huán)境配置和pyspark基本操作就基本介紹完畢了,希望對各位小伙伴有所幫助,更多python學習內(nèi)容也可以關(guān)注W3Cschool的其他文章!