Skip to main content

The Difference Between ROW_NUMBER(), RANK(), and DENSE_RANK()

One of the most obvious and useful set of window functions are ranking functions where rows from your result set are ranked according to a certain scheme. There are three ranking functions:
  • ROW_NUMBER()
  • RANK()
  • DENSE_RANK()
The difference is easy to remember. For the examples, let’s assume we have this table (using PostgreSQL syntax):
1
2
3
4
5
CREATE TABLE t(v) AS
SELECT * FROM (
  VALUES('a'),('a'),('a'),('b'),
        ('c'),('c'),('d'),('e')
) t(v)
ROW_NUMBER()
… assigns unique numbers to each row within the PARTITION given theORDER BY clause. So you’d get:
1
2
SELECT v, ROW_NUMBER() OVER()
FROM t
Note that some SQL dialects (e.g. SQL Server) require an explicit ORDER BYclause in the OVER() clause:
1
2
SELECT v, ROW_NUMBER() OVER(ORDER BY v)
FROM t
The above query returns:
| V | ROW_NUMBER |
|---|------------|
| a |          1 |
| a |          2 |
| a |          3 |
| b |          4 |
| c |          5 |
| c |          6 |
| d |          7 |
| e |          8 |
RANK()
… behaves like ROW_NUMBER(), except that “equal” rows are ranked the same. If we substitute RANK() into our previous query:
1
2
SELECT v, RANK() OVER(ORDER BY v)
FROM t
… then the result we’re getting is this:
| V | RANK |
|---|------|
| a |    1 |
| a |    1 |
| a |    1 |
| b |    4 |
| c |    5 |
| c |    5 |
| d |    7 |
| e |    8 |
As you can see, much like in a sports ranking, we have gaps between the different ranks. We can avoid those gaps by using
DENSE_RANK()
Trivially, DENSE_RANK() is a rank with no gaps, i.e. it is “dense”. We can write:
1
2
SELECT v, DENSE_RANK() OVER(ORDER BY v)
FROM t
… to obtain
| V | DENSE_RANK |
|---|------------|
| a |          1 |
| a |          1 |
| a |          1 |
| b |          2 |
| c |          3 |
| c |          3 |
| d |          4 |
| e |          5 |
One interesting aspect of DENSE_RANK() is the fact that it “behaves like”ROW_NUMBER() when we add the DISTINCT keyword.
1
2
SELECT DISTINCT v, DENSE_RANK() OVER(ORDER BY v)
FROM t
… to obtain
| V | DENSE_RANK |
|---|------------|
| a |          1 |
| b |          2 |
| e |          5 |
| d |          4 |
| c |          3 |
In fact, ROW_NUMBER() prevents you from using DISTINCT, becauseROW_NUMBER() generates unique values across the partition before DISTINCTis applied:
1
2
3
SELECT DISTINCT v, ROW_NUMBER() OVER(ORDER BY v)
FROM t
ORDER BY 1, 2
DISTINCT has no effect:
| V | ROW_NUMBER |
|---|------------|
| a |          1 |
| a |          2 |
| a |          3 |
| b |          4 |
| c |          5 |
| c |          6 |
| d |          7 |
| e |          8 |

Putting it all together

A good way to understand the three ranking functions is to see them all in action side-by-side. Run this query
1
2
3
4
5
6
7
SELECT
  v,
  ROW_NUMBER() OVER(ORDER BY v),
  RANK()       OVER(ORDER BY v),
  DENSE_RANK() OVER(ORDER BY v)
FROM t
ORDER BY 1, 2
… or this one (using the SQL standard WINDOW clause, to reuse window specifications):
1
2
3
4
5
6
7
SELECT
  v,
  ROW_NUMBER() OVER(w),
  RANK()       OVER(w),
  DENSE_RANK() OVER(w)
FROM t
WINDOW w AS (ORDER BY v)
… to obtain:
| V | ROW_NUMBER | RANK | DENSE_RANK |
|---|------------|------|------------|
| a |          1 |    1 |          1 |
| a |          2 |    1 |          1 |
| a |          3 |    1 |          1 |
| b |          4 |    4 |          2 |
| c |          5 |    5 |          3 |
| c |          6 |    5 |          3 |
| d |          7 |    7 |          4 |
| e |          8 |    8 |          5 |
Note that unfortunately, the WINDOW clause is not supported in all databases.

SQL is awesome

These things can be written very easily using SQL window functions. Once you get a hang of the syntax, you won’t want to miss this killer feature in your every day SQL statements any more. Excited?

Comments

Popular posts from this blog

ODI KM Adding Order by Option

You can add Order by statement to queries by editing KM.I have edited IKM SQL Control Append to provide Order by.  1) Add an option to KM named USE_ORDER_BY, its type is Checkbox and default value is False. This option determines you want an order by statement at your query. 2)Add second option to KM named ORDER_BY, type is Text. You will get order by values to your query by this option. 3) Editing Insert New Rows detail of KM. Adding below three line code after having clause. That's it! <% if (odiRef.getOption("USE_ORDER_ BY").equals("1")) { %> ORDER BY <%=odiRef.getOption("ORDER_BY" )%> <%} %>  If USE_ORDER_BY option is not used, empty value of ORDER_BY option get error. And executions of KM appears as such below; At this execution, I checked the KM to not get errors if ORDER_BY option value is null. There is no prove of ORDER BY I'm glad.  Second execution to get  Ord...

Creating Yellow Interface in ODI

Hello everyone! In Oracle data integrator (ODI), an  interface  is an object which populates one datastore, called the  target , with data coming from one or more other datastores, known as  sources . The fields of the source datastore are linked to those in the target datastore using the concept of  Mapping . Temporary interfaces used in ODI are popularly known as  Yellow Interfaces . It is because ODI generates a yellow icon at the time of creation of a yellow interface as opposed to the blue icon of a regular interface. The advantage of using a yellow interface is to avoid the creation of  Models each time you need to use it in an interface. Since they are temporary, they are not a part of the data model and hence don’t need to be in the Model. So let’s begin and start creating our yellow interface! Pre-requisites : Oracle 10g Express Edition with *SQL Plus, Oracle Data Integrator 11g. Open *SQL Plus and create a new table  Sales ...

Synchronous and Asynchronous execution in ODI

In data warehouse designing, an important step is to deciding which step is before/after. Newly added packages and required DW data must be analyzed carefully. Synchronous addings can lengthen ETL duration. Interfaces, procedures without generated scenario cannot be executed in parallel. Only scenario executions can be parallel in ODI. Default scenario execution is synch in ODI. If you want to set a scenario to executed in parallel then you will write “-SYNC_MODE=2″ on command tab or select Synchronous / Asynchronous option Asynchronous in General tab. I have created a package as interfaces executes as; INT_JOBS parallel  INT_REGIONS synch  INT_REGIONS synch  INT_COUNTRIES synch  INT_LOCATIONS parallel  INT_EMPLOYEES parallel (Interfaces are independent.) Selecting beginning and ending times and durations from repository tables as ODI 11g operator is not calculating these values. It is obvious in ODI 10g operator. SELECT    sess_no...