Tek-Tips is the largest IT community on the Internet today!
Members share and learn making Tek-Tips Forums the best source of peer-reviewed technical information on the Internet!
-
Congratulations derfloh on being selected by the Tek-Tips community for having the most helpful posts in the forums last week. Way to Go!
sample = random
- Thread starter bkj123
- Start date
- Thread starter
- #3
Dieter, thank you for your reply and I understand what you are saying. However, I had heard from a reputable consultant that sample is simply returning the first x records from a given Amp that meet the query's criteria. Each time you run the sample query, Teradata is just selecting a different Amp, but still taking the FIRST x obs. Thus, not truly random. Any thoughts?
Thank you.
Thank you.
Maybe that consultant has a bad reputation ;-)
Just submit a
SEL PI_col, HASHAMP(HASHBUCKET(HASHROW(PI_col)))
FROM tab
SAMPLE xx
and you'll see that there are rows from every AMP.
The requested sample size is divided by the number of rows and each AMP retrieves a part of the sample.
If you use SAMPLE RANDOMIZED ALLOCATION in V2R5 the you'll get an even better sample, but you need to be a statistician to recognize that ;-)
Dieter
Running on the demo version with 2 AMPs:
SEL "AMP", count(*)
FROM
(
SEL HASHAMP(HASHBUCKET(HASHROW(cdate))) AS "AMP"
FROM sys_calendar.caldates
SAMPLE 1000
) dt
GROUP BY 1;
*** Query completed. 2 rows found. 2 columns returned.
*** Total elapsed time was 1 second.
AMP Count(*)
----------- -----------
0 500
1 500
SEL "AMP", count(*)
FROM
(
SEL HASHAMP(HASHBUCKET(HASHROW(cdate))) AS "AMP"
FROM sys_calendar.caldates
SAMPLE RANDOMIZED ALLOCATION 1000
) dt
GROUP BY 1;
*** Query completed. 2 rows found. 2 columns returned.
*** Total elapsed time was 1 second.
AMP Count(*)
----------- -----------
0 493
1 507
Just submit a
SEL PI_col, HASHAMP(HASHBUCKET(HASHROW(PI_col)))
FROM tab
SAMPLE xx
and you'll see that there are rows from every AMP.
The requested sample size is divided by the number of rows and each AMP retrieves a part of the sample.
If you use SAMPLE RANDOMIZED ALLOCATION in V2R5 the you'll get an even better sample, but you need to be a statistician to recognize that ;-)
Dieter
Running on the demo version with 2 AMPs:
SEL "AMP", count(*)
FROM
(
SEL HASHAMP(HASHBUCKET(HASHROW(cdate))) AS "AMP"
FROM sys_calendar.caldates
SAMPLE 1000
) dt
GROUP BY 1;
*** Query completed. 2 rows found. 2 columns returned.
*** Total elapsed time was 1 second.
AMP Count(*)
----------- -----------
0 500
1 500
SEL "AMP", count(*)
FROM
(
SEL HASHAMP(HASHBUCKET(HASHROW(cdate))) AS "AMP"
FROM sys_calendar.caldates
SAMPLE RANDOMIZED ALLOCATION 1000
) dt
GROUP BY 1;
*** Query completed. 2 rows found. 2 columns returned.
*** Total elapsed time was 1 second.
AMP Count(*)
----------- -----------
0 493
1 507
- Status