spark dataframe操作集锦（提取前几行，合并，入库等）

spark dataframe派生于RDD类，但是提供了非常强大的数据操作功能。当然主要对类SQL的支持。

在实际工作中会遇到这样的情况，主要是会进行两个数据集的筛选、合并，重新入库。

首先加载数据集，然后在提取数据集的前几行过程中，才找到limit的函数。

而合并就用到union函数，重新入库，就是registerTemple注册成表，再进行写入到HIVE中。

不得不赞叹dataframe的强大。

具体示例：为了得到样本均衡的训练集，需要对两个数据集中各取相同的训练样本数目来组成，因此用到了这个功能。

scala> val fes = hiveContext.sql(sqlss)
fes: org.apache.spark.sql.DataFrame = [caller_num: string, is_sr: int, call_count: int, avg_talk_time: double, max_talk_time: int, min_talk_time: int, called_num_count: int, called_lsd: double, null_called_count: int]


scala> val fcount = fes.count()
fcount: Long = 4371029

scala> val zcfea = hiveContext.sql(sqls2)
zcfea: org.apache.spark.sql.DataFrame = [caller_num: string, is_sr: int, call_count: int, avg_talk_time: double, max_talk_time: int, min_talk_time: int, called_num_count: int, called_lsd: double, null_called_count: int]


scala> val zcount = zcfea.count()
zcount: Long = 14208117


scala> val f01 = fes.limit(25000)
f01: org.apache.spark.sql.DataFrame = [caller_num: string, is_sr: int, call_count: int, avg_talk_time: double, max_talk_time: int, min_talk_time: int, called_num_count: int, called_lsd: double, null_called_count: int]


scala> val f02 = zcfea.limit(25000)
f02: org.apache.spark.sql.DataFrame = [caller_num: string, is_sr: int, call_count: int, avg_talk_time: double, max_talk_time: int, min_talk_time: int, called_num_count: int, called_lsd: double, null_called_count: int]


scala> val ff=f01.unionAll(f02)
ff: org.apache.spark.sql.DataFrame = [caller_num: