[postgis-users] Geocode function failswhencallingnormalize_address

Fri Sep 28 21:09:21 PDT 2012

Robert,

Your normalize_address syntax is no good as that will call normalize_address
numerous times per address (because of the .*) .  
Compare with speed of:
SELECT (normalize_address(address)) As addy FROM respondents_addr WHERE
address IS NOT NULL LIMIT 1000;

There are a lot of reasons sadly mostly with teh nomralize logic and not one
we've had time to go thru.  The best bet is to loop thru each and see which
one is holding things up.

We use pgScript for that and just set the limit to 1 offset i or something
like that.

There are some fixes we've put in 2.1.0 that we haven't yet backported to
2.0.  So 2.0.1 is already kind of antiquated as it is also based on tiger
2010 rather than 2011.

You can send us the 100 off list, and time permitting we'll take a look or
maybe a quick scan we'll see something that sticks out.  Kind of strapped
for time at moment.

Mike Pease did a good job of itemizing some issues.

e.g.
http://trac.osgeo.org/postgis/ticket/1614
http://trac.osgeo.org/postgis/ticket/1669

You may want to see if you suffer from any of those.
Thanks,
Regina and Leo
http://www.postgis.us

  _____  

From: postgis-users-bounces at postgis.refractions.net
[mailto:postgis-users-bounces at postgis.refractions.net] On Behalf Of
Robert_Clift at doh.state.fl.us
Sent: Friday, September 28, 2012 4:04 PM
To: postgis-users at postgis.refractions.net
Subject: Re: [postgis-users] Geocode function
failswhencallingnormalize_address

Dear Leo and Regina (and everyone else too):

Thanks very much for your detailed reply. I'm just now to the point where
I've implemented your recommendations, but no positive outcome yet. I
simplified the addresses to address the regex expression variable error,
performed a soft upgrade to 2.0.1, and cleaned up my query (both with and
without the use of pgScript).

New PostGIS version:
"POSTGIS="2.0.1 r9979" GEOS="3.3.5-CAPI-1.7.5" PROJ="Rel. 4.8.0, 6 March
2012" GDAL="GDAL 1.9.1, released 2012/05/15" LIBXML="2.7.8" ...etc.

New Query:
SELECT a.rid, a.respondent, a.s_year, Coalesce((g.geo).rating,-1) "rating",
(g.geo).geomout "the_geom"
FROM respondents_addr a
LEFT JOIN
  (
    SELECT respondent, s_year, (geocode(address, 1)) "geo"
    FROM respondents_addr
    ORDER BY rid LIMIT 100
  ) g
ON a.respondent = g.respondent
AND a.s_year = g.s_year
WHERE a.address IS NOT NULL
ORDER BY rid LIMIT 100;

Same Error Message:
ERROR:  invalid regular expression: quantifier operand invalid
CONTEXT:  PL/pgSQL function "normalize_address" line 386 at assignment
PL/pgSQL function "geocode" line 10 at assignment

And yet the following query still returns good results in 43320 ms:
SELECT (normalize_address(address)).* FROM respondents_addr WHERE address IS
NOT NULL LIMIT 1000;

Wait. As I typed the last bit I realized the normalization test query is not
ordered the same way as the geocode query.
This fails with the exact same error:
SELECT (normalize_address(address)).* FROM respondents_addr WHERE address IS
NOT NULL ORDER BY rid LIMIT 100;

I'm not very familiar with regular expressions. Is anyone willing to look at
these first 100 addresses, sent off-list, and tell me which are tripping me
up and why?

Thanks,
Rob

  _____  

From: postgis-users-bounces at postgis.refractions.net
[mailto:postgis-users-bounces at postgis.refractions.net] On Behalf Of Paragon
Corporation
Sent: Thursday, September 06, 2012 8:14 PM
To: 'PostGIS Users Discussion'
Subject: Re: [postgis-users] Geocode function fails
whencallingnormalize_address

Rob,

Regarding your normalize issue, it's probably one record causing it.  I
thought we fixed some of these issues in 2.0.1 and 2.1.0SVN so you might
want to upgrade your script to the 2.0.1 -- there is an upgrade script you
can use packaged in the tar ball.
If that still doesn't fix your issue, look for variables often found in
regex expressions in your address (things like (, )  and . for example might
be throwing it off and we might not be escaping right.  Hard to tell which
one until your narrow down to the one bad record.
You might have to run each update as a single to see which one fails

Couple of other tips
1) you don't want to geocode 100K records all at once since it has to
complete as a single transaction.  You'll need to do it in batches.  We use
pgScript for that batch processing so we can just run in a pgAdmin window.
though any scripting tool of your choice would do.
http://www.postgresonline.com/journal/archives/181-pgAdmin-pgScript.html

2) Don't do (geocode(address)).*   That often calls geocode for each field
in output so would slow down your processing n fold (e.g. you'd have like 5
calls instead of 1)

You really want to do geocode(address) As geo

(geo).geomout etc. 

See example in docs --
http://www.postgis.org/documentation/manual-svn/Geocode.html  -- we do a
left join to guarantee that even if no records returned for an address, we
can stamp the address as attempted with a -1 rating
I think we thought about this after we wrote the book :(

-- SHAMELESS PLUG STARTS HERE ---
On bright side, we just signed our contract for Second Edition of PostGIS in
Action and have started writing it :)  more on that later.  that will cover
newer enhancements in geocoder, raster, topology, PostGIS 2.0-2.1 (basically
at least 2.0 and 2.1 changes)  and PostgreSQL 9.1-9.3 

We'll be posting the new Table Of Contents soon here, and let people know
when they can start purchasing
http://www.postgis.us

Those who buy the new book, is our understanding, will automatically get
electronic copies of the first edition, so even if you buy early, you'll get
our drafts as we write them and also have the older book to work with.

-- END SHAMELESS PLUG

3) The book was written before we introduced the limit, so again look at:

http://www.postgis.org/documentation/manual-svn/Geocode.html  (feature is
available in 2.0 as well)

Use: geocode(address,1)

for faster performance and then you don't even need the DISTINCT ON since
you'll get back at most one answer and the best one.

Hope that helps,
Leo and Regina
http://www.postgis.us

  _____  

From: postgis-users-bounces at postgis.refractions.net
[mailto:postgis-users-bounces at postgis.refractions.net] On Behalf Of
Robert_Clift at doh.state.fl.us
Sent: Thursday, September 06, 2012 1:01 PM
To: postgis-users at postgis.refractions.net
Subject: [postgis-users] Geocode function fails when
callingnormalize_address

Hi All: 

My batch geocode of 100K records is failing with error messages that refer
to an operand in the normalize_address function. I based my single-column
UPDATE on the multi-column example in "PostGIS In Action" and have tweaked
the syntax only to achieve slight variation in the error message.

The query is something like: 

UPDATE respondents_addr SET the_geom = 
   g.geomout 
   FROM (SELECT DISTINCT ON (respondent, s_year) respondent, s_year,
(geocode(address)).* 
      FROM respondents_addr As ra 
         WHERE ra.address IS NOT NULL 
      ORDER BY respondent, s_year, rating) As g 
      WHERE g.respondent = respondents_addr.respondent 
      AND g.s_year = respondents_addr.s_year; 

The most frequent error text is: 

ERROR:  invalid regular expression: quantifier operand invalid 
CONTEXT:  PL/pgSQL function "normalize_address" line 386 at assignment 
PL/pgSQL function "geocode" line 10 at assignment 
********** Error ********** 
ERROR: invalid regular expression: quantifier operand invalid 
SQL state: 2201B 
Context: PL/pgSQL function "normalize_address" line 386 at assignment 
PL/pgSQL function "geocode" line 10 at assignment 

Also the query runs for more than an hour, (xp box with 3.5 GB ram,
PosgreSQL 9.0.6, PostGIS 2.0.0 (might be the problem?)) and no matter how
the error message changes it always references the normalization function,
so I tried:

SELECT (normalize_address(address)).* FROM respondents_addr WHERE address IS
NOT NULL LIMIT 1000; 

With the LIMIT it ran quickly and returned the expected result, so I tried: 

SELECT (normalize_address(address)).* FROM respondents_addr WHERE address IS
NOT NULL LIMIT 5000; 

Bumping the LIMIT up reproduced the error from the geocode attempts. I'd say
"aha" if only I knew why. 
I checked the list archives and googled several combinations of words from
the error but alas, to no avail. I suspect that I've either repeated a
simple syntax error or that my setup is somehow off -- any help will be
appreciatedand I might as well close with results of PostGIS_full_version():

POSTGIS="2.0.0 r9605" GEOS="3.3.3-CAPI-1.7.4" PROJ="Rel. 4.8.0, 6 March
2012" GDAL="GDAL 1.9.0, released 2011/12/29" LIBXML="2.7.8"
LIBJSON="UNKNOWN" TOPOLOGY RASTER

Thank you, 
Rob 

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.osgeo.org/pipermail/postgis-users/attachments/20120929/6e8f649c/attachment.html>