Fluently adding values in t-sql - sql

Question

I have a table like this:
Items Date Price
1 2016-01-01 10
1 2016-01-02 15
1 2016-01-03 null
1 2016-01-04 null
1 2016-01-05 8
1 2016-01-06 null
1 2016-01-07 null
1 2016-01-08 null
2 2016-01-01 14
2 2016-01-02 7
2 2016-01-03 null
2 2016-01-04 null
2 2016-01-05 16
2 2016-01-06 null
2 2016-01-07 null
2 2016-01-08 5
Now I want to update the null values. The difference between the price before and after null values must be evenly added.
Example:
1 2016-01-02 15 to
1 2016-01-05 8
15 to 8 = -7
-7 / 3 = -2,333333
1 2016-01-02 15
1 2016-01-03 12,6666
1 2016-01-04 10,3333
1 2016-01-05 8
Shouldn't be made with cursors. Helptables would be OK.

Answer 1

This is really where you want the ignore nulls option on lag() and lead(). Alas.
An alternative is to use outer apply:
select t.*,
coalesce(t.price,
tprev.price +
datediff(day, tprev.date, t.date) * (tnext.price - tprev.price) / datediff(day, tprev.date, tnext.date)
) as est_price
from t outer apply
(select top 1 t2.*
from t t2
where t2.item = t.item and
t2.date <= t.date and
t2.price is not null
order by t2.date desc
) tprev outer apply
(select top 1 t2.*
from t t2
where t2.item = t.item and
t2.date >= t.date and
t2.price is not null
order by t2.date asc
) tnext ;
The complex arithmetic is just calculating the difference, dividing by the number of days, and then allocating the days to the current day.

Answer 2

WITH T1 AS
(
SELECT *,
ROW_NUMBER() OVER (PARTITION BY Items ORDER BY Date) AS RN,
FORMAT(ROW_NUMBER() OVER (PARTITION BY Items ORDER BY Date),'D10') + FORMAT(Price,'0000000000.000000') AS RnPr
FROM YourTable
), T2 AS
(
SELECT *,
MAX(RnPr) OVER (PARTITION BY Items ORDER BY Date ROWS BETWEEN UNBOUNDED PRECEDING AND CURRENT ROW) AS prev,
MIN(RnPr) OVER (PARTITION BY Items ORDER BY Date ROWS BETWEEN CURRENT ROW AND UNBOUNDED FOLLOWING) AS next
FROM T1
), T3 AS
(
SELECT Items,
Date,
Price,
RnPr,
InterpolatedPrice = IIF(Price IS NOT NULL,prevPrice,prevPrice + (RN - prevRN) * (nextPrice - prevPrice)/NULLIF(nextRN - prevRN,0))
FROM T2
CROSS APPLY (VALUES(CAST(SUBSTRING(prev,11,17) AS decimal(16,6)),
CAST(LEFT(prev, 10) AS INT),
CAST(SUBSTRING(next,11,17) AS decimal(16,6)),
CAST(LEFT(next, 10) AS INT)
)) V(prevPrice,prevRN,nextPrice,nextRN)
)
--UPDATE T3 SET Price = InterpolatedPrice
SELECT *
FROM T3
ORDER BY Items,
Date
Which returns
the row_number and price are bundled together in a single column (RnPr above). The order of RnPr is the same as the order by row_number. MIN and MAX both ignore NULLS. So finding the MAX(RnPr) between UNBOUNDED PRECEDING AND CURRENT ROW will include the value of the previous NOT NULL price if the price in the current row is null. And similarly MIN(RnPr) will find the next with a frame between CURRENT ROW AND UNBOUNDED FOLLOWING.
This can then be cracked apart to get the price and row number as above.
If happy with the results the final SELECT can be removed and the UPDATE uncommented as in this demo.

Answer 3

Just replace "YourTable" (4 of them) with your actual table name.
If you are happy with the results, then comment out the select and un-comment the UPDATE and WHERE.
Select A.Items,A.Date,
--Update YourTable Set
Price = IsNull(A.Price,((DateDiff(DD,B.Date,A.Date)/(DateDiff(DD,B.Date,C.Date)+0.0))*(C.Price - B.Price)) + B.Price)
From YourTable A
Outer Apply (Select Top 1 Date,Price from YourTable Where Items=A.Items and Date<A.Date and Price is not Null and A.Price is null Order by Price Desc) B
Outer Apply (Select Top 1 Date,Price from YourTable Where Items=A.Items and Date>A.Date and Price is not Null and A.Price is null Order by Price) C
--Where Price is NULL
Returns
Now, you'll notice nulls between 01/06 and 01/08 for Items 1. This is because there is no cap to interpolate with.

Answer 4

You can do this by using a series of window functions in common table expressions.
T1 assigns a row number to each row (rn)
T2 finds the first non null Price row number before and after the current row (rnb / rna)
T3 calculates the adjusted Price by looking up the prices before and after
declare #T table (Items int, Date date, Price float)
insert into #T (Items, Date, Price) values
(1, '2016-01-01', 10), (1, '2016-01-02', 15), (1, '2016-01-03', null), (1, '2016-01-04', null), (1, '2016-01-05', 8), (1, '2016-01-06', null), (1, '2016-01-07', null), (1, '2016-01-08', null), (2, '2016-01-01', 14), (2, '2016-01-02', 7), (2, '2016-01-03', null), (2, '2016-01-04', null), (2, '2016-01-05', 16), (2, '2016-01-06', null), (2, '2016-01-07', null), (2, '2016-01-08', 5)
;with T1 as
(
select *,
row_number() over (order by Items, Date) as rn
from #T
),
T2 as
(
select *,
max(case when price is null then null else rn end) over (partition by Items order by Date) as rnb,
min(case when price is null then null else rn end) over (partition by Items order by Date desc) as rna
from T1
)
select Items, Date,
isnull(price,
lag(Price, rn-rnb, Price) over (order by rn) -
(
lag(Price, rn-rnb, Price) over (order by rn) -
lead(Price, rna-rn, Price) over (order by rn)
) / (rna-rnb) * (rn-rnb)
) as Price
from T2
order by Items, Date