Oracle分析函数参考手册三

PERCENTILE_DISC
功能描述：返回一个与输入的分布百分比值相对应的数据值，分布百分比的计算方法见函数CUME_DIST，如果没有正好对应的数据值，就取大于该分布值的下一个值。
注意：本函数与PERCENTILE_CONT的区别在找不到对应的分布值时返回的替代值的计算方法不同

SAMPLE：下例中0.7的分布值在部门30中没有对应的Cume_Dist值，所以就取下一个分布值0.83333333所对应的SALARY来替代

SELECT last_name, salary, department_id,
PERCENTILE_DISC(0.7) WITHIN GROUP (ORDER BY salary )
OVER (PARTITION BY department_id) "Percentile_Disc",
CUME_DIST() OVER (PARTITION BY department_id ORDER BY salary) "Cume_Dist"
FROM employees
WHERE department_id in (30, 60);

LAST_NAME SALARY DEPARTMENT_ID Percentile_Disc Cume_Dist
------------------------- ---------- ------------- --------------- ----------
Colmenares 2500 30 3100 .166666667
Himuro 2600 30 3100 .333333333
Tobias 2800 30 3100 .5
Baida 2900 30 3100 .666666667
Khoo 3100 30 3100 .833333333
Raphaely 11000 30 3100 1
Lorentz 4200 60 6000 .2
Austin 4800 60 6000 .6
Pataballa 4800 60 6000 .6
Ernst 6000 60 6000 .8
Hunold 9000 60 6000 1


RANK
功能描述：根据ORDER BY子句中表达式的值，从查询返回的每一行，计算它们与其它行的相对位置。组内的数据按ORDER BY子句排序，然后给每一行赋一个号，从而形成一个序列，该序列从1开始，往后累加。每次ORDER BY表达式的值发生变化时，该序列也随之增加。有同样值的行得到同样的数字序号（认为null时相等的）。然而，如果两行的确得到同样的排序，则序数将随后跳跃。若两行序数为1，则没有序数2，序列将给组中的下一行分配值3，DENSE_RANK则没有任何跳跃。
SAMPLE：下例中计算每个员工按部门分区再按薪水排序，依次出现的序列号（注意与DENSE_RANK函数的区别）

SELECT d.department_id , e.last_name, e.salary, RANK()
OVER (PARTITION BY e.department_id ORDER BY e.salary) as drank
FROM employees e, departments d
WHERE e.department_id = d.department_id
AND d.department_id IN ('60', '90');

DEPARTMENT_ID LAST_NAME SALARY DRANK
------------- ------------------------- ---------- ----------
60 Lorentz 4200 1
60 Austin 4800 2
60 Pataballa 4800 2
60 Ernst 6000 4
60 Hunold 9000 5
90 Kochhar 17000 1
90 De Haan 17000 1
90 King 24000 3


RATIO_TO_REPORT
功能描述：该函数计算expression/(sum(expression))的值，它给出相对于总数的百分比，即当前行对sum(expression)的贡献。
SAMPLE：下例计算每个员工的工资占该类员工总工资的百分比

SELECT last_name, salary, RATIO_TO_REPORT(salary) OVER () AS rr
FROM employees
WHERE job_id = 'PU_CLERK';

LAST_NAME SALARY RR
------------------------- ---------- ----------
Khoo 3100 .223021583
Baida 2900 .208633094
Tobias 2800 .201438849
Himuro 2600 .18705036
Colmenares 2500 .179856115


REGR_ (Linear Regression) Functions
功能描述：这些线性回归函数适合最小二乘法回归线，有9个不同的回归函数可使用。
REGR_SLOPE：返回斜率，等于COVAR_POP(expr1, expr2) / VAR_POP(expr2)
REGR_INTERCEPT：返回回归线的y截距，等于
AVG(expr1) - REGR_SLOPE(expr1, expr2) * AVG(expr2)
REGR_COUNT：返回用于填充回归线的非空数字对的数目
REGR_R2：返回回归线的决定系数，计算式为：
If VAR_POP(expr2) = 0 then return NULL
If VAR_POP(expr1) = 0 and VAR_POP(expr2) != 0 then return 1
If VAR_POP(expr1) > 0 and VAR_POP(expr2 != 0 then
return POWER(CORR(expr1,expr),2)
REGR_AVGX：计算回归线的自变量(expr2)的平均值，去掉了空对(expr1, expr2)后，等于AVG(expr2)
REGR_AVGY：计算回归线的应变量(expr1)的平均值，去掉了空对(expr1, expr2)后，等于AVG(expr1)
REGR_SXX： 返回值等于REGR_COUNT(expr1, expr2) * VAR_POP(expr2)
REGR_SYY： 返回值等于REGR_COUNT(expr1, expr2) * VAR_POP(expr1)
REGR_SXY: 返回值等于REGR_COUNT(expr1, expr2) * COVAR_POP(expr1, expr2)

（下面的例子都是在SH用户下完成的）
SAMPLE 1：下例计算1998年最后三个星期中两种产品（260和270）在周末的销售量中已开发票数量和总数量的累积斜率和回归线的截距

SELECT t.fiscal_month_number "Month", t.day_number_in_month "Day",
REGR_SLOPE(s.amount_sold, s.quantity_sold)
OVER (ORDER BY t.fiscal_month_desc, t.day_number_in_month) AS CUM_SLOPE,
REGR_INTERCEPT(s.amount_sold, s.quantity_sold)
OVER (ORDER BY t.fiscal_month_desc, t.day_number_in_month) AS CUM_ICPT
FROM sales s, times t
WHERE s.time_id = t.time_id
AND s.prod_id IN (270, 260)
AND t.fiscal_year=1998
AND t.fiscal_week_number IN (50, 51, 52)
AND t.day_number_in_week IN (6,7)
ORDER BY t.fiscal_month_desc, t.day_number_in_month;

Month Day CUM_SLOPE CUM_ICPT
---------- ---------- ---------- ----------
12 12 -68 1872
12 12 -68 1872
12 13 -20.244898 1254.36735
12 13 -20.244898 1254.36735
12 19 -18.826087 1287
12 20 62.4561404 125.28655
12 20 62.4561404 125.28655
12 20 62.4561404 125.28655
12 20 62.4561404 125.28655
12 26 67.2658228 58.9712313
12 26 67.2658228 58.9712313
12 27 37.5245541 284.958221
12 27 37.5245541 284.958221
12 27 37.5245541 284.958221

SAMPLE 2：下例计算1998年4月每天的累积交易数量

SELECT UNIQUE t.day_number_in_month,
REGR_COUNT(s.amount_sold, s.quantity_sold)
OVER (PARTITION BY t.fiscal_month_number ORDER BY t.day_number_in_month)
"Regr_Count"
FROM sales s, times t
WHERE s.time_id = t.time_id
AND t.fiscal_year = 1998 AND t.fiscal_month_number = 4;

DAY_NUMBER_IN_MONTH Regr_Count
------------------- ----------
1 825
2 1650
3 2475
4 3300
.
.
.
26 21450
30 22200

SAMPLE 3：下例计算1998年每月销售量中已开发票数量和总数量的累积回归线决定系数

SELECT t.fiscal_month_number,
REGR_R2(SUM(s.amount_sold), SUM(s.quantity_sold))
OVER (ORDER BY t.fiscal_month_number) "Regr_R2"
FROM sales s, times t
WHERE s.time_id = t.time_id
AND t.fiscal_year = 1998
GROUP BY t.fiscal_month_number
ORDER BY t.fiscal_month_number;

FISCAL_MONTH_NUMBER Regr_R2
------------------- ----------
1
2 1
3 .927372984
4 .807019972
5 .932745567
6 .94682861
7 .965342011
8 .955768075
9 .959542618
10 .938618575
11 .880931415
12 .882769189

SAMPLE 4：下例计算1998年12月最后两周产品260的销售量中已开发票数量和总数量的累积平均值

SELECT t.day_number_in_month,
REGR_AVGY(s.amount_sold, s.quantity_sold)
OVER (ORDER BY t.fiscal_month_desc, t.day_number_in_month)
"Regr_AvgY",
REGR_AVGX(s.amount_sold, s.quantity_sold)
OVER (ORDER BY t.fiscal_month_desc, t.day_number_in_month)
"Regr_AvgX"
FROM sales s, times t
WHERE s.time_id = t.time_id
AND s.prod_id = 260
AND t.fiscal_month_desc = '1998-12'
AND t.fiscal_week_number IN (51, 52)
ORDER BY t.day_number_in_month;

DAY_NUMBER_IN_MONTH Regr_AvgY Regr_AvgX
------------------- ---------- ----------
14 882 24.5
14 882 24.5
15 801 22.25
15 801 22.25
16 777.6 21.6
18 642.857143 17.8571429
18 642.857143 17.8571429
20 589.5 16.375
21 544 15.1111111
22 592.363636 16.4545455
22 592.363636 16.4545455
24 553.846154 15.3846154
24 553.846154 15.3846154
26 522 14.5
27 578.4 16.0666667
SAMPLE 5：下例计算产品260和270在1998年2月周末销售量中已开发票数量和总数量的累积REGR_SXY, REGR_SXX, and REGR_SYY统计值

SELECT t.day_number_in_month,
REGR_SXY(s.amount_sold, s.quantity_sold)
OVER (ORDER BY t.fiscal_year, t.fiscal_month_desc) "Regr_sxy",
REGR_SYY(s.amount_sold, s.quantity_sold)
OVER (ORDER BY t.fiscal_year, t.fiscal_month_desc) "Regr_syy",
REGR_SXX(s.amount_sold, s.quantity_sold)
OVER (ORDER BY t.fiscal_year, t.fiscal_month_desc) "Regr_sxx"
FROM sales s, times t
WHERE s.time_id = t.time_id
AND prod_id IN (270, 260)
AND t.fiscal_month_desc = '1998-02'
AND t.day_number_in_week IN (6,7)
ORDER BY t.day_number_in_month;

DAY_NUMBER_IN_MONTH Regr_sxy Regr_syy Regr_sxx
------------------- ---------- ---------- ----------
1 18870.4 2116198.4 258.4
1 18870.4 2116198.4 258.4
1 18870.4 2116198.4 258.4
1 18870.4 2116198.4 258.4
7 18870.4 2116198.4 258.4
8 18870.4 2116198.4 258.4
14 18870.4 2116198.4 258.4
15 18870.4 2116198.4 258.4
21 18870.4 2116198.4 258.4
22 18870.4 2116198.4 258.4


ROW_NUMBER
功能描述：返回有序组中一行的偏移量，从而可用于按特定标准排序的行号。
SAMPLE：下例返回每个员工再在每个部门中按员工号排序后的顺序号

SELECT department_id, last_name, employee_id, ROW_NUMBER()
OVER (PARTITION BY department_id ORDER BY employee_id) AS emp_id
FROM employees
WHERE department_id < 50;

DEPARTMENT_ID LAST_NAME EMPLOYEE_ID EMP_ID
------------- ------------------------- ----------- ----------
10 Whalen 200 1
20 Hartstein 201 1
20 Fay 202 2
30 Raphaely 114 1
30 Khoo 115 2
30 Baida 116 3
30 Tobias 117 4
30 Himuro 118 5
30 Colmenares 119 6
40 Mavris 203 1


STDDEV
功能描述：计算当前行关于组的标准偏离。（Standard Deviation）
SAMPLE：下例返回部门30按雇佣日期排序的薪水值的累积标准偏离

SELECT last_name, hire_date,salary,
STDDEV(salary) OVER (ORDER BY hire_date) "StdDev"
FROM employees
WHERE department_id = 30;

LAST_NAME HIRE_DATE SALARY StdDev
------------------------- ---------- ---------- ----------
Raphaely 07-12月-94 11000 0
Khoo 18-5月 -95 3100 5586.14357
Tobias 24-7月 -97 2800 4650.0896
Baida 24-12月-97 2900 4035.26125
Himuro 15-11月-98 2600 3649.2465
Colmenares 10-8月 -99 2500 3362.58829


STDDEV_POP
功能描述：该函数计算总体标准偏离，并返回总体变量的平方根，其返回值与VAR_POP函数的平方根相同。（Standard Deviation－Population）
SAMPLE：下例返回部门20、30、60的薪水值的总体标准偏差

SELECT department_id, last_name, salary,
STDDEV_POP(salary) OVER (PARTITION BY department_id) AS pop_std
FROM employees
WHERE department_id in (20,30,60);

DEPARTMENT_ID LAST_NAME SALARY POP_STD
------------- ------------------------- ---------- ----------
20 Hartstein 13000 3500
20 Fay 6000 3500
30 Raphaely 11000 3069.6091
30 Khoo 3100 3069.6091
30 Baida 2900 3069.6091
30 Colmenares 2500 3069.6091
30 Himuro 2600 3069.6091
30 Tobias 2800 3069.6091
60 Hunold 9000 1722.32401
60 Ernst 6000 1722.32401
60 Austin 4800 1722.32401
60 Pataballa 4800 1722.32401
60 Lorentz 4200 1722.32401


STDDEV_SAMP
功能描述： 该函数计算累积样本标准偏离，并返回总体变量的平方根，其返回值与VAR_POP函数的平方根相同。（Standard Deviation－Sample）
SAMPLE：下例返回部门20、30、60的薪水值的样本标准偏差

SELECT department_id, last_name, hire_date, salary,
STDDEV_SAMP(salary) OVER
(PARTITION BY department_id ORDER BY hire_date
ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW) AS cum_sdev
FROM employees
WHERE department_id in (20,30,60);

DEPARTMENT_ID LAST_NAME HIRE_DATE SALARY CUM_SDEV
------------- ------------------------- ---------- ---------- ----------
20 Hartstein 17-2月 -96 13000
20 Fay 17-8月 -97 6000 4949.74747
30 Raphaely 07-12月-94 11000
30 Khoo 18-5月 -95 3100 5586.14357
30 Tobias 24-7月 -97 2800 4650.0896
30 Baida 24-12月-97 2900 4035.26125
30 Himuro 15-11月-98 2600 3649.2465
30 Colmenares 10-8月 -99 2500 3362.58829
60 Hunold 03-1月 -90 9000
60 Ernst 21-5月 -91 6000 2121.32034
60 Austin 25-6月 -97 4800 2163.33077
60 Pataballa 05-2月 -98 4800 1982.42276
60 Lorentz 07-2月 -99 4200 1925.61678