PySpark: How to create a nested JSON from spark data frame?

问题

I am trying to create a nested json from my spark dataframe which has data in following structure. The below code is creating a simple json with key and value. Could you please help

df.coalesce(1).write.format('json').save(data_output_file+"createjson.json", overwrite=True)

Update1: As per @MaxU answer,I converted the spark data frame to pandas and used group by. It is putting the last two fields in a nested array. How could i first put the category and count in nested array and then inside that array i want to put subcategory and count.

Sample text data:

Vendor_Name,count,Categories,Category_Count,Subcategory,Subcategory_Count
Vendor1,10,Category 1,4,Sub Category 1,1
Vendor1,10,Category 1,4,Sub Category 2,2
Vendor1,10,Category 1,4,Sub Category 3,3
Vendor1,10,Category 1,4,Sub Category 4,4

j = (data_pd.groupby(['vendor_name','vendor_Cnt','Category','Category_cnt'], as_index=False)
             .apply(lambda x: x[['Subcategory','subcategory_cnt']].to_dict('r'))
             .reset_index()
             .rename(columns={0:'subcategories'})
             .to_json(orient='records'))

[{
        "vendor_name": "Vendor 1",
        "count": 10,
        "categories": [{
            "name": "Category 1",
            "count": 4,
            "subCategories": [{
                    "name": "Sub Category 1",
                    "count": 1
                },
                {
                    "name": "Sub Category 2",
                    "count": 1
                },
                {
                    "name": "Sub Category 3",
                    "count": 1
                },
                {
                    "name": "Sub Category 4",
                    "count": 1
                }
            ]
        }]

回答1:

The easiest way to do this in python/pandas would be to use a series of nested generators using groupby I think:

def split_df(df):
    for (vendor, count), df_vendor in df.groupby(["Vendor_Name", "count"]):
        yield {
            "vendor_name": vendor,
            "count": count,
            "categories": list(split_category(df_vendor))
        }

def split_category(df_vendor):
    for (category, count), df_category in df_vendor.groupby(
        ["Categories", "Category_Count"]
    ):
        yield {
            "name": category,
            "count": count,
            "subCategories": list(split_subcategory(df_category)),
        }

def split_subcategory(df_category):
    for row in df.itertuples():
        yield {"name": row.Subcategory, "count": row.Subcategory_Count}

list(split_df(df))

[
    {
        "vendor_name": "Vendor1",
        "count": 10,
        "categories": [
            {
                "name": "Category 1",
                "count": 4,
                "subCategories": [
                    {"name": "Sub Category 1", "count": 1},
                    {"name": "Sub Category 2", "count": 2},
                    {"name": "Sub Category 3", "count": 3},
                    {"name": "Sub Category 4", "count": 4},
                ],
            }
        ],
    }
]

To export this to json, you'll need a way to export the np.int64

回答2:

You need to re-structure the whole dataframe for that.

"subCategories" is a struct stype.

from pyspark.sql import functions as F
df.withColumn(
  "subCategories",
  F.struct(
    F.col("subCategories").alias("name"),
    F.col("subcategory_count").alias("count")
  )
)

and then, groupBy and use F.collect_list to create the array.

At the end, you need to have only 1 record in your dataframe to get the result you expect.

来源：https://stackoverflow.com/questions/53477724/pyspark-how-to-create-a-nested-json-from-spark-data-frame