Оценить размер срезов nans

Я добавил комментарии в код, чтобы объяснить каждый шаг, пока не достигну желаемого результата.

Конечно, нет необходимости создавать все столбцы из приведенного ниже примера, и, вероятно, этот код можно значительно улучшить, но я думаю, что важно показать вам шаг за шагом и сделать начальный старт, чтобы решить ваш вопрос.

x = sc.parallelize([
    [0, None],
    [1, 3.],
    [2, 7.],
    [3, None],
    [4, 4.],
    [5, 3.],
    [6, None],
    [7, None],
    [8, 5.],
    [9, 2.],
    [10, None]
])
# Assigned values ​​in columns A and B to facilitate manipulation
x = x.toDF(['A', 'B'])

# Prints initial DF
x.show()

Выход:

+---+----+
|  A|   B|
+---+----+
|  0|null|
|  1| 3.0|
|  2| 7.0|
|  3|null|
|  4| 4.0|
|  5| 3.0|
|  6|null|
|  7|null|
|  8| 5.0|
|  9| 2.0|
| 10|null|
+---+----+

# Transform null values into "1"
x = x.withColumn('C', when(x.B.isNull(), 1))
x.show()

Выход:

+---+----+----+
|  A|   B|   C|
+---+----+----+
|  0|null|   1|
|  1| 3.0|null|
|  2| 7.0|null|
|  3|null|   1|
|  4| 4.0|null|
|  5| 3.0|null|
|  6|null|   1|
|  7|null|   1|
|  8| 5.0|null|
|  9| 2.0|null|
| 10|null|   1|
+---+----+----+

# Creates a spec that order column A
order_spec = Window().orderBy('A')

# Doing a cumulative sum. See the explanation
# https://stackoverflow.com/questions/56384625/pyspark-cumulative-sum-with-reset-condition
x = x \
    .withColumn('tmp', sum((x.C.isNull()).cast('int')).over(order_spec)) \
    .withColumn('D', sum(x.C).over(order_spec.partitionBy("tmp"))) \
    .drop('tmp')
x.show()

Выход:

+---+----+----+----+
|  A|   B|   C|   D|
+---+----+----+----+
|  0|null|   1|   1|
|  1| 3.0|null|null|
|  2| 7.0|null|null|
|  3|null|   1|   1|
|  4| 4.0|null|null|
|  5| 3.0|null|null|
|  6|null|   1|   1|
|  7|null|   1|   2|
|  8|null|   1|   3|
|  9| 5.0|null|null|
| 10| 2.0|null|null|
| 11|null|   1|   1|
+---+----+----+----+

# Put values from column D to one row above and select the desired output values
x = x.withColumn('E', lag(x.D, ).over(order_spec)) \
    .select(x.A, x.B, when(col('E').isNotNull(), col('E')).otherwise(0).alias('nan_count'))
x.show()

Выход:

+---+----+---------+
|  A|   B|nan_count|
+---+----+---------+
|  0|null|        0|
|  1| 3.0|        1|
|  2| 7.0|        0|
|  3|null|        0|
|  4| 4.0|        1|
|  5| 3.0|        0|
|  6|null|        0|
|  7|null|        1|
|  8|null|        2|
|  9| 5.0|        3|
| 10| 2.0|        0|
| 11|null|        0|
+---+----+---------+

Весь код:

from pyspark.shell import sc
from pyspark.sql import Window
from pyspark.sql.functions import lag, when, sum, col

x = sc.parallelize([
    [0, None], [1, 3.], [2, 7.], [3, None], [4, 4.],
    [5, 3.], [6, None], [7, None], [8, None], [9, 5.], [10, 2.], [11, None]])
x = x.toDF(['A', 'B'])

# Transform null values into "1"
x = x.withColumn('C', when(x.B.isNull(), 1))

# Creates a spec that order column A
order_spec = Window().orderBy('A')

# Doing a cumulative sum with reset condition. See the explanation
# https://stackoverflow.com/questions/56384625/pyspark-cumulative-sum-with-reset-condition
x = x \
    .withColumn('tmp', sum((x.C.isNull()).cast('int')).over(order_spec)) \
    .withColumn('D', sum(x.C).over(order_spec.partitionBy("tmp"))) \
    .drop('tmp')

# Put values from column D to one row above and select the desired output values
x = x.withColumn('E', lag(x.D, ).over(order_spec)) \
    .select(x.A, x.B, when(col('E').isNotNull(), col('E')).otherwise(0).alias('nan_count'))
x.show()