SQL: Running total count of distinct values - sql

Question

I'm trying to obtain rolling number of unique values in a window.
Here's how my table looks like:
SELECT
user_id
, order_date
, product
FROM example_table
WHERE user_id = 1
ORDER BY order_date ASC
user_id
order_date
product
1
2021-01-01
A
1
2021-01-01
B
1
2021-01-04
A
1
2021-01-07
C
1
2021-01-09
C
1
2021-01-20
A
Here's what I'm trying to achieve:
user_id
order_date
product
cum_dist_count
1
2021-01-01
A
1
1
2021-01-02
B
2
1
2021-01-04
A
2
1
2021-01-07
C
3
1
2021-01-09
C
3
1
2021-01-20
A
3
In other words, I want to be able to see how many unique items a customer has bough so far, and be able to see that for particular date (so for the example above: on 2021-01-04 they have bought 2 unique items and for 2021-01-07 that number was 3).
I've tried grouping by selecting user_id and product, and min(order_date) in a CTE, then doing ROW_NUMBER over user_id and product in that CTE and that worked partially - I'm able to seethe dates the countof unique products has changed (so for this example: 2021-01-01, 2021-01-02 and 2021-01-07, but then I loose the rows "between" which I still want to be able to access.
with cte as (
SELECT
user_id
, product
, min(order_date) as first_order
FROM example_table
GROUP BY 1,2
ORDER BY order_date ASC
)
SELECT
user_id
, first_order
, product
, ROW_NUMBER() OVER (PARTITION BY user_id, product ORDER BY first_order) AS number_of_unique_products
WHERE user_id = 1
With the above, I would get:
user_id
order_date
product
cum_dist_count
1
2021-01-01
A
1
1
2021-01-02
B
2
1
2021-01-07
C
3
The DB is in BigQuery StandardSQL.
Any help is much appreciated!

Answer 1

For each item, you can record the earliest date it appears. Then add those up:
select et.* except (seqnum),
countif(seqnum = 1) over (partition by user_id order by order_date) as running_distinct_count
from (select et.*,
row_number() over (partition by user_id, product order by order_date) as seqnum
from example_table et
) et

Answer 2

Below is for BigQuery
select * except(cum_products),
(select count(distinct product) from t.cum_products product) as cum_dist_count
from (
select *,
array_agg(product) over prev_rows as cum_products
from example_table
window prev_rows as (partition by user_id order by order_date)
) t
if applied to sample data in your question
with example_table as (
select 1 user_id, '2021-01-01' order_date, 'A' product union all
select 1, '2021-01-02', 'B' union all
select 1, '2021-01-04', 'A' union all
select 1, '2021-01-07', 'C' union all
select 1, '2021-01-09', 'C' union all
select 1, '2021-01-20', 'A'
)
output is