hive窗口函數/分析函數
在sql中有一類函數叫做聚合函數,例如sum()、avg()、max()等等,這類函數可以將多行數據按照規則聚集爲一行,一般來講聚集後的行數是要少於聚集前的行數的。但是有時我們想要既顯示聚集前的數據,又要顯示聚集後的數據,這時我們便引入了窗口函數。窗口函數又叫OLAP函數/分析函數，窗口函數兼具分組和排序功能。
窗口函數最重要的關鍵字是 partition by 和 order by。
具體語法如下：over (partition by xxx order by xxx)
sum,avg,min,max 函數
準備數據
1建表語句:
2create table bigdata_t1(
3cookieid string,
4createtime string, --day
5pv int
6) row format delimited
7fields terminated by ',';
8
9加載數據：
10load data local inpath '/root/hivedata/bigdata_t1.dat' into table bigdata_t1;
11
12cookie1,2018-04-10,1
13cookie1,2018-04-11,5
14cookie1,2018-04-12,7
15cookie1,2018-04-13,3
16cookie1,2018-04-14,2
17cookie1,2018-04-15,4
18cookie1,2018-04-16,4
19
20開啓智能本地模式
21SET hive.exec.mode.local.auto=true;
SUM函數和窗口函數的配合使用：結果和ORDER BY相關,默認爲升序。
1#pv1
2select cookieid,createtime,pv,
3sum(pv) over(partition by cookieid order by createtime) as pv1
4from bigdata_t1;
5
6#pv2
7select cookieid,createtime,pv,
8sum(pv) over(partition by cookieid order by createtime rows between unbounded preceding and current row) as pv2
9from bigdata_t1;
10
11#pv3
12select cookieid,createtime,pv,
13sum(pv) over(partition by cookieid) as pv3
14from bigdata_t1;
15
16#pv4
17select cookieid,createtime,pv,
18sum(pv) over(partition by cookieid order by createtime rows between 3 preceding and current row) as pv4
19from bigdata_t1;
20
21#pv5
22select cookieid,createtime,pv,
23sum(pv) over(partition by cookieid order by createtime rows between 3 preceding and 1 following) as pv5
24from bigdata_t1;
25
26#pv6
27select cookieid,createtime,pv,
28sum(pv) over(partition by cookieid order by createtime rows between current row and unbounded following) as pv6
29from bigdata_t1;
30
31
32pv1: 分組內從起點到當前行的pv累積，如，11號的pv1=10號的pv+11號的pv, 12號=10號+11號+12號
33pv2: 同pv1
34pv3: 分組內(cookie1)所有的pv累加
35pv4: 分組內當前行+往前3行，如，11號=10號+11號， 12號=10號+11號+12號，
36 13號=10號+11號+12號+13號， 14號=11號+12號+13號+14號
37pv5: 分組內當前行+往前3行+往後1行，如，14號=11號+12號+13號+14號+15號=5+7+3+2+4=21
38pv6: 分組內當前行+往後所有行，如，13號=13號+14號+15號+16號=3+2+4+4=13，
39 14號=14號+15號+16號=2+4+4=10
如果不指定rows between,默認爲從起點到當前行;
如果不指定order by，則將分組內所有值累加;
關鍵是理解rows between含義,也叫做window子句：
preceding：往前
following：往後
current row：當前行
unbounded：起點
unbounded preceding 表示從前面的起點
unbounded following：表示到後面的終點
AVG，MIN，MAX，和SUM用法一樣。
row_number,rank,dense_rank,ntile 函數
準備數據
1cookie1,2018-04-10,1
2cookie1,2018-04-11,5
3cookie1,2018-04-12,7
4cookie1,2018-04-13,3
5cookie1,2018-04-14,2
6cookie1,2018-04-15,4
7cookie1,2018-04-16,4
8cookie2,2018-04-10,2
9cookie2,2018-04-11,3
10cookie2,2018-04-12,5
11cookie2,2018-04-13,6
12cookie2,2018-04-14,3
13cookie2,2018-04-15,9
14cookie2,2018-04-16,7
15
16CREATE TABLE bigdata_t2 (
17cookieid string,
18createtime string, --day
19pv INT
20) ROW FORMAT DELIMITED
21FIELDS TERMINATED BY ','
22stored as textfile;
23
24加載數據：
25load data local inpath '/root/hivedata/bigdata_t2.dat' into table bigdata_t2;
ROW_NUMBER()使用
ROW_NUMBER()從1開始，按照順序，生成分組內記錄的序列。
1SELECT
2cookieid,
3createtime,
4pv,
5ROW_NUMBER() OVER(PARTITION BY cookieid ORDER BY pv desc) AS rn
6FROM bigdata_t2;
RANK 和 DENSE_RANK使用
RANK() 生成數據項在分組中的排名，排名相等會在名次中留下空位 。
DENSE_RANK()生成數據項在分組中的排名，排名相等會在名次中不會留下空位。
1SELECT
2cookieid,
3createtime,
4pv,
5RANK() OVER(PARTITION BY cookieid ORDER BY pv desc) AS rn1,
6DENSE_RANK() OVER(PARTITION BY cookieid ORDER BY pv desc) AS rn2,
7ROW_NUMBER() OVER(PARTITION BY cookieid ORDER BY pv DESC) AS rn3
8FROM bigdata_t2
9WHERE cookieid = 'cookie1';
NTILE
有時會有這樣的需求:如果數據排序後分爲三部分，業務人員只關心其中的一部分，如何將這中間的三分之一數據拿出來呢?NTILE函數即可以滿足。
ntile可以看成是：把有序的數據集合平均分配到指定的數量（num）個桶中, 將桶號分配給每一行。如果不能平均分配，則優先分配較小編號的桶，並且各個桶中能放的行數最多相差1。
然後可以根據桶號，選取前或後 n分之幾的數據。數據會完整展示出來，只是給相應的數據打標籤；具體要取幾分之幾的數據，需要再嵌套一層根據標籤取出。
1SELECT
2cookieid,
3createtime,
4pv,
5NTILE(2) OVER(PARTITION BY cookieid ORDER BY createtime) AS rn1,
6NTILE(3) OVER(PARTITION BY cookieid ORDER BY createtime) AS rn2,
7NTILE(4) OVER(ORDER BY createtime) AS rn3
8FROM bigdata_t2
9ORDER BY cookieid,createtime;
其他一些窗口函數
lag,lead,first_value,last_value 函數
- LAG
LAG(col,n,DEFAULT) 用於統計窗口內往上第n行值第一個參數爲列名，第二個參數爲往上第n行（可選，默認爲1），第三個參數爲默認值（當往上第n行爲NULL時候，取默認值，如不指定，則爲NULL）
1 SELECT cookieid,
2 createtime,
3 url,
4 ROW_NUMBER() OVER(PARTITION BY cookieid ORDER BY createtime) AS rn,
5 LAG(createtime,1,'1970-01-01 00:00:00') OVER(PARTITION BY cookieid ORDER BY createtime) AS last_1_time,
6 LAG(createtime,2) OVER(PARTITION BY cookieid ORDER BY createtime) AS last_2_time
7 FROM bigdata_t4;
8
9
10 last_1_time: 指定了往上第1行的值，default爲'1970-01-01 00:00:00'
11 cookie1第一行，往上1行爲NULL,因此取默認值 1970-01-01 00:00:00
12 cookie1第三行，往上1行值爲第二行值，2015-04-10 10:00:02
13 cookie1第六行，往上1行值爲第五行值，2015-04-10 10:50:01
14 last_2_time: 指定了往上第2行的值，爲指定默認值
15 cookie1第一行，往上2行爲NULL
16 cookie1第二行，往上2行爲NULL
17 cookie1第四行，往上2行爲第二行值，2015-04-10 10:00:02
18 cookie1第七行，往上2行爲第五行值，2015-04-10 10:50:01
LEAD
與LAG相反
LEAD(col,n,DEFAULT) 用於統計窗口內往下第n行值
第一個參數爲列名，第二個參數爲往下第n行（可選，默認爲1），第三個參數爲默認值（當往下第n行爲NULL時候，取默認值，如不指定，則爲NULL）
1 SELECT cookieid,
2 createtime,
3 url,
4 ROW_NUMBER() OVER(PARTITION BY cookieid ORDER BY createtime) AS rn,
5 LEAD(createtime,1,'1970-01-01 00:00:00') OVER(PARTITION BY cookieid ORDER BY createtime) AS next_1_time,
6 LEAD(createtime,2) OVER(PARTITION BY cookieid ORDER BY createtime) AS next_2_time
7 FROM bigdata_t4;
FIRST_VALUE
取分組內排序後，截止到當前行，第一個值
1 SELECT cookieid,
2 createtime,
3 url,
4 ROW_NUMBER() OVER(PARTITION BY cookieid ORDER BY createtime) AS rn,
5 FIRST_VALUE(url) OVER(PARTITION BY cookieid ORDER BY createtime) AS first1
6 FROM bigdata_t4;
LAST_VALUE
取分組內排序後，截止到當前行，最後一個值
1 SELECT cookieid,
2 createtime,
3 url,
4 ROW_NUMBER() OVER(PARTITION BY cookieid ORDER BY createtime) AS rn,
5 LAST_VALUE(url) OVER(PARTITION BY cookieid ORDER BY createtime) AS last1
6 FROM bigdata_t4;
如果想要取分組內排序後最後一個值，則需要變通一下：
1 SELECT cookieid,
2 createtime,
3 url,
4 ROW_NUMBER() OVER(PARTITION BY cookieid ORDER BY createtime) AS rn,
5 LAST_VALUE(url) OVER(PARTITION BY cookieid ORDER BY createtime) AS last1,
6 FIRST_VALUE(url) OVER(PARTITION BY cookieid ORDER BY createtime DESC) AS last2
7 FROM bigdata_t4
8 ORDER BY cookieid,createtime;
特別注意order by
如果不指定ORDER BY，則進行排序混亂，會出現錯誤的結果
1 SELECT cookieid,
2 createtime,
3 url,
4 FIRST_VALUE(url) OVER(PARTITION BY cookieid) AS first2
5 FROM bigdata_t4;
cume_dist,percent_rank 函數
這兩個序列分析函數不是很常用，注意： 序列函數不支持WINDOW子句
- 數據準備
1 d1,user1,1000
2 d1,user2,2000
3 d1,user3,3000
4 d2,user4,4000
5 d2,user5,5000
6
7 CREATE EXTERNAL TABLE bigdata_t3 (
8 dept STRING,
9 userid string,
10 sal INT
11 ) ROW FORMAT DELIMITED
12 FIELDS TERMINATED BY ','
13 stored as textfile;
14
15 加載數據：
16 load data local inpath '/root/hivedata/bigdata_t3.dat' into table bigdata_t3;
CUME_DIST 和order by的排序順序有關係
CUME_DIST 小於等於當前值的行數/分組內總行數 order 默認順序 正序 升序
比如，統計小於等於當前薪水的人數，所佔總人數的比例
1 SELECT
2 dept,
3 userid,
4 sal,
5 CUME_DIST() OVER(ORDER BY sal) AS rn1,
6 CUME_DIST() OVER(PARTITION BY dept ORDER BY sal) AS rn2
7 FROM bigdata_t3;
8
9 rn1: 沒有partition,所有數據均爲1組，總行數爲5，
10 第一行：小於等於1000的行數爲1，因此，1/5=0.2
11 第三行：小於等於3000的行數爲3，因此，3/5=0.6
12 rn2: 按照部門分組，dpet=d1的行數爲3,
13 第二行：小於等於2000的行數爲2，因此，2/3=0.6666666666666666
PERCENT_RANK
PERCENT_RANK 分組內當前行的RANK值-1/分組內總行數-1
1 SELECT
2 dept,
3 userid,
4 sal,
5 PERCENT_RANK() OVER(ORDER BY sal) AS rn1, --分組內
6 RANK() OVER(ORDER BY sal) AS rn11, --分組內RANK值
7 SUM(1) OVER(PARTITION BY NULL) AS rn12, --分組內總行數
8 PERCENT_RANK() OVER(PARTITION BY dept ORDER BY sal) AS rn2
9 FROM bigdata_t3;
10
11 rn1: rn1 = (rn11-1) / (rn12-1)
12 第一行,(1-1)/(5-1)=0/4=0
13 第二行,(2-1)/(5-1)=1/4=0.25
14 第四行,(4-1)/(5-1)=3/4=0.75
15 rn2: 按照dept分組，
16 dept=d1的總行數爲3
17 第一行，(1-1)/(3-1)=0
18 第三行，(3-1)/(3-1)=1
grouping sets,grouping__id,cube,rollup 函數
這幾個分析函數通常用於OLAP中，不能累加，而且需要根據不同維度上鑽和下鑽的指標統計，比如，分小時、天、月的UV數。
- 數據準備
1 2018-03,2018-03-10,cookie1
2 2018-03,2018-03-10,cookie5
3 2018-03,2018-03-12,cookie7
4 2018-04,2018-04-12,cookie3
5 2018-04,2018-04-13,cookie2
6 2018-04,2018-04-13,cookie4
7 2018-04,2018-04-16,cookie4
8 2018-03,2018-03-10,cookie2
9 2018-03,2018-03-10,cookie3
10 2018-04,2018-04-12,cookie5
11 2018-04,2018-04-13,cookie6
12 2018-04,2018-04-15,cookie3
13 2018-04,2018-04-15,cookie2
14 2018-04,2018-04-16,cookie1
15
16 CREATE TABLE bigdata_t5 (
17 month STRING,
18 day STRING,
19 cookieid STRING
20 ) ROW FORMAT DELIMITED
21 FIELDS TERMINATED BY ','
22 stored as textfile;
23
24 加載數據：
25 load data local inpath '/root/hivedata/bigdata_t5.dat' into table bigdata_t5;
GROUPING SETS
grouping sets是一種將多個group by 邏輯寫在一個sql語句中的便利寫法。
等價於將不同維度的GROUP BY結果集進行UNION ALL。
GROUPING__ID，表示結果屬於哪一個分組集合。
1 SELECT
2 month,
3 day,
4 COUNT(DISTINCT cookieid) AS uv,
5 GROUPING__ID
6 FROM bigdata_t5
7 GROUP BY month,day
8 GROUPING SETS (month,day)
9 ORDER BY GROUPING__ID;
10
11 grouping_id表示這一組結果屬於哪個分組集合，
12 根據grouping sets中的分組條件month，day，1是代表month，2是代表day
13
14 等價於
15 SELECT month,NULL,COUNT(DISTINCT cookieid) AS uv,1 AS GROUPING__ID FROM bigdata_t5 GROUP BY month UNION ALL
16 SELECT NULL as month,day,COUNT(DISTINCT cookieid) AS uv,2 AS GROUPING__ID FROM bigdata_t5 GROUP BY day;
再如：
1 SELECT
2 month,
3 day,
4 COUNT(DISTINCT cookieid) AS uv,
5 GROUPING__ID
6 FROM bigdata_t5
7 GROUP BY month,day
8 GROUPING SETS (month,day,(month,day))
9 ORDER BY GROUPING__ID;
10
11 等價於
12 SELECT month,NULL,COUNT(DISTINCT cookieid) AS uv,1 AS GROUPING__ID FROM bigdata_t5 GROUP BY month
13 UNION ALL
14 SELECT NULL,day,COUNT(DISTINCT cookieid) AS uv,2 AS GROUPING__ID FROM bigdata_t5 GROUP BY day
15 UNION ALL
16 SELECT month,day,COUNT(DISTINCT cookieid) AS uv,3 AS GROUPING__ID FROM bigdata_t5 GROUP BY month,day;
CUBE
根據GROUP BY的維度的所有組合進行聚合。
1 SELECT
2 month,
3 day,
4 COUNT(DISTINCT cookieid) AS uv,
5 GROUPING__ID
6 FROM bigdata_t5
7 GROUP BY month,day
8 WITH CUBE
9 ORDER BY GROUPING__ID;
10
11 等價於
12 SELECT NULL,NULL,COUNT(DISTINCT cookieid) AS uv,0 AS GROUPING__ID FROM bigdata_t5
13 UNION ALL
14 SELECT month,NULL,COUNT(DISTINCT cookieid) AS uv,1 AS GROUPING__ID FROM bigdata_t5 GROUP BY month
15 UNION ALL
16 SELECT NULL,day,COUNT(DISTINCT cookieid) AS uv,2 AS GROUPING__ID FROM bigdata_t5 GROUP BY day
17 UNION ALL
18 SELECT month,day,COUNT(DISTINCT cookieid) AS uv,3 AS GROUPING__ID FROM bigdata_t5 GROUP BY month,day;
ROLLUP
是CUBE的子集，以最左側的維度爲主，從該維度進行層級聚合。
1 比如，以month維度進行層級聚合：
2 SELECT
3 month,
4 day,
5 COUNT(DISTINCT cookieid) AS uv,
6 GROUPING__ID
7 FROM bigdata_t5
8 GROUP BY month,day
9 WITH ROLLUP
10 ORDER BY GROUPING__ID;
11
12 --把month和day調換順序，則以day維度進行層級聚合：
13
14 SELECT
15 day,
16 month,
17 COUNT(DISTINCT cookieid) AS uv,
18 GROUPING__ID
19 FROM bigdata_t5
20 GROUP BY day,month
21 WITH ROLLUP
22 ORDER BY GROUPING__ID;
23 （這裏，根據天和月進行聚合，和根據天聚合結果一樣，因爲有父子關係，如果是其他維度組合的話，就會不一樣）