2011.12.29 Weekly Check In

• 13:04 -!- demory [demory@gateway/openplans.org/session] has joined #opentripplanner
• 13:17 -!- FrankP [1815504e@gateway/web/freenode/ip.24.21.80.78] has joined #opentripplanner
• 13:20 <novalis_dt> Big crowd for a holiday-ish week!
• 13:22 -!- mele [d819d1d2@gateway/web/freenode/ip.216.25.209.210] has joined #opentripplanner
• 13:28 -!- grant_h [d819d1ba@gateway/web/freenode/ip.216.25.209.186] has joined #opentripplanner
• 13:28 <FrankP> managers are all on vacation...most productive week of the year :-)
• 13:30 <novalis_dt> Well, mine seem to still be around :)
• 13:32 <demory> we'll we're definitely having a productive week in NYC -- nice and quiet!
• 13:32 <demory> shall we get started?
• 13:32 <novalis_dt> Sure!
• 13:32 <andrewbyrd> Hi, sure.
• 13:32 <demory> OK. My update: still focused on the automated deployment. We are actively setting up the system on AWS now -- as of today we have everything from user registration to gtfs validation and graph building working. Next steps are better instance management (automated starting/stopping of AWS instances as needed) and automating the actual deployment of the built graph.
• 13:33 <demory> I think we are still on track to have something to show at TranspoCamp DC, which was the goal
• 13:35 <novalis_dt> I whipped up some OSM street segmentation code which I sent off to Frank last night. It turned out to be a bit more work than expected to get it efficient, but it seems to be working now!
• 13:35 <mele> yay!
• 13:35 <demory> great
• 13:35 <FrankP> good show
• 13:36 <kpw> FrankP: have you had a chance to take a look yet?
• 13:36 <FrankP> not yet. Mele?
• 13:37 <FrankP> Later today / tomorrow...finishing something else up.
• 13:37 <mele> oh sorry i was just looking at it actually
• 13:37 <novalis_dt> I'll be out tomorrow, back Tues. But I might check email
• 13:37 <mele> looks great
• 13:37 <kpw> ok, just keep me in the loop if you have questions
• 13:37 <mele> excited to try it out
• 13:38 <mele> ok
• 13:38 <andrewbyrd> Analytics update: I've got a basic REST service that calculates travel time rasters at 50-meter resolution. The server ships the results back as an 8-bit indexed png, with pixel values in minutes, and a color map that emphasizes the 30/60/90/120 minute marks.
• 13:39 <andrewbyrd> Rather than kludging some way to add georeferencing metadata to the response, I've been considering implementing one of the OGC web service specs.
• 13:39 <kpw> cool! does ogc give you want you need?
• 13:39 <FrankP> Cool...and your fix Andrew for corner names looks good.
• 13:39 <mele> yeah the corner names fix is a great improvement
• 13:40 <andrewbyrd> FrankP : yeah, I think we should snap to corners over a somewhat larger radius
• 13:40 <andrewbyrd> It should be straightforward to provide zoom levels and tiling (though not really necessary - a regionwide PNG for Portland is only about 250k), but looking through the specs, OTP analytics data looks more like a job for OGC WCS rather than WMS.
• 13:40 <novalis_dt> Snapping: For start/end-of-trip only, or also for bus stop locations?
• 13:40 <andrewbyrd> It makes sense to me to express/expose travel time data as a 4D map/coverage, with destination (lon, lat) as the first two dimensions and origin (lon, lat) as sample dimensions for slicing.
• 13:41 <andrewbyrd> novalis_dt: i just meant for start/end
• 13:41 <novalis_dt> sounds good
• 13:41 <andrewbyrd> and maybe just up it to 10-15 instead of 5 meters
• 13:43 <novalis_dt> andrewbyrd, any luck on that slow trip?
• 13:44 -!- Peeeeeejay [d819d414@gateway/web/freenode/ip.216.25.212.20] has joined #opentripplanner
• 13:44 <FrankP> andrewbyrd, why WCS over WMS (WCS to me means raster / aerial data ... and multiple aerial coverages data sets combined into a single layer)?
• 13:45 <andrewbyrd> FrankP: just because it's travel time data rather than imagery, the standard seems more geared toward raw data.
• 13:45 <andrewbyrd> But images do just fine for sending time values.
• 13:46 <FrankP> sounds good...thanks.
• 13:47 <kpw> yeah, ultimately folks will be integrating the coverage data into other tools/data sets for downstream analysis
• 13:47 <andrewbyrd> Originally I thought WCS would be more appropriate for parameterizing the output (since we need to specify origin point) but it turns out WMS supports sample dimensions and slicing as well.
• 13:48 <kpw> also, did everything work as expected with the analytics internals, using STP->triangles?
• 13:48 <kpw> er, SPT
• 13:49 <andrewbyrd> I actually did not use triangulation because once I found all the necessary classes in geotools I discovered they were unimplemented. (I think - please correct me if anyone has discovered otherwise). Anyway the interpolation method is not exactly what we want...
• 13:50 <andrewbyrd> So I just set up a precomputed 50-meter raster of OTP vertices. Cross that with a search result and you get an image. You could up the base resolution and decimate the precomputed grid for different zoom levels.
• 13:51 <novalis_dt> Triangulation has theoretical issues anyway because x,y measure different things than z.
• 13:52 <andrewbyrd> novalis_dt: Often triangulation in GIS applications is 2.5D.
• 13:52 <andrewbyrd> you compute the triangulation in the plane then extrude
• 13:52 <novalis_dt> Yeah, that's more plausible
• 13:53 <andrewbyrd> But more importantly having all your sample points precomputed makes generating the image almost instantaneous.
• 13:54 <kpw> interesting, so at the 50-meter level we're not having to do any interpolation/.
• 13:54 <kpw> ?
• 13:54 <andrewbyrd> A triangulation would have to hit the spatial index for every pixel and would not react in real time.
• 13:55 <andrewbyrd> kpw: well, it means the data are not completely smooth when zoomed in very close
• 13:55 <andrewbyrd> but we are probably already exceeding the true precision of the information
• 13:56 <kpw> right, but what happens at larger scales? how do we interpolate then?
• 13:56 <andrewbyrd> kpw: just downsample
• 13:56 <andrewbyrd> and/or maintain several different grids
• 13:57 <kpw> ah, so just using standard raster downsampling methods gives us accurate enough results?
• 13:58 <andrewbyrd> well, you'd be off by less than 50 meters, which at our standard walk speed is about the same as the travel time quantization
• 13:58 <kpw> ok
• 14:00 <andrewbyrd> but if you want to go for nice smooth surfaces I'd be interested in messing with it.
• 14:01 <kpw> i'm mostly curious if this closes any doors for downstream analysis due to assumptions about the raster or sampling
• 14:02 <andrewbyrd> I don't believe so - I think most applications could make use of a 50 meter grid, but more importantly
• 14:03 <andrewbyrd> given an arbitrary set of sample points (e.g. name your resolution) you can find a new set of representative vertices in a few seconds.
• 14:04 <kpw> ok, this makes sense.
• 14:04 <demory> yeah i think this will serve us well for the initial applications we've discussed
• 14:04 <demory> thanks for the update
• 14:05 <andrewbyrd> I'm now trying to determine whether and how GeoServer supports dynamically generating/slicing multidimensional coverages. I know there were plans to do at least the latter, as they are the reference WCS implementation.
• 14:05 <andrewbyrd> This might be a place to cooperate with OpenGeo. Besides producing WCS responses from OTP, maybe OpenLayers support for basic coverage data visualization (beyond just shoehorning them into WMS and Image layers).
• 14:07 <andrewbyrd> novalis_dt: about the slow trip, you must mean Wilsonville-Sandy bike+transit. I haven't addressed it, I guess my dependency graph goes: class hierarchy refactor -> faster bidi heuristic -> improve multiobjective
• 14:08 <novalis_dt> What do you intend for the faster bidi?
• 14:08 <andrewbyrd> But I will try to experiment with it this week.
• 14:08 <andrewbyrd> novalis_dt: leave out most of the edges in the graph without significantly changing the result
• 14:09 <novalis_dt> Oh.
• 14:09 <novalis_dt> That's not going to fix this one, but yeah, it's probably not a bad idea.
• 14:09 <andrewbyrd> but maybe that is more long term, this needs some attention now i guess
• 14:10 -!- novalis_dt [~[email protected]] has left #opentripplanner["Leaving"]
• 14:10 -!- novalis_dt [~[email protected]] has joined #opentripplanner
• 14:10 <novalis_dt> (oops)
• 14:10 <novalis_dt> I'm going to work on #580 today, I think
• 14:11 <novalis_dt> And maybe take a look at #578.
• 14:11 <novalis_dt> FrankP, have you ever seen anything like #578?
• 14:12 <FrankP> html says it expects utf-8, and all langs render for me.
• 14:13 <FrankP> I had to convert a few .js lang files (a year ago) to utf-8 encoding ... they all should be utf-8 now.
• 14:13 <novalis_dt> Wait, what says it expects utf-8?
• 14:13 <FrankP> that said, only english comes out of OTP for me, so...
• 14:13 <FrankP> a meta tag
• 14:13 <novalis_dt> Oh. But that won't affect the JS
• 14:14 <FrankP> the .js files are encoded as utf-8
• 14:14 <novalis_dt> Yeah, but they're served with no charset
• 14:14 <novalis_dt> Which means that, according to the HTTP spec, they should be read as ISO8859-1
• 14:15 <FrankP> ok...is that a tomcat setting to make all .js files serve up as utf-8?
• 14:15 <novalis_dt> Hm, possibly.
• 14:15 <novalis_dt> I'll research it
• 14:15 <novalis_dt> I just thought you might know off the top of your head
• 14:16 <novalis_dt> OK, anyone have anything else?
• 14:17 <FrankP> No idea. It all seems to work for me, so I'd never questioned it. The Marathi characters didn't work till I converted the file to utf-8, but I think that was more a problem with saving the original edits.
• 14:17 <novalis_dt> I haven't tried the Marathi in Firefox yet -- I'm just going on a bug report.
• 14:18 <FrankP> For me, all langs work when testing locally (with Eclipse / Aptana serving up text files), including the minimized file and individual .js files
• 14:19 <FrankP> (And I would assume the same when served with tomcat, but I haven't tested that)
• 14:19 <novalis_dt> In Firefox specifically?
• 14:19 <novalis_dt> If so, then maybe I'll not worry about it.
• 14:20 <FrankP> Yes, FF (on the mac & windows)...as well as IE 9 and Chrome
• 14:21 <FrankP> I'll give it another test in a few, and chat back....again, I'm testing with Eclipse/Aptana's internal Jetty serving these files locally
• 14:21 <novalis_dt> I'm using Eclipse's internal tomcat.
• 14:21 <novalis_dt> Out of curiosity, if you look at the HTTP headers, what does Content-type say for Marathi.js?
• 14:28 <FrankP> Content-Type application/x-javascript
• 14:28 <FrankP> And Marathi renders fine
• 14:30 <novalis_dt> Oh, hm. I think I was getting it served differently. I'll look into it. Thanks.
• 14:30 <demory> anything else, anyone?
• 14:31 <mele> I don't think there's anything from us
• 14:31 <novalis_dt> Great!
• 14:31 <andrewbyrd> Nothing here.
• 14:31 <FrankP> take care, all
• 14:31 <demory> ok. see you here in a week
• 14:31 <demory> and happy new year all
• 14:32 <mele> yes, happy new year!