Pypsark - Retain null values when using collect_list

Question

According to the accepted answer in pyspark collect_set or collect_list with groupby, when you do a collect_list on a certain column, the null values in this column are removed. I have checked and this is true.

But in my case, I need to keep the null columns -- How can I achieve this?

I did not find any info on this kind of a variant of collect_list function.

---

Background context to explain why I want nulls:

I have a dataframe df as below:

cId   |  eId  |  amount  |  city
1     |  2    |   20.0   |  Paris
1     |  2    |   30.0   |  Seoul
1     |  3    |   10.0   |  Phoenix
1     |  3    |   5.0    |  null

I want to write this to an Elasticsearch index with the following mapping:

"mappings": {
    "doc": {
        "properties": {
            "eId": { "type": "keyword" },
            "cId": { "type": "keyword" },
            "transactions": {
                "type": "nested", 
                "properties": {
                    "amount": { "type": "keyword" },
                    "city": { "type": "keyword" }
                }
            }
        }
    }
 }      

In order to conform to the nested mapping above, I transformed my df so that for each combination of eId and cId, I have an array of transactions like this:

df_nested = df.groupBy('eId','cId').agg(collect_list(struct('amount','city')).alias("transactions"))
df_nested.printSchema()
root
 |-- cId: integer (nullable = true)
 |-- eId: integer (nullable = true)
 |-- transactions: array (nullable = true)
 |    |-- element: struct (containsNull = true)
 |    |    |-- amount: float (nullable = true)
 |    |    |-- city: string (nullable = true)

Saving df_nested as a json file, there are the json records that I get:

{"cId":1,"eId":2,"transactions":[{"amount":20.0,"city":"Paris"},{"amount":30.0,"city":"Seoul"}]}
{"cId":1,"eId":3,"transactions":[{"amount":10.0,"city":"Phoenix"},{"amount":30.0}]}

As you can see - when cId=1 and eId=3, one of my array elements where amount=30.0 does not have the city attribute because this was a null in my original data (df). The nulls are being removed when I use the collect_list function.

However, when I try writing df_nested to elasticsearch with the above index, it errors because there is a schema mismatch. This is basically the reason as to why I want to retain my nulls after applying the collect_list function.

---

Lys

Is it an option to replace the null values with something else, perhaps the string 'null'? – Bakeman

Answer 1

    from pyspark.sql.functions import create_map, collect_list, lit, col, to_json, from_json
    from pyspark import SparkContext, SparkConf
    from pyspark.sql import SQLContext, HiveContext, SparkSession, types, Row
    from pyspark.sql import functions as f
    import os
    
    app_name = "CollList"
    conf = SparkConf().setAppName(app_name)
    spark = SparkSession.builder.appName(app_name).config(conf=conf).enableHiveSupport().getOrCreate()
    
    df = spark.createDataFrame([[1, 2, 20.0, "Paris"], [1, 2, 30.0, "Seoul"],
        [1, 3, 10.0, "Phoenix"], [1, 3, 5.0, None]],
        ["cId", "eId", "amount", "city"])
    print("Actual data")
    df.show(10,False)
```
Actual data
+---+---+------+-------+
|cId|eId|amount|city   |
+---+---+------+-------+
|1  |2  |20.0  |Paris  |
|1  |2  |30.0  |Seoul  |
|1  |3  |10.0  |Phoenix|
|1  |3  |5.0   |null   |
+---+---+------+-------+
```
    #collect_list that skips null columns
    df1 = df.groupBy(f.col('city'))\
            .agg(f.collect_list(f.to_json(f.struct([f.col(x).alias(x) for x in (c for c in df.columns if c != 'cId' and c != 'eId' )])))).alias('newcol')
    print("Collect List Data - Missing Null Columns in the list")
    df1.show(10, False)
```
Collect List Data - Missing Null Columns in the list
+-------+-------------------------------------------------------------------------------------------------------------------+
|city   |collect_list(structstojson(named_struct(NamePlaceholder(), amount AS `amount`, NamePlaceholder(), city AS `city`)))|
+-------+-------------------------------------------------------------------------------------------------------------------+
|Phoenix|[{"amount":10.0,"city":"Phoenix"}]                                                                                 |
|null   |[{"amount":5.0}]                                                                                                   |
|Paris  |[{"amount":20.0,"city":"Paris"}]                                                                                   |
|Seoul  |[{"amount":30.0,"city":"Seoul"}]                                                                                   |
+-------+-------------------------------------------------------------------------------------------------------------------+
``` 
    my_list = []
    for x in (c for c in df.columns if c != 'cId' and c != 'eId' ):
        my_list.append(lit(x))
        my_list.append(col(x))
    
    grp_by = ["eId","cId"]
    df_nested = df.withColumn("transactions", create_map(my_list))\
                  .groupBy(grp_by)\
                  .agg(collect_list(f.to_json("transactions")).alias("transactions"))
    
    print("collect list after create_map")
    df_nested.show(10,False)
```
collect list after create_map
+---+---+--------------------------------------------------------------------+
|eId|cId|transactions                                                        |
+---+---+--------------------------------------------------------------------+
|2  |1  |[{"amount":"20.0","city":"Paris"}, {"amount":"30.0","city":"Seoul"}]|
|3  |1  |[{"amount":"10.0","city":"Phoenix"}, {"amount":"5.0","city":null}]  |
+---+---+--------------------------------------------------------------------+
```   

Paulina

Keep in mind that create_map will cast the key: value as string: string, so the values of amount are strings instead of floats – Neolith