Hi.

On 28.02.2015, at 14:50 , RK Aranas <[email protected]> wrote:
> I was able to identify the source of errors for these observations. Turns
> out, those were readings taken when my phone's clock was messed up. Date on
> my phone during that time was sometime in 1970. Those resulted in wacky
> computed GPS positions. What do I need to do to help you remove the
> erroneous data? Right now, it's making it rather hard to see which spots on
> the map are not yet covered in the Metro Manila area. Thanks!

Sorry, your file has been sitting in my email inbox for a good long while now. 
I haven't forgotten about it, it's just difficult to correct this.

The problem is that all we have for the map data is a table consisting of:

auto-inc id, rounded lat, rounded lon, creation date
1, 27500, 95000, "2014-08-03"
2, 27501, 95000, "2014-08-04"
...

Where the lat/lon fields in that example represent 27.500 / 95.000. We only 
have the creation date since about July 2014 and no modification date. For each 
of those coordinates we randomly generate five points into it, to fill out the 
tile a bit more. That randomness depends on a pseudo-random generator which 
relies on the ordering of the auto-inc ids, to generate the exact same 
pseudo-random sequence with each run. If it where truly random, we'd regenerate 
and change each tile with each run, breaking all caching.

And we don't have online access to any of the old observation data, just the 
data from the last day. The rest is all dumped to the filesystem (or rather 
Amazon S3) in large chunks of 100k records. We can only access that data by the 
months it was uploaded in.

Add to all this, that the map generation process doesn't currently support 
deleting map tiles. The way it's wired up, it only uploads new data to Amazon 
S3 / Cloudfront when the data in a tile has changed. But it doesn't issue a 
delete when there's no more data in a tile.

This is all excuses, but there's a good number of things here which makes 
correcting the map currently difficult and time consuming, so it always ends up 
at the end of my backlog.

Sorry, I'll try to work on the map generation in the next weeks and see if I 
can correct this. I'll have to work on the map generation to speed it up as 
well, since it is now taking close to two hours to run each night.

Hanno
_______________________________________________
dev-geolocation mailing list
[email protected]
https://lists.mozilla.org/listinfo/dev-geolocation

Reply via email to