Re: [OSM-dev] Straw man design for OSM postcode database creator
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Gervase Markham wrote: | Robert (Jamie) Munro wrote: | As no one else has replied: | | Thank you :-) | | I think this is a good idea, but if I am using random addresses from my | address book, I'm not sure that I will be able to know if the road | called high street that is shown to me on the slippy map is the high | street for which I have a postcode. | | Possibly not; but presumably they are people you know at least vaguely. | And you can use your intelligence; most towns, at least don't have more | than one street of any name. If you aren't sure, you can use some of the | hints you suggest (which the interface should give you) or you can just | pass on that one. | | The namefinder code can probably | estimate automatically just as well as I can in many cases based on it's | ~ search results xxx found 100m away from yyy. If the distance is more | than 5 miles, it's probably wrong. The other factor it could take into | account is the neighbouring postcodes. If there are no other postcodes | nearby with the same prefix, the confidence level in the result is low. | | Very good idea. It should present five closest postcodes and their | distances. | | On the other hand, a user searching for a postcode on the home page is | possibly more likely to know if the result is correct, because they will | see (or fail to see) whatever it is they are looking for. Perhaps if the | namefinder postcode system could return a level of confidence, and a | yes/no button, we could get a lot of postcodes out of it. | | I'm not sure I understand this. How would it work, in detail? If I just | search for a postcode, say N12 5BQ, and the postcode finder knows where | N12 5?? is but no better, and takes me there, and the person clicks Yes, | what has been gained? I'm talking about the name finder postcode results, not the NPEMap / FreeThePostcode results. I don't think we should show the approximate NPEMap / FreeThePostcode matches ahead of the NameFinder results. We should only put NPE / FreeThePostcode matches first if they are exact matches, or we don't have anything useful from the namefinder (all the distances are greater than 10km or something). Can we set up the home page so that if NPEMap / FreeThePostcode finds an exact match, we just zoom straight to it, without asking Name Finder, (and probably without displaying a sidebar). If it doesn't find an exact match we list the namefinder results first, with the partial match listed last. Also, can we make the zoom levels of the approximate matches much lower than for exact or name finder matches. Robert (Jamie) Munro -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.6 (Darwin) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFHqKMUz+aYVHdncI0RAghKAKDtjp3Pr+jPql9lYiUFemG+Wi6RGQCfU0q8 ty3bTKBorGtLqSM56TYgDak= =zWsr -END PGP SIGNATURE- ___ dev mailing list dev@openstreetmap.org http://lists.openstreetmap.org/cgi-bin/mailman/listinfo/dev
Re: [OSM-dev] Straw man design for OSM postcode database creator
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Tom Hughes wrote: | In message [EMAIL PROTECTED] | Robert (Jamie) Munro [EMAIL PROTECTED] wrote: | | Can we set up the home page so that if NPEMap / FreeThePostcode finds an | exact match, we just zoom straight to it, without asking Name Finder, | (and probably without displaying a sidebar). If it doesn't find an exact | match we list the namefinder results first, with the partial match | listed last. Also, can we make the zoom levels of the approximate | matches much lower than for exact or name finder matches. | | As you yourself have just demonstrated in your first post, the | name finder results really aren't reliable enough for that. If | anything I'm considering removing them again. I think it's ok for the results to be bad as long as we indicate that they might be bad. The main purpose of the maps on OSM's home page is for people to see if their area has been mapped yet, and encourage people to The majority of postcode searches are good, or are obviously bad where the area hasn't been mapped yet. The one in my other post was subtle and rare - there was a building in one street that had a similar name to a nearby street. Robert (Jamie) Munro -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.6 (Darwin) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFHqPvzz+aYVHdncI0RAqZZAKCvg3ZxzFjeuXu+eeOZag9H5DT8lgCgnN7Y EGP8nFg+0NdJUeEkGOCREwE= =UMpZ -END PGP SIGNATURE- ___ dev mailing list dev@openstreetmap.org http://lists.openstreetmap.org/cgi-bin/mailman/listinfo/dev
Re: [OSM-dev] Straw man design for OSM postcode database creator
On 06/02/2008 00:14, Robert (Jamie) Munro wrote: -BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Tom Hughes wrote: | In message [EMAIL PROTECTED] | Robert (Jamie) Munro [EMAIL PROTECTED] wrote: | | Can we set up the home page so that if NPEMap / FreeThePostcode finds an | exact match, we just zoom straight to it, without asking Name Finder, | (and probably without displaying a sidebar). If it doesn't find an exact | match we list the namefinder results first, with the partial match | listed last. Also, can we make the zoom levels of the approximate | matches much lower than for exact or name finder matches. | | As you yourself have just demonstrated in your first post, the | name finder results really aren't reliable enough for that. If | anything I'm considering removing them again. I think it's ok for the results to be bad as long as we indicate that they might be bad. The main purpose of the maps on OSM's home page is for people to see if their area has been mapped yet, and encourage people to The majority of postcode searches are good, or are obviously bad where the area hasn't been mapped yet. The one in my other post was subtle and rare - there was a building in one street that had a similar name to a nearby street. I think it is possible to improve the pattern matching for addresses as well. Because this is by definition UK specific, I could make more use of clues: words like 'Road' and 'Street' (I already do this to some extent), try searching for different components of something that looks like and address, try several hits rather than working on the first likely looking one and so on. This was very much a first attempt to see if it might work - and it was much more successful than I had expected. I think I ought to patent it! David ___ dev mailing list dev@openstreetmap.org http://lists.openstreetmap.org/cgi-bin/mailman/listinfo/dev
Re: [OSM-dev] Straw man design for OSM postcode database creator
On 3 Feb 2008, at 09:11, Gervase Markham wrote: Robert (Jamie) Munro wrote: As no one else has replied: Thank you :-) I think this is a good idea, but if I am using random addresses from my address book, I'm not sure that I will be able to know if the road called high street that is shown to me on the slippy map is the high street for which I have a postcode. Possibly not; but presumably they are people you know at least vaguely. And you can use your intelligence; most towns, at least don't have more than one street of any name. If you aren't sure, you can use some of the hints you suggest (which the interface should give you) or you can just pass on that one. I haven't been paying attention to this so I'm not trying to pass judgement here, the claim about most towns, at least don't have more than one street of any name piqued my interest though. I just checked, there's 15 Church Roads in Liverpool, 4 High Streets and 5-7 Station Roads. Just thought you might be interested in those stats (found using the Multimap geocoder, unfortunately I haven't got anything like all of those mapped). John ___ dev mailing list dev@openstreetmap.org http://lists.openstreetmap.org/cgi-bin/mailman/listinfo/dev
Re: [OSM-dev] Straw man design for OSM postcode database creator
-BEGIN PGP SIGNED MESSAGE- Hash: SHA1 Gervase Markham wrote: | Here are a few thoughts about how one would build a web page allowing | people to take addresses and quickly mark postcode locations. | | (The other half of the project might be something which finds addresses, | perhaps from Google. But even this half could be used with e.g. your | Palm's address book exported as CSV and munged to remove extraneous data.) | | The interface would be a web page containing: | - textbox for pasting in a list of addresses and postcodes, one per | line | - Namefinder box, which shows a list of the results for each address | - slippy map (hardcoded to zoom=17?) which shows the currently-selected | result | - Buttons: That's Right and Beats Me | | The procedure would be as follows: | | - Paste a load of addresses into the box | - Code immediately goes away and gets all the necessary Namefinder |and slippy map data in the background (to work around slowness of some |APIs) | - First Namefinder result comes back | - Slippy map displays first hit | - If it's not right, click on other results to find the right one | - If it is right, click on That's Right | - Postcode is submitted with the lat/long returned by Namefinder | - If you can't find it, click on Beats Me | - Page automatically moves on to next address | - There is a Back button in case you mis-submit | | Note that you don't actually click on the map to mark the postcode. | Instead, it uses the lat/long returned from Namefinder, which is | (ideally) in the middle of the road in question. | | Perhaps we might want to allow map clicking if the person also has local | knowledge... | | Comments? As no one else has replied: I think this is a good idea, but if I am using random addresses from my address book, I'm not sure that I will be able to know if the road called high street that is shown to me on the slippy map is the high street for which I have a postcode. The namefinder code can probably estimate automatically just as well as I can in many cases based on it's ~ search results xxx found 100m away from yyy. If the distance is more than 5 miles, it's probably wrong. The other factor it could take into account is the neighbouring postcodes. If there are no other postcodes nearby with the same prefix, the confidence level in the result is low. On the other hand, a user searching for a postcode on the home page is possibly more likely to know if the result is correct, because they will see (or fail to see) whatever it is they are looking for. Perhaps if the namefinder postcode system could return a level of confidence, and a yes/no button, we could get a lot of postcodes out of it. Do we know how many people search for postcodes on the home page? Robert (Jamie) Munro -BEGIN PGP SIGNATURE- Version: GnuPG v1.4.6 (Darwin) Comment: Using GnuPG with Mozilla - http://enigmail.mozdev.org iD8DBQFHpQyPz+aYVHdncI0RAqxKAKC8/OdBEyNIUyrCo/gSHzEpdpp9GQCguGVz UYM0UtCJiTMULrMjnl+vaes= =c/TV -END PGP SIGNATURE- ___ dev mailing list dev@openstreetmap.org http://lists.openstreetmap.org/cgi-bin/mailman/listinfo/dev