Compare two datasets in pyspark

问题

I have 2 datasets.

Example Dataset 1:

id     |   model |   first_name   |      last_name
-----------------------------------------------------------
1234   |   32    |    456765      |   [456700,987565]
-----------------------------------------------------------
4539   |   20    |    123211      |   [893456,123456]
-----------------------------------------------------------

Some times one of the columns first_name and last_name is empty.

Example dataset 2:

number  |  matricule   | name       |    model
----------------------------------------------------------
AA      |  0009        |  456765    |     32
----------------------------------------------------------
AA      |  0009        |  893456    |     32
----------------------------------------------------------
AA      |  0009        |  456700    |     32
----------------------------------------------------------
AA      |  0008        |  456700    |     32
----------------------------------------------------------
AA      |  0008        |  987565    |     32

For one matricule we can find more name and model, like in my example just above. What I should do:

For each row from the Dataset 1, I take the 3 columns: model, first_name and last_name and look for them in Dataset 2 if exist / match according the matricule elements.

I should compare:

• model by model ==> if model (dataset 1) exist in model (dataset 2) ==> match

• if first_name exist in name ==> no match. If first_name not exist in name ==> match

• if last_name exist in name ==> match. When I have two values of last_name, the both should exist in name of dataset 2 to be matched.

Example:

Rows 1 from Dataset 1 is:

id     |   model |   first_name   |      last_name
------------------------------------------------------
1234   |   32    |    456765      |   [456700,987565]

For matricule 0009 in dataset 2, I have:

number  |  matricule   | name       |    model
----------------------------------------------------------
AA      |  0009        |  456765    |     32
----------------------------------------------------------
AA      |  0009        |  893456    |     32
----------------------------------------------------------
AA      |  0009        |  456700    |     32

So:

first_name (456765) is exist in name of dataset 2 when matriule =0009 ==> no match

last_name, only 456700 is exist ==> no match

model (32) is exist in model of dataset 2 ==> match

So I skip the matricule 0009. And pass to compare second line in dataset 1 with the elements of matricule 0008.

For matricule 0008 in dataset 2, I have:

----------------------------------------------------------
AA      |  0008        |  456700    |     32
----------------------------------------------------------
AA      |  0008        |  987565    |     32

Always we are in the first rows of dataset 1:

first_name (456765) is not exist in name of dataset 2 when matricule=0008 ==> match

last_name, the both values are exist in name of dataset 2 when matricule = 0008, ==> match

model is exist in model of dataset 2 when matricule =0008==> match

When I find all match, I create a new dataset contain:

number | id     |  matricule
-----------------------------------
AA     | 1234   | 0008
-----------------------------------

I hope that I was clear. Someone can help me please.

回答1:

You can use join on the conditions of matching.

First, you can group by the second DataFrame and collect name column into a list:

df2 = df2.groupBy("number", "model", "matricule").agg(collect_list("name").alias("names"))
f2.show(truncate=False)

#+------+-----+---------+------------------------+
#|number|model|matricule|names                   |
#+------+-----+---------+------------------------+
#|AA    |32   |0009     |[456765, 893456, 456700]|
#|AA    |32   |0008     |[456700, 987565]        |
#+------+-----+---------+------------------------+

Now, join df1 and df2. For conditions 1 and 2, it is somehow simple to check. For the third one, you can use array_except avaliable from Spark 2.4+ (there should be no elements from last_name column that are not in names and vice-versa):

join_condition = (col("df1.model") == col("df2.model")) \
                 & ~expr("array_contains(df2.names, df1.first_name)") \
                 & (size(expr("array_except(df2.names, df1.last_name)")) == lit(0)) \
                 & (size(expr("array_except(df1.last_name, df2.names)")) == lit(0))


df_result = df1.alias("df1").join(df2.alias("df2"), join_condition)

Finally, select desired columns from the join result:

df_result.select("number", "id", "matricule").show(truncate=False)

#+------+----+---------+
#|number|id  |matricule|
#+------+----+---------+
#|AA    |1234|0008     |
#+------+----+---------+

来源：https://stackoverflow.com/questions/60190375/compare-two-datasets-in-pyspark