1.RANK,DENSE_RANK,ROW_NUMBER:
这三个函数的功能都是显示排名，不同之处在于对于相同的值的处理，看下面的例子，得到2001年每个地区和顾客的销售额排名
SELECT region_id, cust_nbr,
SUM(tot_sales) cust_sales,
RANK( ) OVER (ORDER BY SUM(tot_sales) DESC) sales_rank,
DENSE_RANK( ) OVER (ORDER BY SUM(tot_sales) DESC) sales_dense_rank,
ROW_NUMBER( ) OVER (ORDER BY SUM(tot_sales) DESC) sales_number
FROM orders
WHERE year = 2001
GROUP BY region_id, cust_nbr
ORDER BY 6;
REGION_ID CUST_NBR CUST_SALES SALES_RANK SALES_DENSE_RANK SALES_NUMBER
---------- ---------- ---------- ---------- ---------------- ------------
9 25 2232703 1 1 1
8 17 1944281 2 2 2
7 14 1929774 3 3 3
5 4 1878275 4 4 4
10 26 1808949 5 5 5
6 6 1788836 6 6 6
8 20 1412006 7 7 7
10 27 1322747 8 8 8
7 13 1310434 9 9 9
7 15 1255591 10 10 10
8 18 1253840 11 11 11
5 2 1224992 12 12 12
9 23 1224992 12 12 13
9 24 1224992 12 12 14
10 30 1216858 15 13 15
6 9 1208959 16 14 16
6 10 1196748 17 15 17
7 11 1190421 18 16 18
7 12 1182275 19 17 19
8 19 1174421 20 18 20
5 5 1169926 21 19 21
5 3 1161286 22 20 22
5 1 1151162 23 21 23
6 8 1141638 24 22 24
8 16 1068467 25 23 25
9 22 1036146 26 24 26
9 21 1020541 27 25 27
10 28 986964 28 26 28
6 7 971585 29 27 29
10 29 903383 30 28 30可以看出三个函数的不同之处在于：RANK和DENSE_RANK对于相同的值都采用相同的排名，但是前者会跳过接下来的若干名次，而后者不会。
ROW_NUMBER则不会出现相同的名次,对于相同的值的排序是任意的。
PARTITION BY：
PARTITION BY的功能就是将结果集分组。我们看看下面的例子,同样是上面的查询，但是我要得到每个地区内的排名，而不是全部地区的排名，
这时就会用到PARTITION BY。
SELECT region_id, cust_nbr, SUM(tot_sales) cust_sales,
RANK( ) OVER (PARTITION BY region_id
ORDER BY SUM(tot_sales) DESC) sales_rank,
DENSE_RANK( ) OVER (PARTITION BY region_id
ORDER BY SUM(tot_sales) DESC) sales_dense_rank,
ROW_NUMBER( ) OVER (PARTITION BY region_id
ORDER BY SUM(tot_sales) DESC) sales_number
FROM orders
WHERE year = 2001
GROUP BY region_id, cust_nbr
ORDER BY 1,6;
REGION_ID CUST_NBR CUST_SALES SALES_RANK SALES_DENSE_RANK SALES_NUMBER
---------- ---------- ---------- ---------- ---------------- ------------
5 4 1878275 1 1 1
5 2 1224992 2 2 2
5 5 1169926 3 3 3
5 3 1161286 4 4 4
5 1 1151162 5 5 5
6 6 1788836 1 1 1
6 9 1208959 2 2 2
6 10 1196748 3 3 3
6 8 1141638 4 4 4
6 7 971585 5 5 5
7 14 1929774 1 1 1
7 13 1310434 2 2 2
7 15 1255591 3 3 3
7 11 1190421 4 4 4
7 12 1182275 5 5 5
8 17 1944281 1 1 1
8 20 1412006 2 2 2
8 18 1253840 3 3 3
8 19 1174421 4 4 4
8 16 1068467 5 5 5
9 25 2232703 1 1 1
9 23 1224992 2 2 2
9 24 1224992 2 2 3
9 22 1036146 4 3 4
9 21 1020541 5 4 5
10 26 1808949 1 1 1
10 27 1322747 2 2 2
10 30 1216858 3 3 3
10 28 986964 4 4 4
10 29 903383 5 5 52.FIRST/LAST
用来返回排名第一和最后一位的记录。看下面的例子，找到销售最好和最差的地区。
SELECT
MIN(region_id)
KEEP (DENSE_RANK FIRST ORDER BY SUM(tot_sales) DESC) best_region,
MIN(region_id)
KEEP (DENSE_RANK LAST ORDER BY SUM(tot_sales) DESC) worst_region
FROM orders
WHERE year = 2001
GROUP BY region_id;
BEST_REGION WORST_REGION
----------- ------------
7 10上面的min(region_id)这句可能有些迷惑，MIN的意思是说，如果有相同销售额的地区，取比较小的一个。
3.NTILE
这个函数的功能是将结果集分为几个分区，例如，我们想知道哪些地区的销售情况是在前25%，哪些地区的销售情况是在最后的25%.看下面的例
子
SELECT region_id, cust_nbr, SUM(tot_sales) cust_sales,
NTILE(4) OVER (ORDER BY SUM(tot_sales) DESC) sales_quartile
FROM orders
WHERE year = 2001
GROUP BY region_id, cust_nbr
ORDER BY 4,3 DESC;
REGION_ID CUST_NBR CUST_SALES SALES_QUARTILE
---------- ---------- ---------- --------------
9 25 2232703 1
8 17 1944281 1
7 14 1929774 1
5 4 1878275 1
10 26 1808949 1
6 6 1788836 1
8 20 1412006 1
10 27 1322747 1
7 13 1310434 2
7 15 1255591 2
8 18 1253840 2
5 2 1224992 2
9 23 1224992 2
9 24 1224992 2
10 30 1216858 2
6 9 1208959 2
6 10 1196748 3
7 11 1190421 3
7 12 1182275 3
8 19 1174421 3
5 5 1169926 3
5 3 1161286 3
5 1 1151162 3
6 8 1141638 4
8 16 1068467 4
9 22 1036146 4
9 21 1020541 4
10 28 986964 4
6 7 971585 4
10 29 903383 4如果遇到不能均分的时候，按这样的原则分配，上面的例子中有30行，分为4组，每个组8行，这样1，2组8行，3，4组7行。
4.WIDTH_BUCKET
和NTILE函数不同的是，WIDTH_BUCKER是按照值的区间来分组的。看以下的例子：将销售额分为三组，1-1000000，1000001-2000000，2000001
-3000000。
SELECT region_id, cust_nbr,
SUM(tot_sales) cust_sales,
WIDTH_BUCKET(SUM(tot_sales), 1, 3000000, 3) sales_buckets
FROM orders
WHERE year = 2001
GROUP BY region_id, cust_nbr
ORDER BY 3;
REGION_ID CUST_NBR CUST_SALES SALES_BUCKETS
---------- ---------- ---------- -------------
10 29 903383 1
6 7 971585 1
10 28 986964 1
9 21 1020541 2
9 22 1036146 2
8 16 1068467 2
6 8 1141638 2
5 1 1151162 2
5 3 1161286 2
5 5 1169926 2
8 19 1174421 2
7 12 1182275 2
7 11 1190421 2
6 10 1196748 2
6 9 1208959 2
10 30 1216858 2
5 2 1224992 2
9 24 1224992 2
9 23 1224992 2
8 18 1253840 2
7 15 1255591 2
7 13 1310434 2
10 27 1322747 2
8 20 1412006 2
6 6 1788836 2
10 26 1808949 2
5 4 1878275 2
7 14 1929774 2
8 17 1944281 2
9 25 2232703 3如果遇到不在设定的范围内的值，会自动产生两个组，0和N+1，看下面的例子：
SELECT region_id, cust_nbr,
SUM(tot_sales) cust_sales,
WIDTH_BUCKET(SUM(tot_sales), 1000000, 2000000, 3) sales_buckets
FROM orders
WHERE year = 2001
GROUP BY region_id, cust_nbr
ORDER BY 3;
REGION_ID CUST_NBR CUST_SALES SALES_BUCKETS
---------- ---------- ---------- -------------
10 29 903383 0
6 7 971585 0
10 28 986964 0
9 21 1020541 1
9 22 1036146 1
8 16 1068467 1
6 8 1141638 1
5 1 1151162 1
5 3 1161286 1
5 5 1169926 1
8 19 1174421 1
7 12 1182275 1
7 11 1190421 1
6 10 1196748 1
6 9 1208959 1
10 30 1216858 1
5 2 1224992 1
9 24 1224992 1
9 23 1224992 1
8 18 1253840 1
7 15 1255591 1
7 13 1310434 1
10 27 1322747 1
8 20 1412006 2
6 6 1788836 3
10 26 1808949 3
5 4 1878275 3
7 14 1929774 3
8 17 1944281 3
9 25 2232703 45.CUME_DIST and PERCENT_RANK
CUME_DIST(Cumulative Distribution)：计算小于或等于当前排名的行数在总行数中的百分比。计算方法：当前排名/总行数
PERCENT_RANK:计算排名和总行数的百分比。 计算方法：当前排名-1/总行数-1
所有的排名都采用DENSE_RANK.我们以第二名为例，CUME_DIST=2/30=.066666667，PERCENT_RANK=2-1/30-1=1/29=.034482759
SELECT region_id, cust_nbr,
SUM(tot_sales) cust_sales,
CUME_DIST( ) OVER (ORDER BY SUM(tot_sales) DESC) sales_cume_dist,
PERCENT_RANK( ) OVER (ORDER BY SUM(tot_sales) DESC) sales_percent_rank
FROM orders
WHERE year = 2001
GROUP BY region_id, cust_nbr
ORDER BY 3 DESC;
REGION_ID CUST_NBR CUST_SALES SALES_CUME_DIST SALES_PERCENT_RANK
---------- ---------- ---------- --------------- ------------------
9 25 2232703 .033333333 0
8 17 1944281 .066666667 .034482759
7 14 1929774 .1 .068965517
5 4 1878275 .133333333 .103448276
10 26 1808949 .166666667 .137931034
6 6 1788836 .2 .172413793
8 20 1412006 .233333333 .206896552
10 27 1322747 .266666667 .24137931
7 13 1310434 .3 .275862069
7 15 1255591 .333333333 .310344828
8 18 1253840 .366666667 .344827586
5 2 1224992 .466666667 .379310345
9 23 1224992 .466666667 .379310345
9 24 1224992 .466666667 .379310345
10 30 1216858 .5 .482758621
6 9 1208959 .533333333 .517241379
6 10 1196748 .566666667 .551724138
7 11 1190421 .6 .586206897
7 12 1182275 .633333333 .620689655
8 19 1174421 .666666667 .655172414
5 5 1169926 .7 .689655172
5 3 1161286 .733333333 .724137931
5 1 1151162 .766666667 .75862069
6 8 1141638 .8 .793103448
8 16 1068467 .833333333 .827586207
9 22 1036146 .866666667 .862068966
9 21 1020541 .9 .896551724
10 28 986964 .933333333 .931034483
6 7 971585 .966666667 .965517241
10 29 903383 1 16.Hypothetical Functions
这个函数的作用是我们需要查看数据库集中包括我们假定的值的情况。看下面的例子：
SELECT cust_nbr, SUM(tot_sales) cust_sales,
RANK( ) OVER (ORDER BY SUM(tot_sales) DESC) rank,
DENSE_RANK( ) OVER (ORDER BY SUM(tot_sales) DESC) dense_rank,
CUME_DIST( ) OVER (ORDER BY SUM(tot_sales) DESC) cume_dist,
PERCENT_RANK( ) OVER (ORDER BY SUM(tot_sales) DESC) percent_rank
FROM orders
WHERE year = 2001
GROUP BY cust_nbr
ORDER BY 3;
CUST_NBR CUST_SALES RANK DENSE_RANK CUME_DIST PERCENT_RANK
---------- ---------- ---------- ---------- ---------- ------------
25 2232703 1 1 .033333333 0
17 1944281 2 2 .066666667 .034482759
14 1929774 3 3 .1 .068965517
4 1878275 4 4 .133333333 .103448276
26 1808949 5 5 .166666667 .137931034
6 1788836 6 6 .2 .172413793
20 1412006 7 7 .233333333 .206896552
27 1322747 8 8 .266666667 .24137931
13 1310434 9 9 .3 .275862069
15 1255591 10 10 .333333333 .310344828
18 1253840 11 11 .366666667 .344827586
2 1224992 12 12 .466666667 .379310345
23 1224992 12 12 .466666667 .379310345
24 1224992 12 12 .466666667 .379310345
30 1216858 15 13 .5 .482758621
9 1208959 16 14 .533333333 .517241379
10 1196748 17 15 .566666667 .551724138
11 1190421 18 16 .6 .586206897
12 1182275 19 17 .633333333 .620689655
19 1174421 20 18 .666666667 .655172414
5 1169926 21 19 .7 .689655172
3 1161286 22 20 .733333333 .724137931
1 1151162 23 21 .766666667 .75862069
8 1141638 24 22 .8 .793103448
16 1068467 25 23 .833333333 .827586207
22 1036146 26 24 .866666667 .862068966
21 1020541 27 25 .9 .896551724
28 986964 28 26 .933333333 .931034483
7 971585 29 27 .966666667 .965517241
29 903383 30 28 1 1下面我们需要看看假定销售额是100000的排名情况。
SELECT
RANK(1000000) WITHIN GROUP
(ORDER BY SUM(tot_sales) DESC) hyp_rank,
DENSE_RANK(1000000) WITHIN GROUP
(ORDER BY SUM(tot_sales) DESC) hyp_dense_rank,
CUME_DIST(1000000) WITHIN GROUP
(ORDER BY SUM(tot_sales) DESC) hyp_cume_dist,
PERCENT_RANK(1000000) WITHIN GROUP
(ORDER BY SUM(tot_sales) DESC) hyp_percent_rank
FROM orders
WHERE year = 2001
GROUP BY cust_nbr;
HYP_RANK HYP_DENSE_RANK HYP_CUME_DIST HYP_PERCENT_RANK
---------- -------------- ------------- ----------------
28 26 .903225806 .9WITHIN GROUP的意思说：决定这个假定值的排名时，是把这个值插入到实际的数据集中。如果没有的话，就是把假定值的排名和实际值的排名
进行比较。
117
被折叠的 条评论
为什么被折叠?
到【灌水乐园】发言
