Recognize time zone for an activity and display local time about activitylog2 HOT 4 CLOSED

The new weather data provider is DarkSky.net (#33), the JSON response for a weather query also returns the timezone (both named and as an hour offset) for the GPS coordinates -- this could be used to determine the local time and it saves us the trouble of implementing our own timezone mapping service.

from activitylog2.

Found this on Google+, but with Google+ going away, it seems prudent to save the comment here. The original link is at:

https://plus.google.com/u/0/+BradFitzpatrick/posts/XVyy1bAzkZd

Fun with rendering timezones. (with #golang , naturally)

Goal: map (lat, long) -> timezone name, with minimal memory footprint. Camlistore needs this, to know the actual time of a photo when EXIF only has local time and GPS, but not timezone (which approximately 0 cameras on the market populate). Without knowing the actual time, it's impossible to sort your photos in the web UI with respect to related foursquare checkins, tweets, etc.

I started going down the route of using S2 geometry to do this properly, but rendering a map and doing a lat-long -> pixel -> color lookup works well enough.

But keeping a 10800 x 5400 pixel image is 111 MB of memory uncompressed for our "lookup table". Or storing 9 bits per pixel (tricky) is still ~65 MB of memory. We can do better.

Most the world isn't interesting. We only need to keep memory resident for the tricky areas.

How to find the tricky areas?

• get timezone data files from http://efele.net/maps/tz/world/

• parse them with http://github.com/jonas-p/go-shp

• render the polygons with +Nigel Tao's http://godoc.org/code.google.com/p/freetype-go/freetype/raster and its MonochromePainter so there are hard-edges. the color is the RGBA CRC32 of the timezone name ("America/Los_Angeles") with the alpha set to 0xff. There are no conflicts in the 408 zones.

• cut up the world into 100x100 pixels, then 50x50, then 25x25, then 10x10, looking for regions that are all the same color (or ocean, which doesn't count). note that region is all one zone.

• Instead, store in memory only the pixel data for the tricky borders

Now, instead of 111 MB of pixel data to keep around in memory, I can keep 280 KB of pixel data and about a half MB of hashmap data for the 28722 regions of various sizes.

Zoom in on this image to see more detail.

/cc +Mathieu Lonjaret, +Brett Slatkin

Edit: now available at https://github.com/bradfitz/latlong

from activitylog2.

NOTE: the linked commits are intended for issue #7

from activitylog2.

This was implemented, see referenced commit.

from activitylog2.