Re: [NOVICE] Question About Aggregate Functions

I think I mistakenly sent this to General instead of Novice. Oops.

Yeah, I either skipped over or forgot the bit in the OP about bools.
Mea culpa.

You should be able to use OR instead of AND in any logical expression.

Well this sounds more like what I want. Given t2.fielda, t2.fieldb,
t2.fieldc, any one (or all three) could be true, but frequently at least
one of the fields is false. Initially, all of the fields might be
unknown (thus NULL) for a given item until I am able to investigate the
items to determine TRUE/FALSE. I frequently have items that are
inactive, and thus unable to determine any of attributes in t2.

My end result needs to be a count of all the values in each field where
the value is TRUE, as opposed to FALSE or NULL.

Yeah, I would probably run 4 separate, simple queries. That will get
you the best performance since you're doing no JOINs and no composite
queries.

If you need to enter the results into another table, try INSERT ... to
insert the defaults and any primary key you have (like timestamp), then
four UPDATE ... SELECT statements.

The real problem with NULLs is some of the (in my mind) nonsensical
results you get, especially with logical operators:

NULL AND TRUE => NULL

NULL OR TRUE => TRUE

NULL AND FALSE => FALSE

NULL OR FALSE => NULL

Plus you have to use IS instead of = since any NULL in an = expression
makes the result NULL (yes, this is an error in my previous queries).
NULL just has all these special cases. I find it much nicer to avoid it
wherever possible since it has somewhat unpredictable results.

________________________________

From: pgsql-general-owner(at)postgresql(dot)org
[mailto:pgsql-general-owner(at)postgresql(dot)org] On Behalf Of Don Parris
Sent: Wednesday, September 13, 2006 12:50 PM
To: pgsql-general(at)postgresql(dot)org
Subject: Re: [GENERAL] [NOVICE] Question About Aggregate Functions

On 9/13/06, Brandon Aiken <BAiken(at)winemantech(dot)com> wrote:

Ah, I did not know what was in your fields, so I did not assume
they were Boolean values. It looked to me like you were trying to use
IS TRUE to substitute for the lack of a GROUP BY, so I didn't know what
to do.

That was in the first paragraph of my OP. "How do I create a query
that (1) evaluates each boolean field for TRUE/FALSE and (2) counts the
number of rows where each field is TRUE?" Maybe you just hadn't had
your first cup of coffee? ;-) Seriously, though, I really do appreciate
your help.

Yes, count() will include all non-NULL values. Sorry if I
sounded unclear there.

If you do typecasting the value zero is false (and non-zero is
true). NULL in an expression always returns NULL, and many programs
will interpret that result as false. So I'm not sure of what results
you might get with a Boolean test against a non-Boolean field,
especially if it's an integer field.

postgres=# select 0::boolean = FALSE;

?column?

----------

t

(1 row)

You should just be able to take the previous query and add in
your WHERE clauses:

SELECT count(t1.fielda), count(t2.fielda), count(t2.fieldb),
AVG(t2.fieldc)

FROM t1 JOIN t2 ON t1.item_id = t2.item_id

WHERE t1.fielda = TRUE AND t2.fielda = TRUE AND t2.fieldb = TRUE

GROUP BY NULL;

Now, the INNER JOIN you're using is only selecting fields where
both t1.item_id and t2.item_id exist and the respective fields are TRUE.
That is, it's only going to run the count and average functions against
the results of this query:

SELECT *

FROM t1 JOIN t2 ON ON t1.item_id = t2.item_id

WHERE t1.fielda = TRUE AND t2.fielda = TRUE AND t2.fieldb =
TRUE;

If that's what you want, that's great.

Can I use OR instead of AND here?

However, you might want a count of each field where that field
is TRUE. In that case, I would use either temporary tables, compound
queries and derived tables, or multiple simple queries.

It's also possible that you might want a count of fields where
t1.item_id and t2.item_id exist, but where only each respective field is
TRUE. That is, you want a count of t1.fielda where it is TRUE no matter
what t2.fielda and t2.fieldb are as long as t1.item_id matches
t2.item_id. In that case you have to do even more joins, and that could
take a fair bit of time especially if you haven't indexed your item_id
fields.

Well this sounds more like what I want. Given t2.fielda, t2.fieldb,
t2.fieldc, any one (or all three) could be true, but frequently at least
one of the fields is false. Initially, all of the fields might be
unknown (thus NULL) for a given item until I am able to investigate the
items to determine TRUE/FALSE. I frequently have items that are
inactive, and thus unable to determine any of attributes in t2.

My end result needs to be a count of all the values in each field where
the value is TRUE, as opposed to FALSE or NULL.

You really have to look at your result sets. Sometimes it is
better to run multiple simple queries instead of one big complex query
to be sure you're getting the data you want and the query executes in a
reasonable amount of time.

Also, consider that NULL values are generally considered bad to
purposefully enter. Logically, It would be better to create one table
for each field and then create a record for each item_id as you need it
so you never have NULLs. The problem with that is one of performance if
you end up doing large number of JOINs. In that case, it might be
better to use integers instead of Boolean fields, since you have three
explicit states of TRUE, FALSE, and NOT YET DETERMINED.

Regarding the NULL Values:
I have been thinking that I might want to leave my NULLs as they are.
However, I will be concentrating mostly on the items that are active.
Inactive items are only counted as part of the total number of items.
Their attributes are a moot point, but technically FALSE. (If they are
inactive, the attributes are no longer TRUE in any case.)

I am counting only those fields that I know (through verification) to be
TRUE. I can use FALSE where the actual value is not known, and just
change the attribute to TRUE when I discover that to be the case. I
just need to be sure in my counts that I note the fact that FALSE values
include the unverified values (meaning some of those might actually be
true). Does that sound fairly logical to you?

Regarding the Table Layout:
The columns in t2 are ordered, essentially according to the category of
attributes involved. I had thought about using multiple tables, one for
each category of attributes. However, I chose a monolithic table for
the attributes to represent a single survey of each item. Each item
might be surveyed again in the future to determine any changes, which
would introduce a new version of the current table.

I'll tinker around with the queries a bit, and see what I come up with.

Thanks for the input.