On 2/19/07, Oleg Bartunov <oleg@sai.msu.su> wrote:
> pg_trgm was developed for spelling corrrection and there is a threshold of
> similarity, which is 0.3 by default. Readme explains what does it means.
Yes, I read it.
> Similarity could be very low, since you didn't make separate column and length
> of the full string is used to normalize similarity.
Yep, that's probably my problem. Ignored records are a bit longer than
the others.
I tried the tip in README.pg_trgm to generate a table with all the words.
It can do the work in conjunction of tsearch2 and a bit of AJAX to
suggest the full words to the users. The reason why I was not using
tsearch2 is that it's sometimes hard to spell location names
correctly.
The only problem is that it is still quite slow on a 50k rows words
table but I'll make further tests on a decent server this afternoon.
--
Guillaume