Question

I’m trying to analyze some baseball stats and I’m having some trouble achieving what seems like it should be a simple task. Take a look at the following result set:

GAME_PK  REC_SEQ  BatterId  PlayNumber  EventType
287576   6        462101    1           single
287576   14       519048    2           single
287576   25       435079    3           strikeout
287576   26       435079    4           stolen_base_home
287576   28       435079    5           stolen_base_2b

The PlayNumber column is being generated by me using ROW_NUMBER() OVER (ORDER BY GAME_PK, REC_SEQ). The rest comes directly from an MLB stats database. REC_SEQ is the sequence number of the event within the game. EventType is essentially the result of an at-bat.

I would like PlayNumber to increment only when the BatterId changes. But it must respect the ordering of REC_SEQ. So I don’t think I can use RANK or DENSE_RANK, but those seem to be very close to what I need.

I would like my result set to look like this:

GAME_PK  REC_SEQ  BatterId  PlayNumber  EventType
287576   6        462101    1           single
287576   14       519048    2           single
287576   25       435079    3           strikeout
287576   26       435079    3           stolen_base_home
287576   28       435079    3           stolen_base_2b

Any help is appreciated.

Thanks!

EDIT: A batter can appear more than once during a game. He should be assigned a new PlayNumber for each appearance. Basically, each new at-bat requires a new PlayNumber.

Answer 1

Edit: It seems like the only way this can be accomplished is to figure out where each group begins and ends by determining which sequential records share a batterId. This is done by joining the records with themselves offset by 1 rownum to determine where each group begins. Once we have a collection of the starts of each group (GroupSets), we can determine to which group each individual record belongs to produce the correct numbering:

with GroupSets as (
select
    row_number() over (order by s1.rec_seq) as rownum,
    s1.game_pk, s1.rec_seq, s1.batterid, s2.batterid as nextbatterid,
    s1.eventtype
from (select *, row_number() over (order by rec_seq) as rownum from stats) s1
left join (select rec_seq, batterid,
           row_number() over (order by rec_seq) as rownum from stats) s2
    on s1.rownum = s2.rownum + 1
where s1.batterid != s2.batterid or s2.batterid is null
)
select
    game_pk,
    rec_seq,
    batterid,
    (select max(rownum) from GroupSets gs where gs.Rec_Seq <= s1.rec_seq) as PlayNumber,
    eventtype
from
    stats s1;

Demo: http://www.sqlfiddle.com/#!3/a5e68/50

---

old code that doesn’t handle interleaving:

Actually the DENSE_RANK() function should do it. However, we need to rank over the values of the MIN(REC_SEQ) per BatterId group in order to use REC_SEQ to control the order. Something like this should do it:

select
    s1.game_pk,
    s1.rec_seq,
    s1.batterID,
    dense_rank() over (order by s2.rec_seq) as PlayNumber,
    s1.EventType
from
    stats s1
join
    (select batterid, min(rec_seq) rec_seq
     from stats group by batterid) s2 on s1.batterid = s2.batterid
order by
    rec_seq

Demo: http://www.sqlfiddle.com/#!3/0682e/4