Comments on: Driving Time Google Map Mash Up?

Question: Driving Time Google Map Mash Up?

John Frum — Thu, 21 May 2009 20:16:31 -0800

Does a Google Maps (or other mapping service) mash-up exist that will show me the driving time to various places in a heat map type format? So basically what TripTop does for the NYC subway system, but for driving instead?

I'm in the St. Louis area. So, what I'm envisioning is a Google Map that would show me that in 4 hours I could drive to Kansas City, Memphis, Indianapolis, and Louisville. In less than 6 hours I could drive to Madison, Cincinnati, Little Rock, and Tulsa. Something like UPS does for shipping times, but on a driving scale. This would be a great tool for planning weekend getaways.

By: McBearclaw

McBearclaw — Fri, 22 May 2009 00:19:22 -0800

Hm. I was waiting to see if some responses would come in, but none yet. Once upon a time, I did some driving time computations for a handful of counties in Iowa - from maybe two or three points (potential hospital sites) to every Census tract in those counties. The computations took hours. Now imagine doing that from every address to every other address in the US. You could cut the times down by reducing the granularity (as I did by using the centroids of Census tracts instead of addresses), but you're still looking at some brutal computation.

Google, UPS and TripTrop can pull it off. Google is a huge corporation with the computational power of God's calculator, and they only have to do one trip at a time (yeah, for thousands of people, but that's still not the same scale). No problem. UPS is a huge corporation whose business depends on delivery times; they have a relatively small number of origin points that only need to account for trips within a certain region, and even then their unit of time is days. TripTrop is not a giant business, but covers a much, much tinier area, has fixed points in the NYC subway stations, and still depends on "a couple hundred EC2 instances".

It's just too damn expensive to do on a large scale, and there isn't really a business model to go with it.

By: McBearclaw

McBearclaw — Fri, 22 May 2009 00:22:37 -0800

Actually, now that I think about it, you might be able to do this by looking only at the state and national highway systems. For time calculation purposes, you could snap towns not on those systems to the nearest highway, then add some standard penalty for the "snap distance"... you'd lose accuracy, but that might put it in the realm of computational feasibility, especially with EC2 on the loose now.

By: madcaptenor

madcaptenor — Fri, 22 May 2009 05:56:35 -0800

It seems like doing this only for towns that are, say, on an Interstate would be pretty similar to doing it for a subway system. It's getting every damn little road that makes this hard. I've played around with this (by hand with Google Maps) and these distance functions do some really counterintuitive things. But for the purposes you want it for, just looking at the big roads would do.

Also, from an algorithmic point of view, calculating N point-to-point driving times on the same map shouldn't take N times as long as calculating one of them. There has to be a faster algorithm for this than just asking some drive-time calculator over and over again, although I don't know what it is.

As for your actual question, why not just fire up Google Maps, ask for directions from St. Louis to some point X, and then drag X around and observe how the distance changes?

By: soma lkzx

soma lkzx — Fri, 22 May 2009 07:12:14 -0800

I was waiting for this to pan out, too. I swear I saw this for London, but I can't seem to track it down again. In general, though, heat maps for transit take a lot of computing power, and either the market isn't there to support the costs or no one's thought hard enough about it.

There has to be a faster algorithm for this than just asking some drive-time calculator over and over again, although I don't know what it is.

That's what I did for Triptrop and it was murder. For something continuous like car travel it would have to be even worse. I did figure out a new quick method for discrete points, though, but as it stands right now you can't apply it to cars.

By: bDiddy

bDiddy — Fri, 22 May 2009 09:28:15 -0800

I swear I saw something somewhere where the xkcd guy (randall) did this exact thing. But probably didn't intend it to be used by anyone but him. Trying to find any mention of it isn't panning out. Maybe it was a video of a talk he gave to Google employees.

By: bDiddy

bDiddy — Fri, 22 May 2009 09:34:38 -0800

http://www.youtube.com/watch?v=zJOS0sV2a24
Look around 4:30. He mentions the London transit maps discussed above.
However, I guess the point is that it was very computer intensive and he didn't get anywhere with it because gMaps stopped serving the data to him because he wasn't using the API.

Soooo. Sorry I didn't help at all.

By: John Frum

John Frum — Fri, 22 May 2009 16:36:05 -0800

Interesting, it sounds like the computing power isn't there yet to support something this intensive. Thanks for the replies.

@ Madcaptenor, your suggestion is basically what I have been doing, but I was hoping that there was a simpler option out there. Especially since as you mentioned it can be pretty counter intuitive at times.

By: soma lkzx

soma lkzx — Fri, 22 May 2009 20:37:23 -0800

I have no idea on the backstory of this, but i totally found something.

By: genghis

genghis — Sat, 23 May 2009 12:43:40 -0800

The 'London one' you're looking for is a project from the ever-clever mySociety. I'm reliably informed that the amount of preprocessing required is alarming.

That said, keep an eye on their site: I know for a fact that there's something very impressive in this area coming down the pipeline. At the very least, you can assume the code they use will be released.