Full blocking join in Pyspark - pyspark

I am trying to use pyspark to dedupe a database and as part of the pipeline I want to create a dataframe from two identical left and right dataframes. These have a shared index.
Initially I was lazy and just used a CROSS JOIN, but that results in duplication of joins (as left to right is the same as right to left). But I now need to optimise the code a bit and I am left wondering what is the best way to achieve this type of join. Can anyone suggest anything?
l_df = spark.createDataFrame([('A',), ('B',), ('C',), ('D',)], ['l_idx'])
r_df = spark.createDataFrame([('A',), ('B',), ('C',), ('D',)], ['r_idx'])
l_df.createOrReplaceTempView('l_df')
r_df.createOrReplaceTempView('r_df')
block_df = spark.sql('''
SELECT *
FROM l_df
CROSS JOIN r_df
''')
block_df.show()
This results in all combinations.
However I am looking for..
A B
A C
A D
B C
B D
C D
only
Thanks jxc!
Solved with your suggestion!

It will work. Please find below.
>>> bdf = spark.sql("select * from l_df CROSS JOIN r_df WHERE l_df.l_idx < r_df.r_idx")
>>> bdf.show()
+-----+-----+
|l_idx|r_idx|
+-----+-----+
| A| B|
| A| C|
| A| D|
| B| C|
| B| D|
| C| D|
+-----+-----+

Related

transform distinct row values to different columns with corresponding rows using Pyspark

I'm new to Pyspark and trying to transform data
Given dataframe
Col1
A=id1a A=id2a B=id1b C=id1c B=id2b
D=id1d A=id3a B=id3b C=id2c
A=id4a C=id3c
Required:
A B C
id1a id1b id1c
id2a id2b id2c
id3a id3b id3b
id4a null null
I have tried pivot, but that gives first value.
There might be a better way , however an approach is splitting the column on spaces to create array of the entries and then using higher order functions(spark 2.4+) to split on the '=' for each entry in the splitted array .Then explode and create 2 columns one with the id and one with the value. Then we can assign a row number to each partition and groupby then pivot:
import pyspark.sql.functions as F
df1 = (df.withColumn("Col1",F.split(F.col("Col1"),"\s+")).withColumn("Col1",
F.explode(F.expr("transform(Col1,x->split(x,'='))")))
.select(F.col("Col1")[0].alias("cols"),F.col("Col1")[1].alias("vals")))
from pyspark.sql import Window
w = Window.partitionBy("cols").orderBy("cols")
final = (df1.withColumn("Rnum",F.row_number().over(w)).groupBy("Rnum")
.pivot("cols").agg(F.first("vals")).orderBy("Rnum"))
final.show()
+----+----+----+----+----+
|Rnum| A| B| C| D|
+----+----+----+----+----+
| 1|id1a|id1b|id1c|id1d|
| 2|id2a|id2b|id2c|null|
| 3|id3a|id3b|id3c|null|
| 4|id4a|null|null|null|
+----+----+----+----+----+
this is how df1 looks like after the transformation:
df1.show()
+----+----+
|cols|vals|
+----+----+
| A|id1a|
| A|id2a|
| B|id1b|
| C|id1c|
| B|id2b|
| D|id1d|
| A|id3a|
| B|id3b|
| C|id2c|
| A|id4a|
| C|id3c|
+----+----+
May be I don't know the full picture, but the data format seems to be strange. If nothing can be done at the data source, then some collects, pivots and joins will be needed. Try this.
import pyspark.sql.functions as F
test = sqlContext.createDataFrame([('A=id1a A=id2a B=id1b C=id1c B=id2b',1),('D=id1d A=id3a B=id3b C=id2c',2),('A=id4a C=id3c',3)],schema=['col1','id'])
tst_spl = test.withColumn("item",(F.split('col1'," ")))
tst_xpl = tst_spl.select(F.explode("item"))
tst_map = tst_xpl.withColumn("key",F.split('col','=')[0]).withColumn("value",F.split('col','=')[1]).drop('col')
#%%
tst_pivot = tst_map.groupby(F.lit(1)).pivot('key').agg(F.collect_list(('value'))).drop('1')
#%%
tst_arr = [tst_pivot.select(F.posexplode(coln)).withColumnRenamed('col',coln) for coln in tst_pivot.columns]
tst_fin = reduce(lambda df1,df2:df1.join(df2,on='pos',how='full'),tst_arr).orderBy('pos')
tst_fin.show()
+---+----+----+----+----+
|pos| A| B| C| D|
+---+----+----+----+----+
| 0|id3a|id3b|id1c|id1d|
| 1|id4a|id1b|id2c|null|
| 2|id1a|id2b|id3c|null|
| 3|id2a|null|null|null|
+---+----+----+----+----

Spark SHA Hash Coming Back Empty

I am trying to create a SHA256 hash of each row in a dataframe.
import org.apache.spark.sql.functions.{col, concat, sha2}
val finalResultWithHash = finalResult.withColumn("ROWHASH", sha2(concat(finalResult.columns.map(col):_*), 256))
When I had only one column in the dataframe it seemed to be working.
Later in the code I write the dataframe as a CSV and the rowhash column is empty.
I haven't been able to find any documentation on what I am doing wrong.
Thank you in advance.
Another way of doing it is by using a foldLeft():
val df2 = df.withColumn("rowsha",sha2(df.columns.foldLeft(lit(""))((x,y)=>concat(x,col(y))),256))
Folding will concat all columns left to right before hashing it:
df.withColumn("rowsha",sha2(df.columns.foldLeft(lit(""))((x,y)=>concat(x,col(y))),256)).explain()
== Physical Plan ==
*(1) Project [c1#10, c2#11, c3#12, c4#13, sha2(cast(concat(, c1#10, c2#11, c3#12, 4#13) as binary), 256) AS rowsha#165]
+- *(1) ...
However, if any of the columns in concatenation contain NULLs, the result will be also NULL. To safeguard against that you might want to use something like
val df2 = df.withColumn("rowsha",sha2(df.columns.foldLeft(lit(""))((x,y)=>concat(x,coalesce(col(y),lit("n/a"))),256))
For some reason below code works for me for multiple columns
val finalResultWithHash = personDF.withColumn("ROWHASH", sha2(concat(personDF.columns.map(col): _*), 256))
+-----+-----+---+------+--------------------+
|FName|LName|Age|Gender| ROWHASH|
+-----+-----+---+------+--------------------+
| A| B| 29| M|c4ae6946a295e9d74...|
| A| C| 12| |89a18fdc3ddb3c2fd...|
| B| D| 35| F|ef1c89dfc765c7e1e...|
| Q| D| 85| |cd91aa387a7e6a180...|
| W| R| 14| |e9ff9bb78fd93a13a...|
+-----+-----+---+------+--------------------+
Could it be just the bracket placement mistake ...

How to use countDistinct using a window function in Spark/Scala?

I need to use window function that is paritioned by 2 columns and do distinct count on the 3rd column and that as the 4th column. I can do count with out any issues, but using distinct count is throwing exception -
rg.apache.spark.sql.AnalysisException: Distinct window functions are not supported:
Is there any workaround for this ?
A previous answer suggested two possible techniques: approximate counting and size(collect_set(...)). Both have problems.
If you need an exact count, which is the main reason to use COUNT(DISTINCT ...) in big data, approximate counting will not do. Also, approximate counting actual error rates can vary quite significantly for small data.
size(collect_set(...)) may cause a substantial slowdown in processing of big data because it uses a mutable Scala HashSet, which is a pretty slow data structure. In addition, you may occasionally get strange results, e.g., if you run the query over an empty dataframe, because size(null) produces the counterintuitive -1. Spark's native distinct counting runs faster for a number of reasons, the main one being that it doesn't have to produce all the counted data in an array.
The typical approach to solving this problem is with a self-join. You group by whatever columns you need, compute the distinct count or any other aggregate function that cannot be used as a window function, and then join back to your original data.
Use approx_count_distinct (or) collect_set and size functions on window to mimic countDistinct functionality.
Example:
df.show()
//+---+---+---+
//| i| j| k|
//+---+---+---+
//| 1| a| c|
//| 2| b| d|
//| 1| a| c|
//| 2| b| e|
//+---+---+---+
import org.apache.spark.sql.expressions.Window
import org.apache.spark.sql.functions._
val windowSpec = Window.partitionBy("i","j")
df.withColumn("cnt",size(collect_set("k").over(windowSpec))).show()
//or using approx_count_distinct
df.withColumn("cnt",approx_count_distinct("k").over(windowSpec)).show()
//+---+---+---+---+
//| i| j| k|cnt|
//+---+---+---+---+
//| 2| b| d| 2|
//| 2| b| e| 2|
//| 1| a| c| 1| //as c value repeated for 1,a partition
//| 1| a| c| 1|
//+---+---+---+---+
Trying to improve Sim's answer, if you want to do this:
//val newColumnName: String = ...
//val colToCount: Column = ...
//val aggregatingCols: Seq[Column] = ...
df.withColumn(newColName, countDistinct(colToCount).over(partitionBy(aggregatingCols:_*)))
You must instead do this:
//val aggregatingCols: Seq[String] = ...
df.groupBy(aggregatingCols.head, aggregatingCols.tail:_*)
.agg(countDistinct(colToCount).as(newColName))
.select(newColName, aggregatingCols:_*)
.join(df, usingColumns = aggregatingCols)
This will return the number of distinct elements in the partition, using dense_rank() function. When we sum ascending and descending rank, we always get the total number of distinct elements + 1 :
dense_rank().over(Window.partitionBy("i").orderBy(c.asc)) + dense_rank().over(Window.partitionBy("i").orderBy(c.desc)) - 1

Advanced join two dataframe spark scala

I have to join two Dataframes.
Sample:
Dataframe1 looks like this
df1_col1 df1_col2
a ex1
b ex4
c ex2
d ex6
e ex3
Dataframe2
df2_col1 df2_col2
1 a,b,c
2 d,c,e
3 a,e,c
In result Dataframe I would like to get result like this
res_col1 res_col2 res_col3
a ex1 1
a ex1 3
b ex4 1
c ex2 1
c ex2 2
c ex2 3
d ex6 2
e ex3 2
e ex3 3
What will be the best way to achieve this join?
I have updated the code below
val df1 = sc.parallelize(Seq(("a","ex1"),("b","ex4"),("c","ex2"),("d","ex6"),("e","ex3")))
val df2 = sc.parallelize(Seq(List(("1","a,b,c"),("2","d,c,e")))).toDF
df2.withColumn("df2_col2_explode", explode(split($"_2", ","))).select($"_1".as("df2_col1"),$"df2_col2_explode").join(df1.select($"_1".as("df1_col1"),$"_2".as("df1_col2")), $"df1_col1"===$"df2_col2_explode","inner").show
You just need to split the values and generate multiple rows by exploding it and then join with the other dataframe.
You can refer this link, How to split pipe-separated column into multiple rows?
I used spark sql for this join, here is a part of code;
df1.createOrReplaceTempView("temp_v_df1")
df2.createOrReplaceTempView("temp_v_df2")
val df_result = spark.sql("""select
| b.df1_col1 as res_col1,
| b.df1_col2 as res_col2,
| a.df2_col1 as res_col3
| from (select df2_col1, exp_col
| from temp_v_df2
| lateral view explode(split(df2_col2,",")) dummy as exp_col) a
| join temp_v_df1 b on a.exp_col = b.df1_col1""".stripMargin)
I used spark scala data frame to achieve your desire output.
val df1 = sc.parallelize(Seq(("a","ex1"),("b","ex4"),("c","ex2"),("d","ex6"),("e","ex3"))).toDF("df1_col1","df1_col2")
val df2 = sc.parallelize(Seq((1,("a,b,c")),(2,("d,c,e")),(3,("a,e,c")))).toDF("df2_col1","df2_col2")
df2.withColumn("_tmp", explode(split($"df2_col2", "\\,"))).as("temp").join (df1,$"temp._tmp"===df1("df1_col1"),"inner").drop("_tmp","df2_col2").show
Desire Output
+--------+--------+--------+
|df2_col1|df1_col1|df1_col2|
+--------+--------+--------+
| 2| e| ex3|
| 3| e| ex3|
| 2| d| ex6|
| 1| c| ex2|
| 2| c| ex2|
| 3| c| ex2|
| 1| b| ex4|
| 1| a| ex1|
| 3| a| ex1|
+--------+--------+--------+
Rename the Column according to your requirement.
Here the screenshot of running code
Happy Hadoooooooooooooooppppppppppppppppppp

Scala: Any better way to join two DataFrames by the relationship from the third one

I have to two DataFrames, and want to outer join them. But the joining mapping is in another dataframe.
Now I am using below way, it works, but I hope there is more efficient way for I have >1,000,000 rows
val ta = sc.parallelize(Array(
(1,1,1),
(2,2,2)
)).toDF("A", "B", "C")
scala> ta.show
+---+---+---+
| A| B| C|
+---+---+---+
| 1| 1| 1|
| 2| 2| 2|
+---+---+---+
val tb = sc.parallelize(Array(
(2,1)
)).toDF("C", "D")
scala> tb.show
+---+---+
| C| D|
+---+---+
| 2| 1|
+---+---+
val tc = sc.parallelize(Array(
(1,1,1),
(2,2,2)
)).toDF("D", "E", "F")
scala> tc.show
+---+---+---+
| D| E| F|
+---+---+---+
| 1| 1| 1|
| 2| 2| 2|
+---+---+---+
scala> val tmp = ta.join(tb, Seq("C"), "left_outer")
tmp: org.apache.spark.sql.DataFrame = [C: int, A: int, B: int, D: int]
scala> tmp.show
+---+---+---+----+
| C| A| B| D|
+---+---+---+----+
| 1| 1| 1|null|
| 2| 2| 2| 1|
+---+---+---+----+
scala> tmp.join(tc, Seq("D"), "outer").show
+----+----+----+----+----+----+
| D| C| A| B| E| F|
+----+----+----+----+----+----+
|null| 1| 1| 1|null|null|
| 1| 2| 2| 2| 1| 1|
| 2|null|null|null| 2| 2|
+----+----+----+----+----+----+
As Umberto noted, a good reference on how to improve performance of your joins is Holden Karau and Rachel Warren's High Performance Spark > Chapter 4. Joins (SQL & Core).
From the standpoint of your code, running it as you noted or the SQL equivalent (as noted below) should result in about the same performance.
// Create initial tables
val ta = sc.parallelize(Array(
(1,1,1),
(2,2,2)
)).toDF("A", "B", "C")
val tb = sc.parallelize(Array(
(2,1)
)).toDF("C", "D")
val tc = sc.parallelize(Array(
(1,1,1),
(2,2,2)
)).toDF("D", "E", "F")
// _.createOrReplaceTempView
ta.createOrReplaceTempView("ta")
tb.createOrReplaceTempView("tb")
tc.createOrReplaceTempView("tc")
// SQL Query
spark.sql("
select tc.D, ta.A, ta.B, ta.C, tc.E, tc.F
from ta
left outer join tb
on tb.C = ta.C
full outer join tc
on tc.D = tb.D
")
The reason why is because the Spark SQL Catalyst Optimizer (as noted in the diagram below) takes the DataFrame query and builds up an optimized logical plan. A number of physical plans are developed and Spark SQL Engine's Cost Optimizer chooses the best physical plan and generates the code to produce the RDDs.
Saying this, the key concern is that when you're working with a lot of rows that use up a lot of memory, you have to take into account of the partitioning. For example, if you can ensure that the mapping DataFrame (tc) have the same / similar partitioning scheme as the other DataFrames (ta, tb) so that way you can have a co-located join (this is Figure 4-3 within High Performance Spark > Chapter 4. Join).
If the partitions for your three DataFrames (ta, tb, tc) all have different partitioning, this means the keys for your DataFrames will not have a 1-to-1 matching between the partitions. That is, this will result in a shuffle join (this is Figure 4-2 within High Performance Spark > Chapter 4. Join) which potentially could be more costly.
Basically, from the standpoint of your query, the concern is less about the query itself and more about the partitioning schemes for your DataFrames. But before experimenting too much with the partitioning schemes of your DataFrames, experiment with your queries to see if the default Spark SQL / DataFrame queries are able to take care of the partitioning by itself.