匹配识别将行数据收集到单列中

Question

I'm following the tutorial for match_recognize found here :

create or replace temporary table stock_price_history (company text, price_date date, price int);
insert into stock_price_history values
    ('ABCD', '2020-10-01', 50),
    ('ABCD', '2020-10-02', 50),
    ('ABCD', '2020-10-03', 51),
    ('ABCD', '2020-10-04', 51),
    ('ABCD', '2020-10-05', 51),
    ('ABCD', '2020-10-06', 52),
    ('ABCD', '2020-10-07', 71),
    ('ABCD', '2020-10-08', 80),
    ('ABCD', '2020-10-09', 90),
    ('ABCD', '2020-10-10', 63),
    ('XYZ' , '2020-10-01', 24),
    ('XYZ' , '2020-10-02', 24),
    ('XYZ' , '2020-10-03', 37),
    ('XYZ' , '2020-10-04', 63),
    ('XYZ' , '2020-10-05', 65),
    ('XYZ' , '2020-10-06', 66),
    ('XYZ' , '2020-10-07', 50),
    ('XYZ' , '2020-10-08', 54),
    ('XYZ' , '2020-10-09', 30),
    ('XYZ' , '2020-10-10', 32);
    
select * from stock_price_history
  match_recognize(
    partition by company
    order by price_date
    measures
      match_number() as match_number,
      price as all_price,
      first(price_date) as start_date,
      last(price_date) as end_date,
      count(*) as rows_in_sequence,
      count(row_with_price_stationary.*) as num_stationary,
      count(row_with_price_increase.*) as num_increases
    one row per match
    after match skip to last row_with_price_increase
    pattern(row_before_increase row_with_price_increase{1} row_with_price_stationary* row_with_price_increase{1})
    define
      row_with_price_increase as price > lag(price),
      row_with_price_stationary as price = lag(price)
  )
order by company, match_number;

The code above is my version of the tutorial code. Everything works fine except the price as all_price part in the measures clause. What I want to do is collect all prices in the pattern and return it as an array into a single column. I know I can do all rows per match to get all rows but that's not what I want.

How would I go about doing that?

Answer 1

You have to specify all rows per match or lose that information out of the match_recognize function. You can use array_agg within group to get the prices in a single array. Since this aggregates row counts down you may want to do the same for the dates of each of these prices - something like this:

select   COMPANY
        ,array_agg(PRICE) within group (order by PRICE_DATE) as ALL_PRICE
        ,array_agg(PRICE_DATE) within group (order by PRICE_DATE) as ALL_PRICE_DATE
from stock_price_history
  match_recognize(
    partition by company
    order by price_date
    measures
      match_number() as match_number,
      price as all_price,
      first(price_date) as start_date,
      last(price_date) as end_date,
      count(*) as rows_in_sequence,
      count(row_with_price_stationary.*) as num_stationary,
      count(row_with_price_increase.*) as num_increases
    all rows per match
    after match skip to last row_with_price_increase
    pattern(row_before_increase row_with_price_increase{1} row_with_price_stationary* row_with_price_increase{1})
    define
      row_with_price_increase as price > lag(price),
      row_with_price_stationary as price = lag(price)
  )
group by company
order by company
;

COMPANY	ALL_PRICE	ALL_PRICE_DATE
ABCD	[ 50, 51, 51, 51, 52, 52, 71, 80 ]	[ "2020-10-02", "2020-10-03", "2020-10-04", "2020-10-05", "2020-10-06", "2020-10-06", "2020-10-07", "2020-10-08" ]
XYZ	[ 24, 37, 63, 63, 65, 66 ]	[ "2020-10-02", "2020-10-03", "2020-10-04", "2020-10-04", "2020-10-05", "2020-10-06" ]

If you want to keep all rows, you can use the window function version of array_agg:

select   * exclude ALL_PRICE
        ,array_agg(PRICE) within group (order by PRICE_DATE) 
            over (partition by COMPANY) as ALL_PRICE
from stock_price_history
  match_recognize(
    partition by company
    order by price_date
    measures
      match_number() as match_number,
      price as all_price,
      first(price_date) as start_date,
      last(price_date) as end_date,
      count(*) as rows_in_sequence,
      count(row_with_price_stationary.*) as num_stationary,
      count(row_with_price_increase.*) as num_increases
    all rows per match
    after match skip to last row_with_price_increase
    pattern(row_before_increase row_with_price_increase{1} row_with_price_stationary* row_with_price_increase{1})
    define
      row_with_price_increase as price > lag(price),
      row_with_price_stationary as price = lag(price)
  )
order by company
;