Question

I am creating a C# application I will explain what I want by giving a simple example:

consider this table:

name   age     reply   choice 
------+-------+-------+-------
John   10-20   yes     apple
Kate   20-30   yes     orange
Sam    10-20   yes     apple
Peter  10-20   no      ----
Tom    20-30   no      ----
Mike   10-20   yes     orange

---

I would like to put together a predictive “age” decision tree for all the persons who have replied. and then predict the choices of the ones who didn’t reply.

The table is saved in an SQL Server 2008 database. And there is a feature in the SQL Server 2008 that allows to do that. I searched the Microsoft help website but I didn’t find any clear guide on how to use it.

How can I use it in my C# code, Anyone got a step by step guide for it?

Answer 1

This would do the trick:

-- create table
    declare @t table (name varchar(50), age varchar(50), reply varchar(3), answer varchar(50))
    insert @t (name, age, reply, answer)
    values ('John', '10-20', 'yes', 'apple'),
    ('Kate', '20-30', 'yes', 'orange'),
    ('Sam', '10-20', 'yes', 'apple'),
    ('Peter', '10-20', 'no', '----'),
    ('Tom', '20-30', 'no', '----'),
    ('Mike', '10-20', 'yes', 'orange')

-- get answer
    select  t.name, t.age, t.reply, case t.reply when 'yes' then t.answer else w.answer end answer
    from    @t t
            left join (
                select age, answer
                from (
                    select  age, answer, count(*) cnt, row_number() over (partition by age order by count(*) desc) rnk
                    from    @t
                    where   reply = 'yes' 
                    group by age, answer
                ) s
                where rnk = 1
            ) w on t.age = w.age 

Just find out what answer as provided the most for each age and then pick that answer if none was given.

When there is a tie between 2 answers, it just picks one. I think the one that comes first but the are no guarantees that it will always do that.

Be aware tho that if you have a group where the answers are A:B = 55%:45%, then all the people with no answer get answer A, so you’d change the averages of the population by doing this. Just so you know.