具有多个条件的 Sparksql 过滤（使用 where 子句进行选择）

Question

Hi I have the following issue:

numeric.registerTempTable("numeric"). 

All the values that I want to filter on are literal null strings and not N/A or Null values.

I tried these three options:

1. numeric_filtered = numeric.filter(numeric['LOW'] != 'null').filter(numeric['HIGH'] != 'null').filter(numeric['NORMAL'] != 'null')

2. numeric_filtered = numeric.filter(numeric['LOW'] != 'null' AND numeric['HIGH'] != 'null' AND numeric['NORMAL'] != 'null')

3. sqlContext.sql("SELECT * from numeric WHERE LOW != 'null' AND HIGH != 'null' AND NORMAL != 'null'")

Unfortunately, numeric_filtered is always empty. I checked and numeric has data that should be filtered based on these conditions.

Here are some sample values:

Low High Normal

3.5 5.0 null

2.0 14.0 null

null 38.0 null

null null null

1.0 null 4.0

Answer 1

Your are using logical conjunction (AND). It means that all columns have to be different than 'null' for row to be included. Lets illustrate that using filter version as an example:

numeric = sqlContext.createDataFrame([
    ('3.5,', '5.0', 'null'), ('2.0', '14.0', 'null'),  ('null', '38.0', 'null'),
    ('null', 'null', 'null'),  ('1.0', 'null', '4.0')],
    ('low', 'high', 'normal'))

numeric_filtered_1 = numeric.where(numeric['LOW'] != 'null')
numeric_filtered_1.show()

## +----+----+------+
## | low|high|normal|
## +----+----+------+
## |3.5,| 5.0|  null|
## | 2.0|14.0|  null|
## | 1.0|null|   4.0|
## +----+----+------+

numeric_filtered_2 = numeric_filtered_1.where(
    numeric_filtered_1['NORMAL'] != 'null')
numeric_filtered_2.show()

## +---+----+------+
## |low|high|normal|
## +---+----+------+
## |1.0|null|   4.0|
## +---+----+------+

numeric_filtered_3 = numeric_filtered_2.where(
    numeric_filtered_2['HIGH'] != 'null')
numeric_filtered_3.show()

## +---+----+------+
## |low|high|normal|
## +---+----+------+
## +---+----+------+

All remaining methods you've tried follow exactly the same schema. What you need here is a logical disjunction (OR).

from pyspark.sql.functions import col 

numeric_filtered = df.where(
    (col('LOW')    != 'null') | 
    (col('NORMAL') != 'null') |
    (col('HIGH')   != 'null'))
numeric_filtered.show()

## +----+----+------+
## | low|high|normal|
## +----+----+------+
## |3.5,| 5.0|  null|
## | 2.0|14.0|  null|
## |null|38.0|  null|
## | 1.0|null|   4.0|
## +----+----+------+

or with raw SQL:

numeric.registerTempTable("numeric")
sqlContext.sql("""SELECT * FROM numeric
    WHERE low != 'null' OR normal != 'null' OR high != 'null'"""
).show()

## +----+----+------+
## | low|high|normal|
## +----+----+------+
## |3.5,| 5.0|  null|
## | 2.0|14.0|  null|
## |null|38.0|  null|
## | 1.0|null|   4.0|
## +----+----+------+

See also: Pyspark: multiple conditions in when clause

Answer 2

from pyspark.sql.functions import col, countDistinct 
totalrecordcount = df.where("ColumnName is not null").select(countDistinct("ColumnName")).collect()[0][0]