将Python类对象转换为DataFrame

Question

How do I convert a Python class object that has fields that instantiate other classes to a DataFrame? I tried the following code below but it does not work.

I can get it to work when I take out self.address = Address() and self.agency_contact_info = ContactInfo()

class Address:
    def __init__(self):
        self.address_one = "address 1"
        self.address_two = "P.O. BOX 1"                  

class ContactInfo:
    def __init__(self):
        self.person_name = "Me"
        self.phone_number = "999-999-9999"    

class AgencyRecord:
    def __init__(self):
        self.agency_code = "00"
        self.agency_id = "000"
        self.agency_name = "Some Agency"
        self.address = Address()
        self.agency_contact_info = ContactInfo()            

def create_data():
    data = {}

    for i in range(0, 3):
        alc = AgencyRecord()                    
        data[i] = alc   

    column_list = [
        'agency_code', 'agency_id', 'agency_name', 
        'address_one', 'address_two', 'person_name', 'phone_number'
    ]

    spark.createDataFrame(
        list(data.values()),
        column_list
    ).createOrReplaceTempView("MyTempTable")

Answer 1

Quoting myself again:

I find it's useful to think of the argument to createDataFrame() as a list of [iterables] where each entry in the list corresponds to a row in the DataFrame and each element of the [iterable] corresponds to a column.

---

So you need to convert each of your objects into an interable where each element corresponds to the columns in column_list .

I wouldn't necessarily endorse it (there's almost surely a better way), but here is one hacky approach you can take to modify your code accordingly:

You can take advantage of the fact that python objects have a self.__dict__ that you can use to retrieve parameters by name. First, update your AgencyRecord class to pull in the fields from the Address and ContactInfo classes:

class AgencyRecord:
    def __init__(self):
        self.agency_code = "00"
        self.agency_id = "000"
        self.agency_name = "Some Agency"
        self.address = Address()
        self.agency_contact_info = ContactInfo()

        # makes the variables of the contained classes members of this class
        self.__dict__.update(self.address.__dict__)
        self.__dict__.update(self.agency_contact_info.__dict__)

Now we can reference each column in column_list by name for any instance of an AgencyRecord .

Modify the create_data as follows (I've also changed this to return a DataFrame, rather than registering a temp view)

def create_data():
    data = {}

    for i in range(0, 3):
        alc = AgencyRecord()                    
        data[i] = alc   

    column_list = [
        'agency_code', 'agency_id', 'agency_name', 
        'address_one', 'address_two', 'person_name', 'phone_number'
    ]

    values = [
        [data[record].__dict__[c] for c in column_list]
        for record in data
    ]

    return spark.createDataFrame(values, column_list)

Now you can do:

temp_df = create_data()
temp_df.show()
#+-----------+---------+-----------+-----------+-----------+-----------+------------+
#|agency_code|agency_id|agency_name|address_one|address_two|person_name|phone_number|
#+-----------+---------+-----------+-----------+-----------+-----------+------------+
#|         00|      000|Some Agency|  address 1| P.O. BOX 1|         Me|999-999-9999|
#|         00|      000|Some Agency|  address 1| P.O. BOX 1|         Me|999-999-9999|
#|         00|      000|Some Agency|  address 1| P.O. BOX 1|         Me|999-999-9999|
#+-----------+---------+-----------+-----------+-----------+-----------+------------+