Number based on condition - sql

Question

I'm trying to generate a number based on a condition.
When there is yes in column 'Stop' in the partition of a Client ordered by Start_Date, the Dense Rank has to start over. So I tried several things but it's stil not what I want.
My table with current number and expected number
+-----------+------------+------+------------+-------------+
| Client_No | Start_Date | Stop | Current_No | Expected_No |
+-----------+------------+------+------------+-------------+
| 1 | 1-1-2018 | No | 1 | 1 |
+-----------+------------+------+------------+-------------+
| 1 | 1-2-2018 | No | 2 | 2 |
+-----------+------------+------+------------+-------------+
| 1 | 1-3-2018 | No | 3 | 3 |
+-----------+------------+------+------------+-------------+
| 1 | 1-4-2018 | Yes | 1 | 1 |
+-----------+------------+------+------------+-------------+
| 1 | 1-5-2018 | No | 4 | 2 |
+-----------+------------+------+------------+-------------+
| 1 | 1-6-2018 | No | 5 | 3 |
+-----------+------------+------+------------+-------------+
| 2 | 1-2-2018 | No | 1 | 1 |
+-----------+------------+------+------------+-------------+
| 2 | 1-3-2018 | No | 2 | 2 |
+-----------+------------+------+------------+-------------+
| 2 | 1-4-2018 | Yes | 1 | 1 |
+-----------+------------+------+------------+-------------+
| 2 | 1-5-2018 | No | 3 | 2 |
+-----------+------------+------+------------+-------------+
| 2 | 1-6-2018 | Yes | 2 | 1 |
+-----------+------------+------+------------+-------------+
The query I used so far:
DENSE_RANK() OVER(PARTITION BY Client_No, Stop ORDER BY Start_Date ASC)
This seems not to be the solution because it counts onwart from the value 'no', but I don't no how to handle this in another way.

Answer 1

One way to solve such a Gaps-And-Islands puzzle is to first calculate a rank that starts with the 'Yes' stops.
Then calculate the row_number or dense_rank also over that rank.
For example:
create table test
(
Id int identity(1,1) primary key,
Client_No int,
Start_Date date,
Stop varchar(3)
)
insert into test
(Client_No, Start_Date, Stop) values
(1,'2018-01-01','No')
,(1,'2018-02-01','No')
,(1,'2018-03-01','No')
,(1,'2018-04-01','Yes')
,(1,'2018-05-01','No')
,(1,'2018-06-01','No')
,(2,'2018-02-01','No')
,(2,'2018-03-01','No')
,(2,'2018-04-01','Yes')
,(2,'2018-05-01','No')
,(2,'2018-06-01','Yes')
select *
, row_number() over (partition by Client_no, Rnk order by start_date) as rn
from
(
select *
, sum(case when Stop = 'Yes' then 1 else 0 end) over (partition by Client_No order by start_date) rnk
from test
) q
order by Client_No, start_date
GO
Id | Client_No | Start_Date | Stop | rnk | rn
-: | --------: | :------------------ | :--- | --: | :-
1 | 1 | 01/01/2018 00:00:00 | No | 0 | 1
2 | 1 | 01/02/2018 00:00:00 | No | 0 | 2
3 | 1 | 01/03/2018 00:00:00 | No | 0 | 3
4 | 1 | 01/04/2018 00:00:00 | Yes | 1 | 1
5 | 1 | 01/05/2018 00:00:00 | No | 1 | 2
6 | 1 | 01/06/2018 00:00:00 | No | 1 | 3
7 | 2 | 01/02/2018 00:00:00 | No | 0 | 1
8 | 2 | 01/03/2018 00:00:00 | No | 0 | 2
9 | 2 | 01/04/2018 00:00:00 | Yes | 1 | 1
10 | 2 | 01/05/2018 00:00:00 | No | 1 | 2
11 | 2 | 01/06/2018 00:00:00 | Yes | 2 | 1
db<>fiddle here
The difference between using this:
row_number() over (partition by Client_no, Rnk order by start_date)
versus this:
dense_rank() over (partition by Client_no, Rnk order by start_date)
is that the dense_rank would calculate the same number for the same start_date per Client_no & Rnk.

Answer 2

Below is one approach which gives you the output you want. You can see as a live/working demo here.
The steps involved are:
Create an adjusted stop value where we mark Stop as Yes for the first ever row for every customer
Create a separate table which only includes the rows where we will want to start/restart counting
For each of the rows in this new table we also add an end data, which is basically the date of the next row for every customer, or for the last row a date in the future
We join the original data table with the new table and run a sequence based on this new calculation
-- 1. Creating adjusted stop value
data_adjusted_stop as
(
select *,
case when row_number() over(partition by Client_No order by Start_Date asc) = 1 then 'Yes' else Stop end as adjusted_stop
from data
),
-- 2. Extracting the rows where we will want to (re)start the counting
data_with_cycle as
(
select Client_No,
row_number() over(partition by Client_No order by Start_Date asc) adjusted_stop_cycle,
Start_Date
from data_adjusted_stop
where adjusted_stop = 'Yes'
),
-- 3. Adding an End_Date column for each row where we will want to (re)start counting
data_with_end_date as
(
select *,
coalesce(lead(Start_Date) over (partition by Client_No order by Start_Date asc), '2021-01-01') as End_Date
from data_with_cycle
)
-- 4. Running a sequence partitioned by Client_No and the stop cycle
select data.*,
row_number() over(partition by data.Client_No, data_with_end_date.adjusted_stop_cycle order by data.Start_Date asc) as desired_output_sequence
from data
left join data_with_end_date
on data_with_end_date.Client_no = data.Client_no
where data.Start_Date >= data_with_end_date.Start_Date
and data.Start_Date < data_with_end_date.End_Date
order by 1, 2