I read a .csv file to Spark DataFrame. For a DoubleType column is there a way to specify at the time of the file read that this column should be rounded to 2 decimal places? I'm also supplying a custom schema to the DataFrameReader API call. Here's my schema and API calls:
val customSchema = StructType(Array(StructField("id_1", IntegerType, true),
StructField("id_2", IntegerType, true),
StructField("id_3", DoubleType, true)))
#using Spark's CSV reader with custom schema
#spark == SparkSession()
val parsedSchema = spark.read.format("csv").schema(customSchema).option("header", "true").option("nullvalue", "?").load("C:\\Scala\\SparkAnalytics\\block_1.csv")
After the file read into DataFrame I can round the decimals like:
parsedSchema.withColumn("cmp_fname_c1", round($"cmp_fname_c1", 3))
But this creates a new DataFrame, so I'd also like to know if it can be done in-place instead of creating a new DataFrame.
Thanks
You can specify, say, DecimalType(10, 2) for the DoubleType column in your customSchema when loading your CSV file. Let's say you have a CSV file with the following content:
id_1,id_2,Id_3
1,10,5.555
2,20,6.0
3,30,7.444
Sample code below:
import org.apache.spark.sql.types._
val customSchema = StructType(Array(
StructField("id_1", IntegerType, true),
StructField("id_2", IntegerType, true),
StructField("id_3", DecimalType(10, 2), true)
))
spark.read.format("csv").schema(customSchema).
option("header", "true").option("nullvalue", "?").
load("/path/to/csvfile").
show
// +----+----+----+
// |id_1|id_2|id_3|
// +----+----+----+
// | 1| 10|5.56|
// | 2| 20|6.00|
// | 3| 30|7.44|
// +----+----+----+
Related
I am trying to read csv file and convert into dataframe.
input.txt
4324,'Andy',43.5,20.3,53.21
2342,'Sam',22.1
3248,'Jane',11.05,12.87
6457,'Bob',32.1,75.23,71.6
Schema: Id, Name,Jan,Feb,March
As you see the csv file doesn't have "," if there are no trailing expenses.
Code:
from pyspark.sql.types import *
input1= sc.textFile('/FileStore/tables/input.txt').map(lambda x: x.split(","))
schema = StructType([StructField('Id',StringType(),True), StructField('Name',StringType(),True), StructField('Jan',StringType(),True), StructField('Feb',StringType(),True), StructField('Mar',StringType(),True)])
df3 = sqlContext.createDataFrame(input1, schema)
I get ValueError: Length of object (4) does not match with length of fields (5). How do I resolve this?
I would first import the file using pandas which should handle everything for you. From there you can then convert the pandas DataFrame to spark and do all your usual stuff. I copied your example txt file and quickly wrote up some code to confirm that it would all work:
import pandas as pd
# Reading in txt file as csv
df_pandas = pd.read_csv('<your location>/test.txt',
sep=",")
# Converting to spark dataframe and displaying
df_spark = spark.createDataFrame(df_pandas)
display(df_pandas)
Which produced the following output:
The faster method would be to import through spark:
# Importing csv file using pyspark
csv_import = sqlContext.read\
.format('csv')\
.options(sep = ',', header='true', inferSchema='true')\
.load('<your location>/test.txt')
display(csv_import)
Which gives the same output.
from pyspark.sql.types import *
from pyspark.sql import SparkSession
spark = SparkSession.builder.appName("Test").getOrCreate()
fields = [StructField('Id', StringType(), True), StructField('Name', StringType(), True),
StructField('Jan', StringType(), True), StructField('Feb', StringType(), True),
StructField('Mar', StringType(), True)]
schema = StructType(fields)
data = spark.read.format("csv").load("test2.txt")
df3 = spark.createDataFrame(data.rdd, schema)
df3.show()
Output:
+----+------+-----+-----+-----+
| Id| Name| Jan| Feb| Mar|
+----+------+-----+-----+-----+
|4324|'Andy'| 43.5| 20.3|53.21|
|2342| 'Sam'| 22.1| null| null|
|3248|'Jane'|11.05|12.87| null|
|6457| 'Bob'| 32.1|75.23| 71.6|
+----+------+-----+-----+-----+
Here are a couple options for you to consider. These use the wildcard character, so you can loop through all folders and sub-folders, look for files with names that match a specific pattern, and merge everything into a dingle dataframe.
val myDFCsv = spark.read.format("csv")
.option("sep",",")
.option("inferSchema","true")
.option("header","true")
.load("mnt/rawdata/2019/01/01/client/ABC*.gz")
myDFCsv.show()
myDFCsv.head()
myDFCsv.count()
//////////////////////////////////////////
// If you also need to load the filename
import org.apache.spark.sql.functions.input_file_name
val myDFCsv = spark.read.format("csv")
.option("sep",",")
.option("inferSchema","true")
.option("header","true")
.load("mnt/rawdata/2019/01/01/client/ABC*.gz")
.withColumn("file_name",input_file_name())
myDFCsv.show(false)
myDFCsv.head()
myDFCsv.count()
Spark has a Permissive mode for reading CSV files which stores the corroupt records into a separate column named _corroupt_record.
permissive -
Sets all fields to null when it encounters a corrupted record and places all corrupted records in a string column
called _corrupt_record
However, when I am trying following example, I don't see any column named _corroupt_record. the reocords which doesn't match with schema appears to be null
data.csv
data
10.00
11.00
$12.00
$13
gaurang
code
import org.apache.spark.sql.types.{StructField, StructType, StringType, LongType, DecimalType}
val schema = new StructType(Array(
new StructField("value", DecimalType(25,10), false)
))
val df = spark.read.format("csv")
.option("header", "true")
.option("mode", "PERMISSIVE")
.schema(schema)
.load("../test.csv")
schema
scala> df.printSchema()
root
|-- value: decimal(25,10) (nullable = true)
scala> df.show()
+-------------+
| value|
+-------------+
|10.0000000000|
|11.0000000000|
| null|
| null|
| null|
+-------------+
If I change the mode to FAILFAST I am getting error when I try to see data.
Adding the _corroup_record as suggested by Andrew and Prateek resolved the issue.
import org.apache.spark.sql.types.{StructField, StructType, StringType, LongType, DecimalType}
val schema = new StructType(Array(
new StructField("value", DecimalType(25,10), true),
new StructField("_corrupt_record", StringType, true)
))
val df = spark.read.format("csv")
.option("header", "true")
.option("mode", "PERMISSIVE")
.schema(schema)
.load("../test.csv")
querying Data
scala> df.show()
+-------------+---------------+
| value|_corrupt_record|
+-------------+---------------+
|10.0000000000| null|
|11.0000000000| null|
| null| $12.00|
| null| $13|
| null| gaurang|
+-------------+---------------+
I want to convert RDD to DataFrame using StructType. But item "Broken,Line," would cause error. Is there an elegant way to process record like this? Thanks.
import org.apache.spark.sql.types.{StructType, StructField, StringType}
import org.apache.spark.sql.Row
val mySchema = StructType(Array(
StructField("colA", StringType, true),
StructField("colB", StringType, true),
StructField("colC", StringType, true)))
val x = List("97573,Start,eee", "9713,END,Good", "Broken,Line,")
val inputx = sc.parallelize(x).
| map((x:String) => Row.fromSeq(x.split(",").slice(0,mySchema.size).toSeq))
val df = spark.createDataFrame(inputx, mySchema)
df.show
Error would be like this:
Name: org.apache.spark.SparkException Message: Job aborted due to
stage failure: Task 0 in stage 14.0 failed 1 times, most recent
failure: Lost task 0.0 in stage 14.0 (TID 14, localhost, executor
driver): java.lang.RuntimeException: Error while encoding:
java.lang.ArrayIndexOutOfBoundsException: 2
I'm using:
Spark: 2.2.0
Scala: 2.11.8
And I ran the code in spark-shell.
Row.fromSeq on which we apply your schema throws the error that you are getting. Your third element in your list contains just 2 elements. You can't transform it into a Row with three elements unless you add a null value instead of the missing value.
And when creating your DataFrame, Spark is expecting 3 elements per Row on which to apply the schema, thus the error.
A quick and dirty solution would be to use scala.util.Try to get fields separately :
import org.apache.spark.sql.types.{StructType, StructField, StringType}
import org.apache.spark.sql.Row
import scala.util.Try
val mySchema = StructType(Array(StructField("colA", StringType, true), StructField("colB", StringType, true), StructField("colC", StringType, true)))
val l = List("97573,Start,eee", "9713,END,Good", "Broken,Line,")
val rdd = sc.parallelize(l).map {
x => {
val fields = x.split(",").slice(0, mySchema.size)
val f1 = Try(fields(0)).getOrElse("")
val f2 = Try(fields(1)).getOrElse("")
val f3 = Try(fields(2)).getOrElse("")
Row(f1, f2, f3)
}
}
val df = spark.createDataFrame(rdd, mySchema)
df.show
// +------+-----+----+
// | colA| colB|colC|
// +------+-----+----+
// | 97573|Start| eee|
// | 9713| END|Good|
// |Broken| Line| |
// +------+-----+----+
I wouldn't say that it's an elegant solution like you've asked. Parsing strings is never elegant ! You ought using the csv source to read it correctly (or spark-csv for < 2.x).
How do you reference a pyspark dataframe when in the execution of an UDF on another dataframe?
Here's a dummy example. I am creating two dataframes scores and lastnames, and within each lies a column that is the same across the two dataframes. In the UDF applied on scores, I want to filter on lastnames and return a string found in lastname.
from pyspark import SparkContext
from pyspark import SparkConf
from pyspark.sql import SQLContext
from pyspark.sql.types import *
sc = SparkContext("local")
sqlCtx = SQLContext(sc)
# Generate Random Data
import itertools
import random
student_ids = ['student1', 'student2', 'student3']
subjects = ['Math', 'Biology', 'Chemistry', 'Physics']
random.seed(1)
data = []
for (student_id, subject) in itertools.product(student_ids, subjects):
data.append((student_id, subject, random.randint(0, 100)))
from pyspark.sql.types import StructType, StructField, IntegerType, StringType
schema = StructType([
StructField("student_id", StringType(), nullable=False),
StructField("subject", StringType(), nullable=False),
StructField("score", IntegerType(), nullable=False)
])
# Create DataFrame
rdd = sc.parallelize(data)
scores = sqlCtx.createDataFrame(rdd, schema)
# create another dataframe
last_name = ["Granger", "Weasley", "Potter"]
data2 = []
for i in range(len(student_ids)):
data2.append((student_ids[i], last_name[i]))
schema = StructType([
StructField("student_id", StringType(), nullable=False),
StructField("last_name", StringType(), nullable=False)
])
rdd = sc.parallelize(data2)
lastnames = sqlCtx.createDataFrame(rdd, schema)
scores.show()
lastnames.show()
from pyspark.sql.functions import udf
def getLastName(sid):
tmp_df = lastnames.filter(lastnames.student_id == sid)
return tmp_df.last_name
getLastName_udf = udf(getLastName, StringType())
scores.withColumn("last_name", getLastName_udf("student_id")).show(10)
And the following is the last part of the trace:
Py4JError: An error occurred while calling o114.__getnewargs__. Trace:
py4j.Py4JException: Method __getnewargs__([]) does not exist
at py4j.reflection.ReflectionEngine.getMethod(ReflectionEngine.java:335)
at py4j.reflection.ReflectionEngine.getMethod(ReflectionEngine.java:344)
at py4j.Gateway.invoke(Gateway.java:252)
at py4j.commands.AbstractCommand.invokeMethod(AbstractCommand.java:133)
at py4j.commands.CallCommand.execute(CallCommand.java:79)
at py4j.GatewayConnection.run(GatewayConnection.java:209)
at java.lang.Thread.run(Thread.java:745)
You can't directly reference a dataframe (or an RDD) from inside a UDF. The DataFrame object is a handle on your driver that spark uses to represent the data and actions that will happen out on the cluster. The code inside your UDF's will run out on the cluster at a time of Spark's choosing. Spark does this by serializing that code, and making copies of any variables included in the closure and sending them out to each worker.
What instead you want to do, is use the constructs Spark provides in it's API to join/combine the two DataFrames. If one of the data sets is small, you can manually send out the data in a broadcast variable, and then access it from your UDF. Otherwise, you can just create the two dataframes like you did, then use the join operation to combine them. Something like this should work:
joined = scores.withColumnRenamed("student_id", "join_id")
joined = joined.join(lastnames, joined.join_id == lastnames.student_id)\
.drop("join_id")
joined.show()
+---------+-----+----------+---------+
| subject|score|student_id|last_name|
+---------+-----+----------+---------+
| Math| 13| student1| Granger|
| Biology| 85| student1| Granger|
|Chemistry| 77| student1| Granger|
| Physics| 25| student1| Granger|
| Math| 50| student2| Weasley|
| Biology| 45| student2| Weasley|
|Chemistry| 65| student2| Weasley|
| Physics| 79| student2| Weasley|
| Math| 9| student3| Potter|
| Biology| 2| student3| Potter|
|Chemistry| 84| student3| Potter|
| Physics| 43| student3| Potter|
+---------+-----+----------+---------+
It's also worth noting, that under the hood Spark DataFrames has an optimization where a DataFrame that is part of a join can be converted to a broadcast variable to avoid a shuffle if it is small enough. So if you do the join method listed above, you should get the best possible performance, without sacrificing the ability to handle larger data sets.
Changing pair to dictionary for easy lookup of names
data2 = {}
for i in range(len(student_ids)):
data2[student_ids[i]] = last_name[i]
Instead of creating rdd and making it to df create broadcast variable
//rdd = sc.parallelize(data2)
//lastnames = sqlCtx.createDataFrame(rdd, schema)
lastnames = sc.broadcast(data2)
Now access this in udf with values attr on broadcast variable(lastnames).
from pyspark.sql.functions import udf
def getLastName(sid):
return lastnames.value[sid]
I have a Spark Dataframe in that consists of a series of dates:
from pyspark.sql import SQLContext
from pyspark.sql import Row
from pyspark.sql.types import *
sqlContext = SQLContext(sc)
import pandas as pd
rdd = sc.parallelizesc.parallelize([('X01','2014-02-13T12:36:14.899','2014-02-13T12:31:56.876','sip:4534454450'),
('X02','2014-02-13T12:35:37.405','2014-02-13T12:32:13.321','sip:6413445440'),
('X03','2014-02-13T12:36:03.825','2014-02-13T12:32:15.229','sip:4534437492'),
('XO4','2014-02-13T12:37:05.460','2014-02-13T12:32:36.881','sip:6474454453'),
('XO5','2014-02-13T12:36:52.721','2014-02-13T12:33:30.323','sip:8874458555')])
schema = StructType([StructField('ID', StringType(), True),
StructField('EndDateTime', StringType(), True),
StructField('StartDateTime', StringType(), True)])
df = sqlContext.createDataFrame(rdd, schema)
What I want to do is find duration by subtracting EndDateTime and StartDateTime. I figured I'd try and do this using a function:
# Function to calculate time delta
def time_delta(y,x):
end = pd.to_datetime(y)
start = pd.to_datetime(x)
delta = (end-start)
return delta
# create new RDD and add new column 'Duration' by applying time_delta function
df2 = df.withColumn('Duration', time_delta(df.EndDateTime, df.StartDateTime))
However this just gives me:
>>> df2.show()
ID EndDateTime StartDateTime ANI Duration
X01 2014-02-13T12:36:... 2014-02-13T12:31:... sip:4534454450 null
X02 2014-02-13T12:35:... 2014-02-13T12:32:... sip:6413445440 null
X03 2014-02-13T12:36:... 2014-02-13T12:32:... sip:4534437492 null
XO4 2014-02-13T12:37:... 2014-02-13T12:32:... sip:6474454453 null
XO5 2014-02-13T12:36:... 2014-02-13T12:33:... sip:8874458555 null
I'm not sure if my approach is correct or not. If not, I'd gladly accept another suggested way to achieve this.
As of Spark 1.5 you can use unix_timestamp:
from pyspark.sql import functions as F
timeFmt = "yyyy-MM-dd'T'HH:mm:ss.SSS"
timeDiff = (F.unix_timestamp('EndDateTime', format=timeFmt)
- F.unix_timestamp('StartDateTime', format=timeFmt))
df = df.withColumn("Duration", timeDiff)
Note the Java style time format.
>>> df.show()
+---+--------------------+--------------------+--------+
| ID| EndDateTime| StartDateTime|Duration|
+---+--------------------+--------------------+--------+
|X01|2014-02-13T12:36:...|2014-02-13T12:31:...| 258|
|X02|2014-02-13T12:35:...|2014-02-13T12:32:...| 204|
|X03|2014-02-13T12:36:...|2014-02-13T12:32:...| 228|
|XO4|2014-02-13T12:37:...|2014-02-13T12:32:...| 269|
|XO5|2014-02-13T12:36:...|2014-02-13T12:33:...| 202|
+---+--------------------+--------------------+--------+
Thanks to David Griffin. Here's how to do this for future reference.
from pyspark.sql import SQLContext, Row
sqlContext = SQLContext(sc)
from pyspark.sql.types import StringType, IntegerType, StructType, StructField
from pyspark.sql.functions import udf
# Build sample data
rdd = sc.parallelize([('X01','2014-02-13T12:36:14.899','2014-02-13T12:31:56.876'),
('X02','2014-02-13T12:35:37.405','2014-02-13T12:32:13.321'),
('X03','2014-02-13T12:36:03.825','2014-02-13T12:32:15.229'),
('XO4','2014-02-13T12:37:05.460','2014-02-13T12:32:36.881'),
('XO5','2014-02-13T12:36:52.721','2014-02-13T12:33:30.323')])
schema = StructType([StructField('ID', StringType(), True),
StructField('EndDateTime', StringType(), True),
StructField('StartDateTime', StringType(), True)])
df = sqlContext.createDataFrame(rdd, schema)
# define timedelta function (obtain duration in seconds)
def time_delta(y,x):
from datetime import datetime
end = datetime.strptime(y, '%Y-%m-%dT%H:%M:%S.%f')
start = datetime.strptime(x, '%Y-%m-%dT%H:%M:%S.%f')
delta = (end-start).total_seconds()
return delta
# register as a UDF
f = udf(time_delta, IntegerType())
# Apply function
df2 = df.withColumn('Duration', f(df.EndDateTime, df.StartDateTime))
Applying time_delta() will give you duration in seconds:
>>> df2.show()
ID EndDateTime StartDateTime Duration
X01 2014-02-13T12:36:... 2014-02-13T12:31:... 258
X02 2014-02-13T12:35:... 2014-02-13T12:32:... 204
X03 2014-02-13T12:36:... 2014-02-13T12:32:... 228
XO4 2014-02-13T12:37:... 2014-02-13T12:32:... 268
XO5 2014-02-13T12:36:... 2014-02-13T12:33:... 202
datediff(Column end, Column start)
Returns the number of days from start to end.
https://spark.apache.org/docs/1.6.2/api/java/org/apache/spark/sql/functions.html
This can be done in spark-sql by converting the string date to timestamp and then getting the difference.
1: Convert to timestamp:
CAST(UNIX_TIMESTAMP(MY_COL_NAME,'dd-MMM-yy') as TIMESTAMP)
2: Get the difference between dates using datediff function.
This will be combined in a nested function like:
spark.sql("select COL_1, COL_2, datediff( CAST( UNIX_TIMESTAMP( COL_1,'dd-MMM-yy') as TIMESTAMP), CAST( UNIX_TIMESTAMP( COL_2,'dd-MMM-yy') as TIMESTAMP) ) as LAG_in_days from MyTable")
Below is the result:
+---------+---------+-----------+
| COL_1| COL_2|LAG_in_days|
+---------+---------+-----------+
|24-JAN-17|16-JAN-17| 8|
|19-JAN-05|18-JAN-05| 1|
|23-MAY-06|23-MAY-06| 0|
|18-AUG-06|17-AUG-06| 1|
+---------+---------+-----------+
Reference: https://docs-snaplogic.atlassian.net/wiki/spaces/SD/pages/2458071/Date+Functions+and+Properties+Spark+SQL
Use DoubleType instead of IntegerType
from pyspark.sql import SQLContext, Row
sqlContext = SQLContext(sc)
from pyspark.sql.types import StringType, IntegerType, StructType, StructField
from pyspark.sql.functions import udf
# Build sample data
rdd = sc.parallelize([('X01','2014-02-13T12:36:14.899','2014-02-13T12:31:56.876'),
('X02','2014-02-13T12:35:37.405','2014-02-13T12:32:13.321'),
('X03','2014-02-13T12:36:03.825','2014-02-13T12:32:15.229'),
('XO4','2014-02-13T12:37:05.460','2014-02-13T12:32:36.881'),
('XO5','2014-02-13T12:36:52.721','2014-02-13T12:33:30.323')])
schema = StructType([StructField('ID', StringType(), True),
StructField('EndDateTime', StringType(), True),
StructField('StartDateTime', StringType(), True)])
df = sqlContext.createDataFrame(rdd, schema)
# define timedelta function (obtain duration in seconds)
def time_delta(y,x):
from datetime import datetime
end = datetime.strptime(y, '%Y-%m-%dT%H:%M:%S.%f')
start = datetime.strptime(x, '%Y-%m-%dT%H:%M:%S.%f')
delta = (end-start).total_seconds()
return delta
# register as a UDF
f = udf(time_delta, DoubleType())
# Apply function
df2 = df.withColumn('Duration', f(df.EndDateTime, df.StartDateTime))
Here is a working version for spark 2.x derived from jason's answer
from pyspark import SparkContext, SparkConf
from pyspark.sql import SparkSession,SQLContext
from pyspark.sql.types import StringType, StructType, StructField
sc = SparkContext()
sqlContext = SQLContext(sc)
spark = SparkSession.builder.appName("Python Spark SQL basic example").getOrCreate()
rdd = sc.parallelize([('X01','2014-02-13T12:36:14.899','2014-02-13T12:31:56.876'),
('X02','2014-02-13T12:35:37.405','2014-02-13T12:32:13.321'),
('X03','2014-02-13T12:36:03.825','2014-02-13T12:32:15.229'),
('XO4','2014-02-13T12:37:05.460','2014-02-13T12:32:36.881'),
('XO5','2014-02-13T12:36:52.721','2014-02-13T12:33:30.323')])
schema = StructType([StructField('ID', StringType(), True),
StructField('EndDateTime', StringType(), True),
StructField('StartDateTime', StringType(), True)])
df = sqlContext.createDataFrame(rdd, schema)
# register as a UDF
from datetime import datetime
sqlContext.registerFunction("time_delta", lambda y,x:(datetime.strptime(y, '%Y-%m-%dT%H:%M:%S.%f')-datetime.strptime(x, '%Y-%m-%dT%H:%M:%S.%f')).total_seconds())
df.createOrReplaceTempView("Test_table")
spark.sql("SELECT ID,EndDateTime,StartDateTime,time_delta(EndDateTime,StartDateTime) as time_delta FROM Test_table").show()
sc.stop()