Question

python Pandas库包含以下功能：

In [48]: df1 = pd.DataFrame({'col1': [0, 1], 'col_left':['a', 'b']})

In [49]: df2 = pd.DataFrame({'col1': [1, 2, 2],'col_right':[2, 2, 2]})

In [50]: pd.merge(df1, df2, on='col1', how='outer', indicator=True)
Out[50]: 
   col1 col_left  col_right      _merge
0     0        a        NaN   left_only
1     1        b        2.0        both
2     2      NaN        2.0  right_only
3     2      NaN        2.0  right_only

指标字段与Panda的value_counts（）函数结合使用，可以快速确定连接的执行情况。

示例：

ASchema = StructType([StructField('id', IntegerType(),nullable=False),
                 StructField('name', StringType(),nullable=False)])
BSchema = StructType([StructField('id', IntegerType(),nullable=False),
                 StructField('role', StringType(),nullable=False)])
AData = sc.parallelize ([ Row(1,'michel'), Row(2,'diederik'), Row(3,'rok'), Row(4,'piet')])
BData = sc.parallelize ([ Row(1,'engineer'), Row(2,'lead'), Row(3,'scientist'), Row(5,'manager')])
ADF = hc.createDataFrame(AData,ASchema)
BDF = hc.createDataFrame(BData,BSchema)
DFJOIN = ADF.join(BDF, ADF['id'] == BDF['id'], "outer")
DFJOIN.show()

Input:
+----+--------+----+---------+
|  id|    name|  id|     role|
+----+--------+----+---------+
|   1|  michel|   1| engineer|
|   2|diederik|   2|     lead|
|   3|     rok|   3|scientist|
|   4|    piet|null|     null|
|null|    null|   5|  manager|
+----+--------+----+---------+

from pyspark.sql.functions import *
DFJOINMERGE = DFJOIN.withColumn("_merge", when(ADF["id"].isNull(), "right_only").when(BDF["id"].isNull(), "left_only").otherwise("both"))\
  .withColumn("id", coalesce(ADF["id"], BDF["id"]))\
   .drop(ADF["id"])\
   .drop(BDF["id"])
DFJOINMERGE.show()

Output
+---+--------+---+---------+------+
| id|    name| id|     role|_merge|
+---+--------+---+---------+------+
|  1|  michel|  1| engineer|  both|
|  2|diederik|  2|     lead|  both|
|  3|     rok|  3|scientist|  both|
|  4|    piet|  4|     null|  both|
|  5|    null|  5|  manager|  both|
+---+--------+---+---------+------+

 ==> I would expect id 4 to be left, and id 5 to be right.

Changing join to "left":


Input:
+---+--------+----+---------+
| id|    name|  id|     role|
+---+--------+----+---------+
|  1|  michel|   1| engineer|
|  2|diederik|   2|     lead|
|  3|     rok|   3|scientist|
|  4|    piet|null|     null|
+---+--------+----+---------+

Output
+---+--------+---+---------+------+
| id|    name| id|     role|_merge|
+---+--------+---+---------+------+
|  1|  michel|  1| engineer|  both|
|  2|diederik|  2|     lead|  both|
|  3|     rok|  3|scientist|  both|
|  4|    piet|  4|     null|  both|
+---+--------+---+---------+------+

在Spark Dataframe中检查连接性能的最佳方法是什么？

在其中一个答案中提供了一个自定义函数：它还没有给出正确的结果，但如果它会这样会很好：

C++11

Answer 1

改变了LostInOverflow的答案，让它发挥作用：

from pyspark.sql import Row

ASchema = StructType([StructField('ida', IntegerType(),nullable=False),
                 StructField('name', StringType(),nullable=False)])
BSchema = StructType([StructField('idb', IntegerType(),nullable=False),
                 StructField('role', StringType(),nullable=False)])
AData = sc.parallelize ([ Row(1,'michel'), Row(2,'diederik'), Row(3,'rok'), Row(4,'piet')])
BData = sc.parallelize ([ Row(1,'engineer'), Row(2,'lead'), Row(3,'scientist'), Row(5,'manager')])
ADF = hc.createDataFrame(AData,ASchema)
BDF = hc.createDataFrame(BData,BSchema)
DFJOIN = ADF.join(BDF, ADF['ida'] == BDF['idb'], "outer")
DFJOIN.show()


+----+--------+----+---------+
| ida|    name| idb|     role|
+----+--------+----+---------+
|   1|  michel|   1| engineer|
|   2|diederik|   2|     lead|
|   3|     rok|   3|scientist|
|   4|    piet|null|     null|
|null|    null|   5|  manager|
+----+--------+----+---------+

from pyspark.sql.functions import *
DFJOINMERGE = DFJOIN.withColumn("_merge", when(DFJOIN["ida"].isNull(), "right_only").when(DFJOIN["idb"].isNull(), "left_only").otherwise("both"))\
  .withColumn("id", coalesce(ADF["ida"], BDF["idb"]))\
   .drop(DFJOIN["ida"])\
   .drop(DFJOIN["idb"])
#DFJOINMERGE.show()
DFJOINMERGE.groupBy("_merge").count().show()

+----------+-----+
|    _merge|count|
+----------+-----+
|right_only|    1|
| left_only|    1|
|      both|    3|
+----------+-----+

Answer 2

试试这个：

>>> from pyspark.sql.functions import *
>>> sdf1 = sqlContext.createDataFrame(df1)
>>> sdf2 = sqlContext.createDataFrame(df2)
>>> sdf = sdf1.join(sdf2, sdf1["col1"] == sdf2["col1"], "outer")
>>> sdf.withColumn("_merge", when(sdf1["col1"].isNull(), "right_only").when(sdf2["col1"].isNull(), "left_only").otherwise("both"))\
...  .withColumn("col1", coalesce(sdf1["col1"], sdf2["col1"]))\
...   .drop(sdf1["col1"])\
...   .drop(sdf2["col1"])

Spark Dataframe是否具有Panda合并指标的等效选项？

2 个答案: