Filtering DataFrame using the length of a column

Question

I want to filter a DataFrame using a condition related to the length of a column, this question might be very easy but I didn't find any related question in the SO.

More specific, I have a DataFrame with only one Column which of ArrayType(StringType()), I want to filter the DataFrame using the length as filterer, I shot a snippet below.

df = sqlContext.read.parquet("letters.parquet")
df.show()

# The output will be 
# +------------+
# |      tokens|
# +------------+
# |[L, S, Y, S]|
# |[L, V, I, S]|
# |[I, A, N, A]|
# |[I, L, S, A]|
# |[E, N, N, Y]|
# |[E, I, M, A]|
# |[O, A, N, A]|
# |   [S, U, S]|
# +------------+

# But I want only the entries with length 3 or less
fdf = df.filter(len(df.tokens) <= 3)
fdf.show() # But it says that the TypeError: object of type 'Column' has no len(), so the previous statement is obviously incorrect.

I read Column's Documentation, but didn't find any property useful for this matter. I appreciate any help!

score 88 · Accepted Answer · edited Nov 08 '17 at 22:02

In Spark >= 1.5 you can use size function:

from pyspark.sql.functions import col, size

df = sqlContext.createDataFrame([
    (["L", "S", "Y", "S"],  ),
    (["L", "V", "I", "S"],  ),
    (["I", "A", "N", "A"],  ),
    (["I", "L", "S", "A"],  ),
    (["E", "N", "N", "Y"],  ),
    (["E", "I", "M", "A"],  ),
    (["O", "A", "N", "A"],  ),
    (["S", "U", "S"],  )], 
    ("tokens", ))

df.where(size(col("tokens")) <= 3).show()

## +---------+
## |   tokens|
## +---------+
## |[S, U, S]|
## +---------+

In Spark < 1.5 an UDF should do the trick:

from pyspark.sql.types import IntegerType
from pyspark.sql.functions import udf

size_ = udf(lambda xs: len(xs), IntegerType())

df.where(size_(col("tokens")) <= 3).show()

## +---------+
## |   tokens|
## +---------+
## |[S, U, S]|
## +---------+

If you use HiveContext then size UDF with raw SQL should work with any version:

df.registerTempTable("df")
sqlContext.sql("SELECT * FROM df WHERE size(tokens) <= 3").show()

## +--------------------+
## |              tokens|
## +--------------------+
## |ArrayBuffer(S, U, S)|
## +--------------------+

For string columns you can either use an udf defined above or length function:

from pyspark.sql.functions import length

df = sqlContext.createDataFrame([("fooo", ), ("bar", )], ("k", ))
df.where(length(col("k")) <= 3).show()

## +---+
## |  k|
## +---+
## |bar|
## +---+

What about if the column is a `string` and I pretend to filter by the `string`'s length? — Alberto Bonsanto, Nov 26 '15 at 15:18

score 5 · Answer 2 · edited Oct 15 '18 at 17:41

5

Here is an example for String in scala:

val stringData = Seq(("Maheswara"), ("Mokshith"))
val df = sc.parallelize(stringData).toDF
df.where((length($"value")) <= 8).show
+--------+
|   value|
+--------+
|Mokshith|
+--------+
df.withColumn("length", length($"value")).show
+---------+------+
|    value|length|
+---------+------+
|Maheswara|     9|
| Mokshith|     8|
+---------+------+

edited Oct 15 '18 at 17:41

Jean-François Corbett

37,420
30
139
188

answered Oct 15 '18 at 17:21

mputha

395
5
7

There's a difference with my question, in it I refer to a column that is an array of strings. – Alberto Bonsanto Oct 15 '18 at 17:45

score 1 · Answer 3 · edited Oct 31 '18 at 23:37

@AlbertoBonsanto : below code filters based on array size:

val input = Seq(("a1,a2,a3,a4,a5"), ("a1,a2,a3,a4"), ("a1,a2,a3"), ("a1,a2"), ("a1"))
val df = sc.parallelize(input).toDF("tokens")
val tokensArrayDf = df.withColumn("tokens", split($"tokens", ","))
tokensArrayDf.show
+--------------------+
|              tokens|
+--------------------+
|[a1, a2, a3, a4, a5]|
|    [a1, a2, a3, a4]|
|        [a1, a2, a3]|
|            [a1, a2]|
|                [a1]|
+--------------------+

tokensArrayDf.filter(size($"tokens") > 3).show
+--------------------+
|              tokens|
+--------------------+
|[a1, a2, a3, a4, a5]|
|    [a1, a2, a3, a4]|
+--------------------+

Filtering DataFrame using the length of a column

3 Answers3

Linked

Related