从PySpark中的RDD中的数据中查找最小和最大日期

Question

我使用Spark带Ipython,并有RDD印刷时,其中包含此格式的数据:

print rdd1.collect()

[u'2010-12-08 00:00:00', u'2010-12-18 01:20:00', u'2012-05-13 00:00:00',....]

每个数据都是a datetimestamp,我想在此找到最小值和最大值RDD.我怎样才能做到这一点？

Answer 1

例如,您可以使用aggregate函数(有关其工作原理的说明:使用RDD.aggregateByKey(),RDD.groupByKey()的等效实现是什么？)

from datetime import datetime    

rdd  = sc.parallelize([
    u'2010-12-08 00:00:00', u'2010-12-18 01:20:00', u'2012-05-13 00:00:00'])

def seq_op(acc, x):
    """ Given a tuple (min-so-far, max-so-far) and a date string
    return a tuple (min-including-current, max-including-current)
    """
    d = datetime.strptime(x, '%Y-%m-%d %H:%M:%S')
    return (min(d, acc[0]), max(d, acc[1]))

def comb_op(acc1, acc2):
    """ Given a pair of tuples (min-so-far, max-so-far)
    return a tuple (min-of-mins, max-of-maxs)
    """
    return (min(acc1[0], acc2[0]), max(acc1[1], acc2[1]))

# (initial-min <- max-date, initial-max <- min-date)
rdd.aggregate((datetime.max, datetime.min), seq_op, comb_op)

## (datetime.datetime(2010, 12, 8, 0, 0), datetime.datetime(2012, 5, 13, 0, 0))

或者DataFrames:

from pyspark.sql import Row
from pyspark.sql.functions import from_unixtime, unix_timestamp, min, max

row = Row("ts")
df = rdd.map(row).toDF()

df.withColumn("ts", unix_timestamp("ts")).agg(
    from_unixtime(min("ts")).alias("min_ts"), 
    from_unixtime(max("ts")).alias("max_ts")
).show()

## +-------------------+-------------------+
## |             min_ts|             max_ts|
## +-------------------+-------------------+
## |2010-12-08 00:00:00|2012-05-13 00:00:00|
## +-------------------+-------------------+