如何从PySpark中的RDD创建数据框？

Question

I have a RDD looks like this

[((0, Row(event_type_new=u'ALERT|VEHICLE_HEALTH_DATA|CHANGE_IN_HEALTH|DTC|B109F|', day=u'Fri')), 0), 
 ((1, Row(event_type_new=u'ALERT|VEHICLE_HEALTH_DATA|CHANGE_IN_HEALTH|DTC|B1115|HIGH MOUNTED STOP LAMP CONTROL', day=u'Sat')), 2)]

which has an index, a Row object ( event_type_new and day ), followed by a prediction (integer). How can I create a DataFrame with 3 columns including event_type_new , day , and Prediction .

I am using Spark 1.6.2 with PySpark API.

Thanks!

Answer 1

Transform your list into RDD first. Then map each element to Row . You can transform list of Row to dataframe easily using .toDF() method

from pyspark.sql import Row

ls = [((0, Row(event_type_new=u'ALERT|VEHICLE_HEALTH_DATA|CHANGE_IN_HEALTH|DTC|B109F|', day=u'Fri')), 0),
      ((1, Row(event_type_new=u'ALERT|VEHICLE_HEALTH_DATA|CHANGE_IN_HEALTH|DTC|B1115|HIGH MOUNTED STOP LAMP CONTROL', day=u'Sat')), 2)]
ls_rdd = sc.parallelize(ls)
ls_row = ls_rdd.map(lambda x: Row(**{'day': str(x[0][1].day), 'event_type': str(x[0][1].event_type_new), 'prediction': int(x[1])}))
df = ls_row.toDF()

When you run df.show() , it will look like this:

+---+--------------------+----------+
|day|          event_type|prediction|
+---+--------------------+----------+
|Fri|ALERT|VEHICLE_HEA...|         0|
|Sat|ALERT|VEHICLE_HEA...|         2|
+---+--------------------+----------+

Answer 2

I assume that this a collected RDD , because it looks like you got a list with tuples of a combination of Row and int objects. You can get your desired output with the following:

from pyspark.sql import Row


lst = [((0, Row(event_type_new=u'ALERT|VEHICLE_HEALTH_DATA|CHANGE_IN_HEALTH|DTC|B109F|', day=u'Fri')), 0),
       ((1, Row(event_type_new=u'ALERT|VEHICLE_HEALTH_DATA|CHANGE_IN_HEALTH|DTC|B1115|HIGH MOUNTED STOP LAMP CONTROL', day=u'Sat')), 2)]

output = []
for row in lst:
    vals = tuple(row[0][1]) + (row[1],)
    fields = row[0][1].__fields__ + ['prediction']
    row = Row(*vals)
    row.__fields__ = fields
    output.append(row)

df = sc.parallelize(output).toDF()
df.show()

You should get something like the following:

+---+--------------------+----------+
|day|      event_type_new|prediction|
+---+--------------------+----------+
|Fri|ALERT|VEHICLE_HEA...|         0|
|Sat|ALERT|VEHICLE_HEA...|         2|
+---+--------------------+----------+

I hope this helps.