PostgreSQL: Customers preferred product and second most preferred product - sql

Question

I'm pretty new to SQL (currently using PostgreSQL but interested in knowledge about any SQL), and am trying to figure something that I guess should be relatively straightforward.
I have a table containing one row per customer transaction, for each transaction I know what the customer bought. I am interested in finding out what product is each customers preferred choice, and then their second to most preferred choice (and in the end, on a general basis what is the preferred second choice when the preferred choice is unavailable).
Below is a mock up of what the data could look like:
+---------------------+-----------------+
| Customer_id | Product bought |
+---------------------+-----------------+
| 1 | DVD |
+- -+- -+
| 1 | DVD |
+- -+- -+
| 1 | Blu-ray |
+- -+- -+
| 1 | DVD |
+- -+- -+
| 2 | DVD |
+- -+- -+
| 2 | DVD |
The successful results would be something like this:
+---------------------+--------------------------------+
| Customer_id | Preferred #1 | Preferred #2 |
+---------------------+--------------------------------+
| 1 | DVD | Blu-ray |
+- -+- -+
| 2 | DVD | $NULL$ |
(And as mentioned earlier, the final result (most likely done in Python/R and not in SQL, would be to see a general basis as "If Preferred #1 is DVD, then Preferred #2 is Blu-ray", "If Preferred #1 is Blu-ray, then Preferred #2 is Sandwich"... and so on)
Cheers

Answer 1

This is a combination of a greatest-n-per-group and a pivot problem (sometimes also referred to as crosstab)
The first step you need to do is to identify the two preferred products.
In your case you need to combine a group by query with window functions.
The following query counts how often each customer has bought each product:
select customer_id,
product_bought,
count(*) as num_products
from sales
group by customer_id, product_bought
order by customer_id;
This can be enhanced to include a rank for the number of times a product was bought:
select customer_id,
product_bought,
count(*) as num_products,
dense_rank() over (partition by customer_id order by count(*) desc) as rnk
from sales
group by customer_id, product_bought
order by customer_id;
This would return the following result (based on your sample data):
customer_id | product_bought | num_products | rnk
------------+----------------+--------------+----
1 | DVD | 3 | 1
1 | Blu-ray | 1 | 2
2 | DVD | 2 | 1
We cannot apply a where condition on the rnk column directly, so we need a derived table for that:
select customer_id, product_bought
from (
select customer_id,
product_bought,
count(*) as num_products,
dense_rank() over (partition by customer_id order by count(*) desc) as rnk
from sales
group by customer_id, product_bought
) t
where rnk <= 2
order by customer_id;
Now we need to convert the two rows for each customer into columns. This could e.g. be done using a common table expression:
with preferred_products as (
select *
from (
select customer_id,
product_bought,
count(*) as num_products,
dense_rank() over (partition by customer_id order by count(*) desc) as rnk
from sales
group by customer_id, product_bought
) t
where rnk <= 2
)
select p1.customer_id,
p1.product_bought as "Product #1",
p2.product_bought as "Product #2"
from preferred_products p1
left join preferred_products p2 on p1.customer_id = p2.customer_id and p2.rnk = 2
where p1.rnk = 1
This then returns
customer_id | Product #1 | Product #2
------------+------------+-----------
1 | DVD | Blu-ray
2 | DVD |
The above is standard SQL and will work on any modern DBMS.
Online example: http://rextester.com/VAID15638