PySpark - 从 Dataframe 中删除第一行

Question

I have a.txt file with a header, which I'd like to remove. The file looks like this:

Entry  Per  Account     Description               
 16524  01  3930621977  TXNPUES                     
191675  01  2368183100  OUNHQEX            
191667  01  3714468136  GHAKASC             
191673  01  2632703881  PAHFSAP              
 80495  01  2766389794  XDZANTV                    
 80507  01  4609266335  BWWYEZL                   
 80509  01  1092717420  QJYPKVO                  
 80497  01  3386366766  SOQLCMU                  
191669  01  5905893739  FYIWNKA             
191671  01  2749355876  CBMJTLP 

# Create spark session
spark = SparkSession.builder.master("local").appName("fixed-width"                          )\
                                            .config("spark.some.config.option", "some-value")\
                                            .getOrCreate()

# Read in fixed-width text file into DataFrame
df = spark.read.option("header"     , "true" )\
               .option("inferSchema", "true" )\
               .text(file                    )
df.show()
df.printSchema()

Which returns:

+--------------------+
|               value|
+--------------------+
|Entry  Per  Accou...|
| 16524  01  39306...|
|191675  01  23681...|
|191667  01  37144...|
|191673  01  26327...|
| 80495  01  27663...|
| 80507  01  46092...|
| 80509  01  10927...|
| 80497  01  33863...|
|191669  01  59058...|
|191671  01  27493...|
+--------------------+

root
 |-- value: string (nullable = true)

I can grab the header:

header = df.first()
header

which returns:

Row(value='Entry  Per  GL Account  Description               ')

and then split into distinct columns:

# Take the fixed width file and split into 3 distinct columns
sorted_df = df.select(
    df.value.substr( 1,  6).alias('Entry'      ),
    df.value.substr( 8,  3).alias('Per'        ),
    df.value.substr(12, 11).alias('GL Account' ),
    df.value.substr(24, 11).alias('Description'),
)

sorted_df.show()
sorted_df.printSchema()

which returns:

+------+---+-----------+-----------+
| Entry|Per| GL Account|Description|
+------+---+-----------+-----------+
|Entry |Per| GL Account| Descriptio|
| 16524| 01| 3930621977| TXNPUES   |
|191675| 01| 2368183100| OUNHQEX   |
|191667| 01| 3714468136| GHAKASC   |
|191673| 01| 2632703881| PAHFSAP   |
| 80495| 01| 2766389794| XDZANTV   |
| 80507| 01| 4609266335| BWWYEZL   |
| 80509| 01| 1092717420| QJYPKVO   |
| 80497| 01| 3386366766| SOQLCMU   |
|191669| 01| 5905893739| FYIWNKA   |
|191671| 01| 2749355876|   CBMJTLP |
+------+---+-----------+-----------+

Now you see that the header still appears as the first line in my dataframe here. I'm unsure of how to remove it.

.iloc is not available, and I often see this approach, but this only works on an RDD:

header = rdd.first()
rdd.filter(lambda line: line != header)

So which alternatives are available?

Answer 1

You can use either .csv , .text , .textFile for this case.

Read the file with .csv method so that spark can read the header(we don't have to filter out the header).

1.Using.csv:

.csv results df .

df=spark.read.option("header","true").csv("path")
df.show(10,False)
#+----------------------------------------------------+
#|Entry  Per  Account     Description                 |
#+----------------------------------------------------+
#| 16524  01  3930621977  TXNPUES                     |
#|191675  01  2368183100  OUNHQEX                     |
#|191667  01  3714468136  GHAKASC                     |
#|191673  01  2632703881  PAHFSAP                     |
#| 80495  01  2766389794  XDZANTV                     |
#| 80507  01  4609266335  BWWYEZL                     |
#| 80509  01  1092717420  QJYPKVO                     |
#| 80497  01  3386366766  SOQLCMU                     |
#|191669  01  5905893739  FYIWNKA                     |
#|191671  01  2749355876  CBMJTLP                     |
#+----------------------------------------------------+

2.Using.text:

.text results df .

#can't read header
df=spark.read.text("path")
#get the header
header=df.first()[0]
#filter the header out from data
df.filter(~col("value").contains(header)).show(10,False)
#+----------------------------------------------------+
#|value                                               |
#+----------------------------------------------------+
#| 16524  01  3930621977  TXNPUES                     |
#|191675  01  2368183100  OUNHQEX                     |
#|191667  01  3714468136  GHAKASC                     |
#|191673  01  2632703881  PAHFSAP                     |
#| 80495  01  2766389794  XDZANTV                     |
#| 80507  01  4609266335  BWWYEZL                     |
#| 80509  01  1092717420  QJYPKVO                     |
#| 80497  01  3386366766  SOQLCMU                     |
#|191669  01  5905893739  FYIWNKA                     |
#|191671  01  2749355876  CBMJTLP                     |
#+----------------------------------------------------+

Then use

sorted_df = df.select(
    df.value.substr( 1,  6).alias('Entry'      ),
    df.value.substr( 8,  3).alias('Per'        ),
    df.value.substr(12, 11).alias('GL Account' ),
    df.value.substr(24, 11).alias('Description'),
)

sorted_df.show()
sorted_df.printSchema()

3.Using .textFile :

.textFile results rdd .

#get header into a variable
header=spark.sparkContext.textFile("path").first()

#.textfile and filter out the header
spark.sparkContext.textFile("path").\
filter(lambda l :not str(l).startswith(header)).\
map(lambda x:x.split()).map(lambda x:(str(x[0].strip()),str(x[1].strip()),str(x[2].strip()),str(x[3].strip()))).\
toDF(["Entry","Per","Account","Description"]).\
show()
#+------+---+----------+-----------+
#| Entry|Per|   Account|Description|
#+------+---+----------+-----------+
#| 16524| 01|3930621977|    TXNPUES|
#|191675| 01|2368183100|    OUNHQEX|
#|191667| 01|3714468136|    GHAKASC|
#|191673| 01|2632703881|    PAHFSAP|
#| 80495| 01|2766389794|    XDZANTV|
#| 80507| 01|4609266335|    BWWYEZL|
#| 80509| 01|1092717420|    QJYPKVO|
#| 80497| 01|3386366766|    SOQLCMU|
#|191669| 01|5905893739|    FYIWNKA|
#|191671| 01|2749355876|    CBMJTLP|
#+------+---+----------+-----------+