Spark 分组取Top N运算

大数据处理中，对数据分组后，取TopN是非常常见的运算。

下面我们以一个例子来展示spark如何进行分组取Top的运算。

1、RDD方法分组取TopN

from pyspark import SparkContext
sc = SparkContext()

准备数据，把数据转换为rdd格式

data_list = [
 (0, "cat26", 130.9), (0, "cat13", 122.1), (0, "cat95", 119.6), (0, "cat105", 11.3),
 (1, "cat67", 128.5), (1, "cat4", 126.8), (1, "cat13", 112.6), (1, "cat23", 15.3),
 (2, "cat56", 139.6), (2, "cat40", 129.7), (2, "cat187", 127.9), (2, "cat68", 19.8),
 (3, "cat8", 135.6)
]

data = sc.parallelize(data_list)
data.collect()

[(0, "cat26", 130.9),
 (0, "cat13", 122.1),
 (0, "cat95", 119.6),
 (0, "cat105", 11.3),
 (1, "cat67", 128.5),
 (1, "cat4", 126.8),
 (1, "cat13", 112.6),
 (1, "cat23", 15.3),
 (2, "cat56", 139.6),
 (2, "cat40", 129.7),
 (2, "cat187", 127.9),
 (2, "cat68", 19.8),
 (3, "cat8", 135.6)]

对数据使用groupBy操作来分组。可以看到分组后数据为(key, list_data)

d1 = data.groupBy(lambda x:x[0])
temp = d1.collect()
print(list(temp[0][1]))
print(temp)

[(0, "cat26", 130.9), (0, "cat13", 122.1), (0, "cat95", 119.6), (0, "cat105", 11.3)]
[(0, ), (1, ), (2, ), (3, )]

使用mapValues方法对数据进行排序。

可以根据需要来取Top N 数据。

这里取Top 3 的数据

d2 = d1.mapValues(lambda x: sorted(x, key=lambda y:y[2])[:3])
d2.collect()

[(0, [(0, "cat105", 11.3), (0, "cat95", 119.6), (0, "cat13", 122.1)]),
 (1, [(1, "cat23", 15.3), (1, "cat13", 112.6), (1, "cat4", 126.8)]),
 (2, [(2, "cat68", 19.8), (2, "cat187", 127.9), (2, "cat40", 129.7)]),
 (3, [(3, "cat8", 135.6)])]

使用flatmap方法把结果拉平，变成一个list返回。

d3 = d2.flatMap(lambda x:[i for i in x[1]])
d3.collect()

[(0, "cat105", 11.3),
 (0, "cat95", 119.6),
 (0, "cat13", 122.1),
 (1, "cat23", 15.3),
 (1, "cat13", 112.6),
 (1, "cat4", 126.8),
 (2, "cat68", 19.8),
 (2, "cat187", 127.9),
 (2, "cat40", 129.7),
 (3, "cat8", 135.6)]

整体代码

from pyspark import SparkContext
# sc = SparkContext()

topN = 3
data_list = [
 (0, "cat26", 130.9), (0, "cat13", 122.1), (0, "cat95", 119.6), (0, "cat105", 11.3),
 (1, "cat67", 128.5), (1, "cat4", 126.8), (1, "cat13", 112.6), (1, "cat23", 15.3),
 (2, "cat56", 139.6), (2, "cat40", 129.7), (2, "cat187", 127.9), (2, "cat68", 19.8),
 (3, "cat8", 135.6)
]

data = sc.parallelize(data_list)
d1 = data.groupBy(lambda x:x[0])
d2 = d1.mapValues(lambda x: sorted(x, key=lambda y:y[2])[:topN])
d3 = d2.flatMap(lambda x:[i for i in x[1]])
d3.collect()

[(0, "cat105", 11.3),
 (0, "cat95", 119.6),
 (0, "cat13", 122.1),
 (1, "cat23", 15.3),
 (1, "cat13", 112.6),
 (1, "cat4", 126.8),
 (2, "cat68", 19.8),
 (2, "cat187", 127.9),
 (2, "cat40", 129.7),
 (3, "cat8", 135.6)]

2、Dataframe方法分组取TopN

dataframe数据格式分组取top N，简单的方法是使用Window方法

from pyspark.sql import SparkSession
from pyspark.sql import functions as func
from pyspark.sql import Window

spark = SparkSession.builder.getOrCreate()

data_list = [
 (0, "cat26", 130.9), (0, "cat13", 122.1), (0, "cat95", 119.6), (0, "cat105", 11.3),
 (1, "cat67", 128.5), (1, "cat4", 126.8), (1, "cat13", 112.6), (1, "cat23", 15.3),
 (2, "cat56", 139.6), (2, "cat40", 129.7), (2, "cat187", 127.9), (2, "cat68", 19.8),
 (3, "cat8", 135.6)
]

根据数据创建dataframe，并给数据列命名

df = spark.createDataFrame(data_list, ["Hour", "Category", "TotalValue"])
df.show()

+----+--------+----------+
|Hour|Category|TotalValue|
+----+--------+----------+
| 0| cat26| 130.9|
| 0| cat13| 122.1|
| 0| cat95| 119.6|
| 0| cat105| 11.3|
| 1| cat67| 128.5|
| 1| cat4| 126.8|
| 1| cat13| 112.6|
| 1| cat23| 15.3|
| 2| cat56| 139.6|
| 2| cat40| 129.7|
| 2| cat187| 127.9|
| 2| cat68| 19.8|
| 3| cat8| 135.6|
+----+--------+----------+

1. 使用窗口方法，分片参数为分组的key，

2. orderBy的参数为排序的key，这里使用desc降序排列。

3. withColumn(colName, col)，为df添加一列，数据为对window函数生成的数据编号

4. where方法取rn列值小于3的数据，即取top3数据

w = Window.partitionBy(df.Hour).orderBy(df.TotalValue.desc())
top3 = df.withColumn("rn", func.row_number().over(w)).where("rn <=3")
top3.show()

+----+--------+----------+---+
|Hour|Category|TotalValue| rn|
+----+--------+----------+---+
| 0| cat26| 130.9| 1|
| 0| cat13| 122.1| 2|
| 0| cat95| 119.6| 3|
| 1| cat67| 128.5| 1|
| 1| cat4| 126.8| 2|
| 1| cat13| 112.6| 3|
| 3| cat8| 135.6| 1|
| 2| cat56| 139.6| 1|
| 2| cat40| 129.7| 2|
| 2| cat187| 127.9| 3|
+----+--------+----------+---+

### 代码汇总

from pyspark.sql import SparkSession
from pyspark.sql import functions as func
from pyspark.sql import Window

spark = SparkSession.builder.getOrCreate()

data_list = [
 (0, "cat26", 130.9), (0, "cat13", 122.1), (0, "cat95", 119.6), (0, "cat105", 11.3),
 (1, "cat67", 128.5), (1, "cat4", 126.8), (1, "cat13", 112.6), (1, "cat23", 15.3),
 (2, "cat56", 139.6), (2, "cat40", 129.7), (2, "cat187", 127.9), (2, "cat68", 19.8),
 (3, "cat8", 135.6)
]
df = spark.createDataFrame(data_list, ["Hour", "Category", "TotalValue"])

w = Window.partitionBy(df.Hour).orderBy(df.TotalValue.desc())
top3 = df.withColumn("rn", func.row_number().over(w)).where("rn <=3")

top3.show()