Window sorting functions rank and deny for SQL data analysis_ rank、raw_ Number and lag, lead window offset function [usage sorting]

Be careful ： These functions must be in MySQL3.8 Version above can be used , Otherwise, there will be mistakes .

zero 、 Write it at the front

All the code in this article is in SQL ZOO Platform operation , Data is also available under the platform world Table and some data tables provided by other platforms , All the codes have passed the test .

One 、 Window sorting function

1、 Basics

• Standard grammar ：over (partition by Field name order by Field name asc/desc)
• over() Two clauses are optional ,partition by Specify partition by ,order by Specify sort by
• Compare ：
  rank function ： about 4,4,4,8, That is, if there is a row with parallel ranking , The ranking result is ：1,1,1,4
  dense_rank function ： about 4,4,4,8, That is, if there is a row with parallel ranking , The ranking result is ：1,1,1,2
  row_number function ： about 4,4,4,8, That is, if there is a row with parallel ranking , The ranking result is ：1,2,3,4

2、rank() over()—— Skip sort

（1） explain
For example, the value is 99, 99, 90, 89, So through this
The ranking obtained by the function is 1, 1, 3, 4
Because the front 2 Both are first , And they all occupy a position .

（2） practice ： The land area of each continent is larger than 100 ten thousand 、 The capital city is not named after A,M,C The first is the name of the continent and country in descending order of the population of each continent .（ Countries with equal populations rank the same , But each occupies a position ）

select continent,name,
rank() over(PARTITION BY continent ORDER BY population desc) RO
from world
where area>=1000000 and left(capital,1) not in ('A','M','C')
# stay SQL ZOO This sentence is sometimes needed on the practice platform ：group by continent,name
# If it can be compiled normally, it doesn't matter 

（3） Other conditions remain the same , If we only want to look at the top two countries on each continent , Now RO This naming works （ Sub query is needed ）：

select RF.continent,RF.name
from (select continent,name,
rank() over(PARTITION BY continent ORDER BY population desc) as RO
from world
where area>=1000000 and left(capital,1) not in ('A','M','C')) as RF
where RF.RO=2

3、dense_rank() over()

（1） explain ：dense. _rank(); Parallel continuous sort – For example, the value is 99, 99,90, 89,
Then the ranking obtained by this function is 1, 1, 2, 3

（2） practice ： The query number is ’S14000024’ And the election year is 2017 Parties and votes sorted by election votes in .

select party,votes,
dense_rank() over(ORDER BY votes desc) as RO
from ge
where constituency='S14000024' and yr=2017
group by party,votes
order by party

4、raw_number() over()

（1） explain ：row _number(): Continuous sort – For example, the value is 99, 99, 90, 89, That's right
The ranking obtained by this function is 1, 2, 3, 4
Select the corresponding sorting window function according to the demand for sorting value , Due to the different characteristics of values ( For example, the value does not repeat ) , These three functions can be used in general

（2） practice ： The query date is ’2020-4-25’ All names and dates of diagnosis , List their diagnosis date and death date in descending order

select name,confirmed,
row_number() over(ORDER BY confirmed desc) RC
,deaths
,row_number() over(ORDER BY deaths desc) RD
from covid
where whn='2020-4-25'
order by confirmed desc;

5、 Be careful

1、 Window functions can only be used in select Used in clauses
2、 In the window function partition by Clause can specify the partition of data , and group by To regroup, the difference is ,partition by Only partition without de duplication
3、 There is no partition by When clause , That is, no data partition , Directly divide the whole table into one area
4、 Sorting window function order by Clause is required , In window function order by Clause in partition , Sort the data rows according to the specified fields and sorting method
5、 The window function must have a name , For example, common RO,RI etc.

Two 、 Window offset function

1、lag()

lag( Field name , Offset , The default value is ) over()

select
name,
date_format(whn,'%Y-%m-%d') date,
confirmed  The cumulative number of confirmed patients by the end of the day ,
lag(confirmed,1) over(partition by name order by whn)  The cumulative number of confirmed cases as of yesterday ,
(confirmed-lag(confirmed,1) over(partition by name order by whn))  The number of newly diagnosed people every day 
from covid
where name in ('France','Germany') and month(whn)=1
group by name,whn,confirmed
order by whn;

2、lead()

lead( Field name , Offset , The default value is ) over()

Allow us to partition from the window , According to the given forward offset relative to the current line （LAG） Or post offset （LEAD）, And return the value of the corresponding row , The default offset is 1. When the specified offset is not matched, the line used is ,LAG and LEAD Default return NULL, Of course, other values can be used to replace LAG（val,1,0.00） The first 3 The first parameter is the replacement value .

3、 Comprehensive practice

select name,date_format(whn,'%Y-%m-%d') date,
confirmed - lag(confirmed,1) over(PARTITION BY name ORDER BY whn) Week_New
from covid
where name='Italy' and weekday(whn)=0
group by name,whn,confirmed
order by whn

confirmed The number of confirmed cases this Monday ,weekday(whn)=0 Select Monday （ Limit the calculation scope to Monday , Subtracting from each Monday is the addition of a week ）

The article is summarized in elder martial brother zidai's course ：https://www.bilibili.com/video/BV1ZM4y1u7uF?p=4, Some modifications and verifications have been made on the basis of this course .