Some addresses are very slow to Geocode

[mikepease]

Hi, Regina. I took a break from the geocoder for a couple weeks, but I tried your latest version yesterday. Seems to be getting a lot of the bugs worked out. At first glance, it looks like it correctly geocoded 96% of a list of 5300 addresses.

In this same list of 5300, I still have a small list of addresses that take a curiously long time to run. (about 2 1/2 minutes)

The addresses seem pretty innocuous to me. Can you take a look at these and see if you can figure out why they would take so long compared to all the other addresses?

Thanks!

Original Address Seconds Rating
4051 27th Ave S Minneapolis MN 55405 168.673 1

3625 18th Ave S Minneapolis MN 55406 168.306 1
4057 10th Ave S Minneapolis MN 55406 160.565 1
2745 18th Ave S Minneapolis MN 55406 159.659 1
3332 16th Ave S Minneapolis MN 55406 161.26 1
3436 18th Ave S Minneapolis MN 55406 167.758 1
3841 21st Ave S Minneapolis MN 55406 29.177 1
3925 13th Ave S Minneapolis MN 55406 168.749 1
3820 12th Ave S Minneapolis MN 55406 163.415 1
4112 21st Ave S Minneapolis MN 55406 29.016 1
136 43rd St W Minneapolis MN 55419 23.279 5
3608 20th Ave S Minneapolis MN 55406 161.492 1
3337 18th Ave S Minneapolis MN 55406 169.434 1
3224 20th Ave S Minneapolis MN 55406 162.743 1
4300 19th Ave S Minneapolis MN 55406 165.816 1
3629 18th Ave S Minneapolis MN 55406 172.265 1
8512 141 St Ct Apple Valley MN 55124 27.35 15
5521 44th Ave S Minneapolis MN 55418 170.415 1
103 36th St W Minneapolis MN 55409 147.015 5

[mikepease]

A few more addresses that took 2-3 minutes to geocode. Amy clues why these few addresses take so long?

"2800 160th St W";"Farmington";"MN";"55024" "11532 96th Ave N";"Maple Grove";"MN";"" "3101 W 50th St";"Minneapolis";"MN";"55406"

[robe]

I took a cursory glance at one. Only issue I see is the tiger zip is different than that of the addresss. For me the samples I've run above take 23-72 seconds. So still much much longer than others, but not quite as long as yours. If I do the geocode with limit 1 it reduces down to about 20 seconds.

What do you get if you use the geocode(address,1) option?

I still haven't released my recent changes, so I'm not really working with the same build you are testing against so timings are hard to compare. I'm still fiddling with issues that the geocoder is not using the edge indexes much and that is making sthings much slower than they should be.

[robe]

Mike I think this should be resolved at r7689 for the random sampling of the above I tested with the above I'm getting under 150ms speeds with more than Minnesota data loaded on fairly crappy drives.

I still need to cleanup the second tier check but left that as is because it catches some things the first doesn't. When it falls in 2nd tier things get really slow with some cases. To prevent second tier, set teh max_results to 1 using geocode(address,1)

At some point, I might change the default from 10 to 1 since I would assume most people just want one result and best result within reasonable amount of time.

I'm going to close this out, but feel free to reopen if you still have speed issues.