"What if it Doesn't Belong in OpenStreetMap?" by Kristofor Carle Live captioning by Norma Miller. @whitecoatcapxg ... ... Good afternoon, everyone. My name is Kris Carle, we're the lead developers of Maphubs and I'm going to talk to you about this question of what if it doesn't belong in OpenStreetMap. And back of my slides here, we have an example of one of our maps that's deforestation data in Borneo. So what I want to cover, going to go back to this analogy that we've been hearing about OSM that's Wikipedia for maps, because I think it comes into play when we talk about the data. Give some examples of what doesn't belong, introduce Moabi and Maphubs and what we do, talk about OSM software as a toolset and I'll give a demo of Maphubs and talk a bit about what's next for us. >> So quickly about us, Leo Bottrill founder and CEO. He's the founder of Moabi, and myself I'm the cofounder and lead developer. I like to brag that I have over a million edits in OpenStreetMap so we can compare if anyone has more, but I have over a million. I like to brag about that. Yes, on a tablet like that's the number of clicks, basically. Nodes. >> So what if it doesn't belong? So unlike this photo where I'm pretty sure cookie monster is going to figure this out, because he's like an expert in cookies, so he's going to figure it out. It's not always a straightforward question. I think we have to get back to this discussion of community and what we're trying to do and it's not going to fit everyone's definition of the problem. So OpenStreetMap is Wikipedia for maps, that's how we explain it. I think we can see how that lines up, it requires pa consensus, with sort of a preference for experts. That's the local knowledge. Anyone company edit, community driven, and the data must be releasable under this license, open license, so open data. I think that's the three main things to keep in mind when we're talking about what belongs. So what doesn't belong in Wikipedia, for example? >> Maybe my 200-page manifesto about Rick Astley, it would be awesome to weave that into Wikipedia? No, probably not OK, and sorry, that doesn't really exist. In case you were excited about that. >> So what doesn't belong in OSM? This is the classic example we've all seen bad data, null island data. Obviously doesn't belong in OSM. It's not always this straightforward, though, some bad data, problematic data as we've been hearing, is harder to discern, and then we have other types of data that might not belong. So what else doesn't belong? I mean this is not a complete list, obviously. [laughter] But we have Pokemon Go it's the fad right now, should we load all the Pokestops in? Probably not. The licensing might be an issue and we don't want that type of data in there. It doesn't fit that entire community. Maybe a little bit of a harder one. My favorite places with free WiFi and I think the WiFi part is important. We want to record places to have WiFi, that makes sense, but I think where the problem comes in is the favorite places part. We want to record what exists and not put our opinions in it. There's another platform for that kind of data. And the last one, all the land I own or anyone owns and I think this is kind of the serious one, because property rights, whether ownership or mining or oil, concessions and these other types of rights that we have for data, you know, it's the type of data we work with, we want to match that data up with OpenStreetMap data and look at it and get that context, but it doesn't belong in OpenStreetMap for various reasons. So I still have to introduce Moabi. Moabi is all about monitoring natural resources and deforestation in the democratic republic of the Congo. And you probably heard other people give information about this in previous conferences and it lets us do something like this with all this data that we've loaded in and maintaining through that system. >> So like I said, Moabi is clone of OpenStreetMap tools and we made it that it's because it's easy to edit. It's easy for people who are already members of OpenStreetMap and give us a good starting point to build this system. so it let us make maps like this and this is Virunga national park in DRC. It is one of the few remaining homes for mountain gorillas, and it is also a oil exploration concession, there's a firm that was granted access to explore for oil overlapping with both the park and the gorilla habitat and that's a very controversial issue. There was a documentary so you might want to check it out for this issue. So sort of problems with the system? The first thing is not really a problem, over just the design, but it was only focused on DRC and we want to think about ways to expand this beyond just that reason. It's difficult to import large datasets and this is just coming from OpenStreetMap. The data that we work with it's managed externally, a lot of it, so we had to bring it out, load new versions in and maintain all of this. And that platform wasn't the best for working with that kinds of data and we don't have controls or permission structure to control who can change the data once it's in the system and that was one of the problems we ran into. We wanted to have some control over the management of the data. So this lets me introduce Maphubs. We say that it's a home for the world's open mapping data and a tool to easily make and share maps and it's our baby. We've been working on it for about a year now, and we're happy to show it off and how we can bring some of this GIS data and OSM data all together and make maps with it. >> So here's some of the things we can do with map had you been. So we can import map layers, we can link external services, we can manage, control the map layers in groups inside the system, so we can edit the data, and we can publish, quickly make a map with data in the system or these external data sources, and write map stories or create what we call Maphubs, little mini websites that feature maps and stories, all one page for your project. So map had you been's technology, we'd like to mention we're standing on the shoulders of giants. A lot of you in this room we used the iD editor, we use MapBox GL, we use map box for our base maps, and we are a hundred percent open source and on GitHub. So I want to jump into the demo and make sure I can show you guys this. So this is map pubs and what I would like to do is recreate that same Virunga park map that I showing you from Moabi. So in Maphubs we explore all the data that's already in here but what I want to do is create a new data layer, so I'll go to layers and click create new layer. So let's pretend we were out in this area and we observed a an area where we saw gorillas, so let's make a gorilla habitat layer. Just say that it is a demo. And here I have to pick where I want this data to live, and I made a group for the demo data here. All right, now, I would have the option here to load data in if I had a shape file or something, but I'm going to do a new demo dataset and polygon. And pick the source. And I'll say this is Moabi data and we can pick the license of we have all the creative commons licenses and updated licenses in here. We're going to go ahead and zoom in to the part of DRC where we want this data to live. Little preview map sheer so we can do the right part so now I've made an empty layer and let's put some data in it. The first thing maybe to do is I want to add a field, and so we can count the number of gorillas. So I'll just call that count. OK, now we're ready to draw some data, so we go to edit map. And we get iD editor, this is hopefully familiar. I want to just switch to another base map here so I can find a good spot. A little turned around. All right, so I'm going to draw a polygon up here. So there's the area and we saw some gorillas here, and let's say we saw 9 gorillas, and give it a name. So now we have some data, we'll save that. And we can see our polygon over here that we've made. so now I want to make a map and now we can see all of these layers that are in Maphubs so we can add them in our map. So I want to click my gorilla habitat area that I just made. Zoom in here. All right. Let's add a few more layers. We have a mountain gorilla habitat from IUCN that I can bring in. We have protected areas from DRC from Moabi that I can -- if I can spell. And as we saw in that from before, we have oil blocs, oil concessions in DRC that overlap with this. And so let's bring that in. So we can see all this. All right, and we also have OpenStreetMap data. In Maphubs. Most of it is land use and roads right now, but we plan to add more. So we'll just search for OpenStreetMap. We can see all of these OpenStreetMap layers that we have. I'm going to look for residential areas. So now we can see the residential areas that are bordering this part of the park. And finally I want to add forest loss information so we can see some of the recent human activity in this area. All right, so now let me sort my layers here, and -- oops, go back down. Move residential areas. Let's move that up and save it. All right, so give it a name, and you can save your map. And there we've made a map with Maphubs and now we can share this. Twitter, for example, and we can tweet this with a hashtag. All right. Great. So a lot we covered. We talked a little bit about what doesn't belong. There's definitely a much bigger discussion to have with that. We can do a birds of a feather if we want to have a discussion on things like Pokemon Go stops. And we talked about map hubs and we showed what was available if we bring all of these together with some of these OSM and GIS data layers. So quickly what's next for us? We have another project called map for environment. This is a clone of a tasking manager focused on environmental tap app topics, for example the gorilla habitat that I was showing, we have that up for map for environment, we can degree that up. We're building premium features for Maphubs, the main one being privacy, if you want to make sure that you only share inside your organization or data you only share with your coworkers, we plan to offer that as something that we can build on top of what we have and we offer services to help you get started with some of this stuff and the other thing to mention for map for environment, we do have a mailing list on there, so if you go to that page and sign up we'll have more information about mapathons and all of that. And that's to create all of that data that we overlead developers and some of these datasets that we will combine in Maphubs. All right, thanks, everyone. I guess ... [applause] OK, we have a couple of minutes for questions if we have any questions? >> AUDIENCE MEMBER: Does your platform offer any analytic capabilities. Such as if you want to make a map where gorilla habitat was within a certain buffer of deforestation. >> No, we don't offer of that right now. We do have downloads that are in so if you downloaded them and use your favorite GIS. We might build that in the future depending on what people want. >> AUDIENCE MEMBER: How does Maphubs compare with arc GIS? >> Um, do we have time? Heh-heh. I mean yes and no, maybe, I don't know. Really, a lot of the people that we work with are closer to the OpenStreetMap community and they're familiar with those tools and that's why we're sort of taking that direction of grig some of these more GISy tools to those people. So in some ways, no, we're not really trying to do the same thing, but we're sticking with the people we work with in places like DRC. And yeah. >> Hi, I hope this isn't a sensitive question, but I see also a maphub.com. Are you same or different? They're you're Maphubs and there's a maphub.com. >> There is a maphub singular.com? That's -- I'll have to check into that. That's news to me. There is a maphub.net. >> Maybe that's what I mean. >> Yeah, I mean I'm not going to say a lot about that, but I think they announced themselves in the end of March and we announced ours back in December, so it was kind of a who got to this, I don't know. But I think we're working in different directions, but we have specific people that we're sort of, you know, showing or want to come to our site, so hopefully we don't walk into each other too much. All right, thank you very much. [applause] All right, hello everyone. We ready to get started? >> Cool, so my name is Lindsay Jacks, I work at the Cadasta Foundation as a developer. I'm originally from Arkansas, my background is in anthropology. I decided to want to become a programmer because I thought fundraising was super-galoring but staring at 3 lines of code for three hours was exciting. I'm new to the mapping community but I'm excited to find a place where all of my background can be packaged up into a nice little bow. I'm excited to see Field Papers in so many of the presentations today. I had no idea how many people were actually using it when I started this. And you'll probably hear people saying and we'll use Field Papers and move on without any explanation for what it is. I want to talk about the why before getting to the what. So for the first little bit I'm going to focus on the following question: How do we increase participation in community mapping projects? So this is is a pretty broad question and so let's come up with a goal for our hypothetical community mapping project. Let's say that we want to involve more community members in the mapping project so they feel ownership of the project. It's our belief. At Cadasta that if you want the effects of your project to last, you need to make sure you get buy-in from the people most affected. If they feel ownership of the project throughout the process, it's more likely to have a lasting impact. So we're going to break that down into two more things. We want to lower the barrier to participation for people that are interested but intimidated by the process. How do we insure that we're not leaving people out who are not unfamiliar with the process or the tools that we're using and two, we want to find a way to include people who would normally be left out due to a lack of resources or technology. In many situations, especially in the nonprofit world, you're have limited funding. So when you're look loo being at lowering barriers and trying to work within resource constraints, you need to know what your options are. at Cadasta, we provide a suss tomorrowized set of tools to find the best way for each community to map their own information. What works for community A isn't necessarily going to work for community B because of a difference in resource, demographics and size. For example, when we partnered with Kosovo, our focus was on mapping the property rights of women in the village who didn't have good documentation or had documentation in the names of their missing male husbands. Or relatives. When we joined them, they had already collected high resolution imagery from drones. So what they were interested in doing was using mobile technology to replace paper-based technology. Paper-based methods. So by the time we joined, they already had the imagery and they already had technology, so moving forward with tools like geo OpenDataKit totally made sense, however, not every group we work with is going to necessarily have access to this high resolution imagery. That would be awesome, but totally unrealistic. In some cases they won't even have access to mobile technology. Because of this we make suggestions and recommendations based on what makes sense for each unique case. So let's talk about resource utilization first because there is a topic that many nonprofit organizations lose sleep over. Chances are if you are a nonprofit, you have limited resources and technology is expensive so to make sure that we're choosing the right tools we always ask ourselves the following questions: What is the problem we're trying to solve? What are the existing tools? Do these existing tools meet our needs? If they don't, can they be modified, and finally, what is the cost. Not only the actual cost of the technology, but for staff time and maintenance. Something to keep in mind answering these questions is that introducing any new tool takes time, money and effort. Choosing the tools that you're working with within your projects, if you get the chance to check it out, I highly recommend it. But if you're looking at the cost of the tools and the cost of the staff time and the training, you better make sure that what you're choosing is worth it. New technology and mobile technology andiron technology are all great and in many cases they're the correct answer. When we worked with Kosovo we confirmed that using tablets and tools like geoD collect, they didn't add any additional time and they produced more accurate data. It solved our problem, it was cost-effective and it could be modified to fit our teams. It was a win-win all the way around, so yay, team! We have to put a lot of thought into what we're doing. Does the benefit outweigh the cost? Does the introduction of these tools introduce a barrier to participation that wouldn't have been there otherwise? In the case of cose they had the infrastructure in place to quickly adapt to this form of data collection so it made sense L. In a situation with less funding and less technology you should be able to change. So what are the tools that exist to solve our problem? There are so many. In the mapping universe we have GPS device, we have OpenDataKit, OpenMapKit and drones and the list goes on because it's an exciting time to be alive. Do they meet our needs? If they don't meet our needs, can they be modified to do so? Maybe, if you have want to hack on a drone, be my guest, but that's going to take a little more time. And finally, what is the cost and this is sort of where it falls apart in this example. Technology is expensive, devices break, people need training, so sure it would be awesome if we could have a small fleet of drops, a pile of tablets and enough GPSs to arm the entire village but that's expensive and given the background of the village there's a chance they would need more training. So there are a ton of variables you have to consider before jumping on what's new an exciting. Whether or not we're willing to sacrifice some of the benefits of this technology should depend on how it affects participation, because technology still needs humans. When you're looking at the technology you're choosing, you should think about the number of people that will be able to be involved and what kind of training is required, which is reliant on a number of factors, including sophistication of the technology and the prior knowledge of the community. In Kosovo, training took a half day. In the village of people unfamiliar with mobile technology, it would take a bit longer. Much like the choice of technology, the amount of training will be unique to the community that you're working with. So let's think about that remote village again. Say you have the resource to purchase some tablets, you decide that a few people with tablets is going to be a lot faster and more accurate and ultimately cheaper than training unan entire village on something else so that's where you decide to put your resources and that's a legitimate choice. So you buy a bunch of tablets and you train two people to use them. Great, so these people know how to run the tablets so they can run the project. But because they can now use computers, they decide to move to a city so congratulations you improved the lives of two people. Now, this is not something that happens every single time. But it does happen occasion and it's something that you need to take into consideration. While it might be more cost-effective to train a small group of people, the problem is that you can't rely on them to remain in the community forever. If one person decides to move and take all the knowledge with them, your project dies, if your knowledge is spread across a wider pool of people it will less likely be a problem. In addition to decreasing the risk of your project, the more people you involve the more access to information you have. Participatory mapping, the emphasis here is my own is based on the premise that local inhabitants have excellent knowledge of their local environment. You have to get people involved. If you don't, you're missing out on your greatest pool of information, you are a cheating yourself, you're cheating the project and most of all you're cheating the community you're trying to help. The greatest lesson is you don't know the answers. You don't even know the right questions until you get there and start talking to people that live there day to day. It's also hard to get people to care about the work that you're doing if you don't include them in the process. So what now? Let's look back at that remote village again. What's our problem? Our problem is we're working with a community it limited access to mobile technology and you have limited resources. Some possible exclusions are you could spend your money on drones which would be super-cool but you'd get zero community buy-in aside from vague curiosity. You could get a couple of computers and train a few people. You could maybe buy more tablets, but then you're leaving people out who are intimidated by the process or arbitrarily limiting the number of people who can participate because there's only so much tech to go around. So the technology that we've been implementing in every other situation suddenly doesn't work. So what do we do? Don't spend time and money on new technology if a good solution already exists. And I would argue that sometimes a pen and paper is the best option. At Cadasta, we know we're only filling a void. Even with all these advances we still rely heavily on a paper-based approach to data collection and we are always looking for ways to improve even this simple approach. And all of this led us to Field Papers. So Field Papers is a tool that was built by stayman group that allows you to select an area on a map, print it on to paper, take if out in the field, mark all over it, take a picture of said paper, load it back up and edit your annotations and if that sounds super big it's vague, it's because that it is. And I would argue that vague is a good thing. It's simple, it's easy and it's open ended. There's not a whole lot of say about it and that's really the beauty of the project. So I'll walk through it now. I'm not doing a live demo, because I'm not brave enough. So once you visit Field Papers.org you're presented with this page. You'll click on make yourself an atlas. What you'll see there is a page with grids on it and the grids represent individual pieces of paper. You can adjust the number of grids, you can adjust the paper size, width or orientation, you can change the base map so if you're worried about saving ink, you can use black and white. If you don't care about ink, you can use satellite imagery. You'll click make an atlas and from here you can download it as a PDF. And at this point, you'll get these pieces of paper which you might have seen in some other presentations. The dots there are for geo referencing once you load it back in and your QR code connects the physical map to the map that's in Field Papers. You'll take it out on the field, write all over it could your heart's content and once you're done, you're going to take a picture of this and load it back into Field Papers, which looks super unexciting, but what's cool is you can take it into the iD editor and trace your annotations straight to OSM. You don't need a GPS to make a map or learn fancy software to use Field Papers. Field Papers fills the need of Cadasta and many of its partners because it's easy to use, it's inexpensive, it allows people to stretch their resources and it allows a minimal amount of training. Many of the partners we're working with are already familiar with the participatory mapping but the problem is that so much of this information never goes anywhere. Field Papers bridges that gap because it allows you to load everything that you've connected straight into the OSM. And if you saw the POSM talk yesterday, you're constantly building on the work that you're doing. It gives participatory mapping projects life beyond the paper that they're collected on. If any of you are run or attended community mapping projects, like the mapping DC that was presented earlier, Field Papers provides a unique opportunity to make it easier for people to get up and go. There's less fear about messing up real data in OSM, there's no need for everyone to have the same operating system and there's no concern about running out of battery and at the end all of this information still goes back into OSM. I'm not encouraging people not to use OSM, I'm encouraging you to get new and creative ways to get people to contribute. They don't have to understand anything about open mapping to understand Field Papers. Because Field Papers is open source, Cadasta was able to get involved and make improvements to the project and Cadasta sponsored an intern. The old Field Papers consisted of four steps, search, select, describe, and layout, and the two main issues that continued to pop up on the GitHub account was that this process was a bit confusing, and the actual creation stage was far too small to be able to like navigate comfortably. So what we focused on, and what the final design looked like was we took these four steps, meshed them all onto one page and pushed them on the size so that the focus could be on the full screen map. The process is all there so they can see everything they have to do before moving on and we also simplified the grid interaction, which was pretty tough to navigate, you had to drag the map, then drag the grid, then drag the map then drag the grid. This one behaves a lit more like a lens of a camera, so as you scroll across the grid goes with you. We have a lot more plans. We want to turn it into an isolated reusable component that can be plugged into any software. Currently it's locked into Field Papers.org which is built on a Rails application. Inside of that is the leaflet in Javascript and the actual conversion to PDF happens in Python so we want to tease all of that and make it an isolated component that people can plug and play in any software, because we think this has application outside of where it's currently reaching. We want to up update the georeferencing system, so the dots are fine and they work, but with the advancements and image recognizing software, we think we can take advantage of that. Because currently if you fill in the dots or play connect the dots with the dots, they don't work, and they cover up part of the map which isn't super convenient. We also want to increase the capacity for UAV and high-quality imagery. We actually tried to use Field Papers in Kosovo with the imagery that they had collected but it wasn't quite set up to handle that. It can handle satellite imagery but it's not quite ready for that level of detail but we wanted to take advantage of that. After seeing the POSM talk yesterday I personally would like to make it mobile friendly. We talked to the community a little bit but without seeing how people were using it day to day it was tough to make these assumptions so we focused on desktop only but I would like to make that work within the POSM system that's set up. We're still figuring out the best way to go about it. If any of these you find particularly interesting or you think you have a great solution for it, we're working on it tomorrow during the code sprints and we would love to hear from you and we would love to have people join us. I personally love Field Papers, because it's way to get people to get involved and it would be great if other people joined in on it, so thank you. [applause] Questions yes. AUDIENCE MEMBER: Sorry for my English, I'm from Colombia I have a question. We work with Field Papers and a tasking manager instant. Is there going to be more integration between the tasking manager and the Field Papers, like the same imaging area -- >> I got it absolutely from Seth.'S very excited about this, so yes. Seth is one of the original designers that I've met this weekend for the first time, so yes, yes. >> AUDIENCE MEMBER: OK, great. >> Yes? >> AUDIENCE MEMBER: Just when people are filling these out, particularly in more densely -- areas with a higher density of features, pen thickness, pen color, being able to make out what people are writing or doing, how do you overcome those kinds of challenges? >> Right, so I haven't tested this ha whole lot but just the little bit that I've played with it, the higher quality the image, the better it's going to -- so I took this with my phone and you can tell that it's pretty spotty because the lighting wasn't great and you have to get pretty close to it to be able to actually edit it in OSM, so the more zoomed in you are on the space the better it's going to be because you won't be able to see the actual markings until you get close enough. It won't appear at a certain zoomed-out level. >> I'm also talking about the challenge for users, as well, so if you have a densely featured map and you're asking them to annotate or do things like that, it might be challenging for them to get their annotations in there. There's like an urban density thing going on. >> Yes, yeah, tile resolution, and also what's nice about this is they can do their best and it can be corrected og the end because you're not editing data straight in. You have a chance to toy with it the at end if things aren't going quite right. >> So Field Papers is excellent for very dense areas. We've used field papers in Daka in Bangladesh which is the densest city in the world. That was all done in fold papers using OpenStreetMap. Pen width is not really an issue. You say something about colors, we'd love to see colors OCR, if you draw red, what does that mean, if you draw blue, what does that mean so you'd have to do less drawing. AUDIENCE MEMBER: Question from one of the Wikipediaance in the room. So I've gotten bits and pieces of this throughout the weekend and I'm excited by what I see here, but I'm still not following the total workflow. So I'm wondering if you've got it documented at the website the process. Because I'm thinking of some applications where we might be able to do some joint stuff. >> So I don't know if there is specific lined out start to finish. There definitely should be, but what's nice about it is it's simple enough that you can kind of walk through it and play with it on your own, that's almost the best way to sort of figure it out because it is a fairly simple process and it's fairly vague until you actually get in there and do it. AUDIENCE MEMBER: Hi there. I'm from the field sensors and we're working with a -- [inaudible] or to compliment data of the census. One of the problems they were having there was to what they call segmenting, the unit to run the survey. Is it possible to run some kind of algorithm that may set a second for these people? >> Could you clarify? >> AUDIENCE MEMBER: The point being you have to run a census, OK? And you need a unit which is basic. Sometimes it's basic and you can do it with a usual selection, but sometimes it's not. Is there any experience where you have been working that has set somehow units to survey, or not? Within the -- >> Not in my personal experience. >> What if you have to survey an old land uses or different units that look alike? >> Oh, OK, I see what you're saying. So I think to answer your question, you can change the base map that you're working with and you can also add your own tiles, so if you're looking at specifying certain features, is that what you're saying? >> AUDIENCE MEMBER: Yes, but a geometric feature, not exactly a feature, a special object as a unit. That covers, like, say 10 houses or -- >> Yeah, you could add your own base map, so we can integrate with other tools to make that happen. If you add your own base map, you can sort of cover whatever you want and it will still print out whatever you're seeing. >> It will print within the field maps for your reference? >> Yes. I think that answers it. Time for one more? >> Yes? AUDIENCE MEMBER: So a question I've had about Field Papers as somebody who used to work for the company who made Field Papers, is there an easy way to get or is there an easy way to and is there a reason to not put the data into OpenStreetMap and is there a way to put it into a separate database if you need to? >> Yes, so like the last talk I was talking about, there's sometimes data that's not appropriate to put straight into OSM. You can use it in other editors, so this just creates a base-layer map that you can export to other editors depending on what you want to use. So this picture that you end up with here is just that extra custom base layer. Cool. Thank you. [applause] Dynamic timewarp barycenter averaging: Repairing polyline path information with user trajectory data,. By Matthew Redmond ... ... ... ... ... >> OK, folks, the clock on the wall says 4:00 so we are going to go ahead and get started. This is Matt Redmond. >> Hello! So hi, my name is Matt Redmond, I'm a data scientist at Strava. My background personally is in computational geometry and linear algebra. Strava's goal is to build the most engaged community of athletes in the world. Our athletes upload GPS activity, we allow them to compete asynchronously with each other so if you are running by yourself, you could go run the same course as the pros and compare your times. So here's an example athletic activity. I did this yesterday. We went to bane bridge. It was delightful. Took a little ride around and you can see that we plotted this nice little polylypo. So Strava activities in general are between 0 and 200 miles with, you know, some variance in that. Some people ride crazy long, but for the most part that's a reasonable bound on how far our lines are. So as I alluded to earlier, the primary one of the primitives we have at Strava is the concept of a Strava segment. So a Strava segment is an athlete-created polyline to race over. The athlete with the fastest time is the queen or the king of the mountain. For historical reasons, it is always of the mount aye. It sounds better than Regen of the Levy or ruler of the flat place. Users, when creating a segment choose a start and end point from their own sensor data. When you upload your activity, your activity is matched against all in the database and we do this so you can track your progress against whatever you're interested in. Unfortunately sometimes segment quality is terrible. Here's an example of what our segment experience looks like. So this is a ride in northern California in the peninsula region. There's a famous climb here and you can see an all thetude chart and also an overlay. So Strava uses OpenStreetMaps with some custom graphic design stuff on top of T so that was good segment date Y here an example of fairly atrocious segment data. You can see, I think this is -- this the in center of San Francisco, this is the mount Sutro climb and whoever built this segment had some interesting GPS noise happening with their device, so the segment that the got recorded initially veers fairly off of the bend there and anybody who chooses to ride this segment in the future will see this and have a bad experience so it's in our interest to not have our users see this sort of degraded experience so one of our goals was to clean this up with a segment repair process. So here's the problem with segments, I think fundamentally, segment quality is poor because we have this underlying dependency on activity data and this has a few problems, so for one, the reason that it's based off of activity data is that we expected users to be able to go back and trim their segments and you wanted to store the original stream so you could edit segments after creation. But this is kind of a problem. It's a leaky abstraction. Really in my mind a segment should be just a stretch of the planet. It should be a polyline and it shouldn't have any time-varying data in it. Tying segments to activities also introduces some sampling biases. And what this ends up is oversampling where athletes are slow and undersampling where they're fast. And finally, one another problem with tying segment data to activity is that as sensors improve we'd like our segment quality to improve, but if you're tied to an old fixed activity, you have old sensor data and you lose out all an you will improvements on filtering and all that sort of thing. So one of my goals to fix segments was to build a repair tool and I had a couple of goals in mind for this repair tool. Namely I wanted it to be fast. Ideally I'd run this over our entire dataset. It should be deterministic, I don't want my repair to give me random results. And I'd want it to be item potent and I think the hardest part is I'd like it to be verifiable. I'd like my segment repairs to be quantifiably better in some ways than the original was and if they're not better I'd want them to refuse or reject the repair and keep the original path. So the key idea for me was harnessing our segment-matching engine so when you go for a ride or a run, you upload your GPS data, your path is matched against nearby candidate segments. And we identify some possible candidates. Those are stored as tiles, and so there's a state machine that goes through and plays your activity path over the set of tiles and it keeps track of which tiles you've progressed to and which tiles you've jumped out of and at the end of the day it matches some statistics. It's fairly robust. It's been tested it's been in production for quite a while and in particular it's tuned for lenience. So we already have this large dataset of people who have been matched for segments and instead of using one activity we can use the collective wisdom of many activities. So to repair a target segment, identify a collection of activities that matched that target. From the training set, you either choose some really good example or with some magic, create a new one and we'll talk about that later. As a repair candidate and then finally what we do with that repair candidate is we test it as a comparison against the original against the testing set. If the statistics are stronger for the repair candidate, you can accept it, otherwise can you reject it. So there's an asterisk here. If we are using matching data on what is assumed to be a known bad segment, it might stand to reason that the training data itself is somehow tainted or corrupted and in theory I think this is actually true but in practice we have enough sort of aggregation power that you can get over that and the repairs turn out pretty well. I was hoping to put one of those little shrugging guys images in there but I don't know how to do that but anyway, we'd like to have some way of choosing an exemplar candidate from that training set. What does that mean? Well, ideally we choose the candidate that looks the most representative of the training set. This problem is a little similar to classic calculus and variations problems where we're trying to find the functional that maximizes or minimizes some objectives and some constraints. In this case we're sort of trying to find the element from the set that minimizes the sum distance to the other elements. So mean and median are generally fairly well defined on discrete metric space, so a metric space is a set equipped with a distance function. So on the real line, the common example of, you know, mean to median is for the mean you compute the sum and you divide by the number of elements and for the median you can sort of sort the set and then choose the middle one and this can be extended fairly naturally to points at higher dimensional spaces, you choose your standard Euclidian norm and the mean becomes the centroid. The median is a little trickier. What you can do is define a medoid, which is the element that minimizes the distance to the other points in the side. so as it turns out the elements in set aren't points. But they're sequences of points. We need a slightly more intelligent distance function than the standard Euclidian distance. So there's a couple distances the first one is house dorf distance, the house dorf distance is intuitive the maximum pointwise distance of a set to the nearest point in the other. That doesn't work great in this context. Another alternative notion of a distance measure of the sequences is the Frechet distance, intuitively if you went walking with your dog, what's the shortest leash that you could get away with and still walk at your own paces and this incorporates some notion of sequential points but it's still a lot of wasted information. We lose a lot of valuable information by only looking at the worst distance. >> Dynamic timewarp distance is a robust, well studied distance function between sequences. It's not a metric so it adds a little bit of complexity there. But it historically comes from the speech processing community. People speak at different rates and you'd like to be able to identify one person saying the vowel O and another person saying the vowel O at slightly different frequencies. So they came up with the dynamic timewarp distance by warping sounds in the same domain. Here's a nice picture. City you can look in figure A. Those might be two wave forms of people saying the same letter, and there is a clever way which we'll talk about in a bit to recursively align those sequences, choose the path of indices from Q to C that best represents the correct alignment, so that warp path is shown in figure C. You can see that each point is tied to at least one other point in the sequence and so I don't know if, so you can see here, this point here on the red curve, Q, it looks like it's tied to a couple of points down here in C and similarly, where the pronunciation is made here, you can see those are tied roughly to the corresponding points down there. So timewarp distance then says, we're going to take -- here's how you build it. We're going to define the timewarp distance as the timewarp distance of the last cell in a dynamic programming matrix and we're going to generate it recursively and we're going to say for each index you can either follow -- take one step in the first time sequence, you can take one step in the second time sequence, or you can take a step at the same time in both of them. And then we greedily build this up, and the timewarp distance is the sum of the distance of all pairs of points on that warp path. So OK, so suppose let's take it now that we have some sensible notion of distance between sequences of points. Let's talk about segment repair again. We'd like to choose either a central element or an exemplar element from our training set so in order to do that, we can equip that set with a dynamic timewarp distance and compute some medians, whoo! If you actually go about and do this in theory it should work. You should be able to take your training segment and it does work. So here is some data from Strava's dataset. This is actually an interesting case, I think this is a running segment in San Francisco, so the purple little arrowheads are people's actual GPS points and the red line in the diagram is the original segment data. It's pretty bad. The various other colored lines are samples with various training sets, the median of that training set. So it works -- we get a better segment than we did initially but I'm still not convinced this is good. We have earlier flaws, we're still tying it to a single activity. So what if we had a better way to do this. So this is another example of one of the segments that gets fixed allittle bit by medoid repair, but not sufficiently. Again, the red line here are originals and the colored lines are sampled medians. So I wish I could take credit for this because it's super clever but I cannot. Rather than choose the best choice, we actually try to synthesize a candidate. So Francois Petitjean came up with this. In parallel you're going to take that guess and run the timewarp alignment from that guess to every element in your training set. In practice, we use about 30 elements in the training set. This is pretty fast. Each point in the repair candidate is tied to at least one point from every other training sequence and practice that's usually tied to one or two points which is pretty efficient, and then for each of those points we're going to clump them up and compute the barycenter and update that point to its barycenter and here, again picture it a lot more than talking about it. You can start with some arbitrary sequence here, this is a pretty terrible guess at what the consensus might be for these two pieces of training data and we start down here and we are going to align this whole sequence to the redealing thing and align this sequence to the blue thing, so after this iteration, we replace this point with a centroid of this point, this point, and this point. Next up, this point after alignment gets aligned to this point and this point and we do this iteratively until we see something that looks more like the right diagram. So already you can see that the black line is converging towards a more sensible guess as what the consensus sequence should be, and in practice you run this until you reach some sort of tolerance and you don't get much movement. So in practice, in order to do this quickly, you have to preprocess your data. Training data sometimes if you go for a run and you sit down at a cafe and you leave your GPS running you're going to keep generating data on that bench, but in your run, you might actually sit on that bench. So in order to combat that we pass all this data through a polyline simplification path which is parameterized. Running this filter around over it reduces it to 400 or so, additionally we restrict our training set to data that rooms from devices recorded on barometric alTim terse. Instead of -- the timewarp distance is quadratic, it's quite slow, but you can actually build this approximation algorithm, another very clever guy came up with this, that sort of instead of looking at the whole window for the dynamic timewarp, it instead looks at points near the main diagonal, so it projects a coarse warp path onto the diagonal. Compute's the neighbor's network path, but instead of rerunning all of the cells it only runs over the gray cells. In practice, this is implemented as a C-native library. As far as I know this is the fastest implementation of this. Fairly sure it's the fastest. I started off with the Scala implementation and that was a bad idea. It was quite slow, but switching to sort of closer to the meta code was quite efficient. So let's see finally post processing on this dataset, once we have our repair it's important to clean up sort of the rough edges, see we polyline filter that, we run through a median filter and then we run a quadratic filter to smooth it out. So let's take a look at that. So here is some bad data. This is a climb in California and you can see sort of the start of the climb, I don't know what the heck is going on. Someone is riding like crazy and there's some diagonal cutting across the road and it's basically just GPS noise. But there's enough data we can build a sensible repair. Here's what it looks like after repair. Now, I want to stress that we don't have to be perfect in our segment repair, we just have to be better. So I would not claim that this blue line is perfect. I think there's some strange stuff in this blue line. But we have validation statistics, so we can rerun our matching engine over the original data and over the repair data and see that the match progress in this case got about.46 percent better and the jump quality which is how much it jumps away from the tile set got about 1.5% better so this is a pretty common result. It's not phenomenally better but it's better enough that I would accept this repair. So right now this is an admin tool for Strava support that lets them manually repair segments but over in the process of running this repair tool over all of our dataset. So finally a little bit of a cumulative wisdom. This is mostly from engineering wisdom. It's important to handle altitude extremes separately. There are a lot of variance, especially from barometric data that comes to you as altitude difs. Switching to a native library was very important. Preprocessing data by simplifying the polyline's biggest simplification. And sort of organizational terms, build adviculizations early was really essential. And finally building something you're comfortable throwing away. For me I don't like throwing away my babies, but it was important to have a disposable prototype. Great, so I've stolen a bunch of code and diagrams from these papers but they are delightful and they're quite easy to read, so it's great to read if you're interested. Thanks, time for questions? If you have a question, please wait your hand and please wait for the mic to get to you so that the video can pick up your question. AUDIENCE MEMBER: It will you run into principal curve analysis when you were doing the research on this. >> So the question was, did I run into principal curve analysis and the answer is no, tell me about that. >> It's a paper by a guy at I think the university of Montreal or something like that, where it's taking essentially probe data and trying to figure out what curve best fits a set of points essentially by minimizing distance metrics. It seems somewhat similar. >> It's very similar, yeah. I'll have to look at that. >> I'll send up the papers I have and the sample code that the researcher wrote. >> Awesome, great, thank you, yeah. AUDIENCE MEMBER: How does this relate or compare to the slide tool from Strava a couple of years ago. >> Great. Good question. So the slide tool. I think someone is here who can answer questions about that, but the slide tool uses our heat map data, so it takes all of the activities that pass through a given point in space and it builds sort of a 3D surface that that has a minimizer passed across it and in this way it sort of allows you to run the sort of process over the entire world at once. Because it's using all of the activity data. I was interested in more of a targeted segment level thing. Additionally, the slide tool is very -- it's like it's very well suited for like large polylines and I wanted to focus on something a little smaller. OK? >> Any other questions? AUDIENCE MEMBER: So the last step that you described was manual review and acceptance. Are there any plans to fully automate or how would you approach that? >> Yes, so I've built some code into place that sort of looks like a test statistic like how likely would it be that we would see these matching statistics given that in fact that it is not statistically significantly better. So automating acceptance would basically I would imagine it would be looking at the distribution of those test statistics and then building some robust test to accept or reject automatically. Anyone else? >> No? Oh, one more. OK. AUDIENCE MEMBER: So I guess I'm just wondering, this might be a quick question, but do you have any sense why we would see things like that blue bulge any think it was slide 27, where the quote improved trail has some strange happenings? >> I have some conjectures. One of them has to do with -- I think it could be an artifact of the window width on the median filtering. So choosing -- it's just another parameter to tune and I think you could be seeing some sort of like smoothing artifacts there from people who maybe pulled into the parking lot in that particular case. It's hard to tell. But usually that sort of thing when I was debugging it it was usually window width parameter sort of thing. >> Cool. Thanks. >> Yeah. >> Anyone else? >> OK, thank you, Matt. Thanks. [applause] Behind the Scenes of the Mapen Targeted Editing Series. Indy Hurt OK, on to our next session, this is Indy who's going to talk to us about Mapzen. I have a background in geography, I've actually been studying geography for nearly to years, I can't believe it's been this long. I've attended many of the schools in Southern California that are none FosB geography and while I was at UCSB, I remember my good child writing a paper about citizens as centers and we did a lot of study about OpenStreetMap at the time and I thought, wow, this is really interesting and I wonder if I'll ever come across this again and like here I am now and this is what I do for a living is do a lot of work with OpenStreetMap data and the community. So I'm really excited to present to you a series of blog posts that we have at the Mapzen, at Mapzen on our website. So a series, well, what exactly is this series to begin with for those of you who may not be familiar with a targeted editing ceric it's a series of blog posts and it features a location in the world that we're highlighting, but it's meant to encourage people to edit anywhere in the world and we're trying to keep on a regular schedule and the best laid plans always get changed of my idea of let's make a post every third Thursday or every third week of the month and it just never really kind of works out that way and so I'll tell you a little bit about the madness in the background that causes the schedule to be crazy. It includes some statistics, usually just summary statistics telling you how many of a particular feature is present in the data, whatever we're highlighting. And it includes maps. Everybody loves a map. And these maps highlight the features that are presented as part of the post. And what we really wanted to do is give people a venue to edit, so we've linked those features iD and JOSM so it's really easy to find those features and edit them. And we are in company so we're not definitely the only entity that has been trying to engage editors. We've been inspired by so many other efforts. Map roulette. I'm so addicted to that you can go to map roulette and lose two hours and it's like oh, my God I'm glad I'm paid to do this. But there are a lot of initiatives that are trying to get people to edit and if you want to know how many we have, well, we have 13 posts in the series and the very first one was in December of 2015, and it was the streets without names. And it highlighted segments of roads that did not have a nametag and then we've had several others that were related to points of interest, there were some that were related to different types of routing and different areas of search. I really encourage you to check them out. They're all the quite interesting. But before I get further into this talk, you're probably wondering, like, what's the motivation? Why start a series of a blog posts that are trying to get editors to identify specific features within OpenStreetMap? And what we've discovered, and it's well understood, that a lot of the new editors that are coming to OpenStreetMap are coming from areas that are very heavily edited. It's like you get started and you're like, wow, what is there for me to edit? It just seems like everything is already represented. And so if your first introduction to OpenStreetMap looks like this, you click the edit button, you're like, wow, oh, my goodness, this is very intimidating, but the reality is that there are so many things left that would benefit from an edit, and a lot of these things are things that don't show up from a rendering perspective. They could be extra attributes or extra tags, like for business, it could be the address. It could be the type of cuisine if it's a restaurant. It could be additional things like websites or URLs, or it could be names. It could just be that there is a building and it doesn't have a name. So this top -- this talk is broken into three categories of -- that contribute to making the targeted editing series, I call them the three Ts, the topics, the tags, and the tiles. So if we start with the topics, it's really easy to get lost, because there are so many tags. If you are curious about how many unique tags there are in the OpenStreetMap database, it has exceeded 80 million, I believe, the last time I looked on -- at the statistics for the tags. It's overwhelming. But we decided let's just break this down into use cases and if we're looking at routing, display and search, we can definitely bucket different types of searches to encourage people to edit. So if we start with routing, we might have things that are related to transportation, like driving, we might have things related to transit or bike riding, and the first thing that we thought would be a really straightforward thing to encourage people to edit were names. So we call it the streets without names, but it's really the street segments without names and we also want to give people some understanding of how this characteristic of the data is used for in different applications. So as you know, if you're -- if you have a routing engine and you're doing turn by turn routing and you happen to turn on a segment that that particular segment doesn't have a name, then all you get is, turn right in 500 feet. And it would be so much better if it said, turn right on Smith Street or whatever the case may be. So as we highlight those roads that or those segments of roads that don't have names, it starts to look very overwhelming. This happens to be New Delhi, and you're probably thinking, is that right? Are there really that many segments that don't have names? That seems crazy. And it does, but on the surface, it's -- there's so much positive here that it's hard to envision if you just look at it on the surface. What you really have to do is break down to see what types of roads are missing those names on the segments. And so if we break that down and on this graph I'm just showing the primary roads, and the motorways, which are the highest level roads, there's just a small percentage of those, of the total kilometers that are missing the names. So we're actually doing quite well in this particular subset of cities in Asia. And I have quite a few more cities also in Europe and in North America that are being reported on a -- in a blog post that I encourage you to check out. So once we've talked about our routing, we also have some opportunities to encourage people to edit things that are more geared towards display. And these things, you would think, would be very accessible. People recognize how this might change the map for positive purposes. And here I have this really great example of a convention center in Portugal. I don't know how washed out this is for those of you that are sitting halfway back and beyond, but it is just incredible. It's so well mapped. The thing that really makes it stand out is that it has a polygon around it. It has -- it has the parking areas, it has the walking paths. It really stands out on the map. And this is the OpenStreetMap default rendering, and you can see it in one of Mapzen's tile sets. This is called bubble wrap, this particular style, and it's not as impressive in my opinion as how it's rendered in OpenStreetMap, but it still looks incredible. If we take a look at this in another rendering package or another set of tiles, it kind of looks like, wow, it's missing. Actually, it's not. It's there. It's just represented as a point, and so you can see that if you enhance a feature by adding a polygon, like digitizing the buildings or digitizing the grounds, it adds visual prominence to that feature on the map and this convention center in Portugal ask really well known and a lot of people travel there to attend trade shows. So in our post we wanted to highlight some of those features so in this case we had a post where we encouraged people to edit schools that were only represented as points and we wanted to highlight those schools because it's hard to tell that they doesn't have a polygon. So we can highlight those, and we did the same thing with hospitals and you can see here that interacting with people that are editing based on the targeted editing series and inviting myself to their homes because they live in amazing places and they're doing amazing things by helping to edit features in OpenStreetMap. So these are examples of the types of things that we're trying to encourage when we think of edits that contribute to display. And people search for everything. Predominantly it's addresses and venues. So you might be looking for that great coffee shop that is going to expose you to Seattle's best. You might be looking for a place to ship because you've picked up so many t-shirts in the last couple of days, you can't fit them all in your bag. You might be looking for a building. Does anybody recognize either one of these buildings? Somebody says the empire state building. How many of you are familiar with the empire state build snag. >> OK, keep your hands up. How many of you know the address? To the empire state building? >> Oh, OK, I see your geo coder. She is sitting in the back row. You see what I mean? You really need to have the names of features. You can rely on this young lady here to tell you where the Empire State Building is or you can add the name so that people don't have to rely on the address to get to these features. So popular posts. These are the success baby posts. There's something that they have in common, they all involve highway tags. Now, maybe this is no big surprise, because it's OpenStreetMap but it's really exciting, people really care about the road network in OpenStreetMap. So what you're seeping in those columns is the views, it's the unique visitors to the blog post. That is like collected with Google Analytics and then the tiles is the number of tiles that were served force these maps, so this is saying people came to the blog post and they actually interacted with the map, they searched, they dragged the map around and they edited things. It's protein-of-pretty exciting and unfortunately we have this one here. These topics just didn't do very well. I don't know if it's just that for stadium parking we don't have a big sports contingent within OpenStreetMap editors. I don't know if people just are bored with airports, polygons, or if you just don't care about money, but these posts just they didn't have as much traction, so it's an interesting thing to think about. What do people want to edit, what kind of features do they relate to? So if you're interested in the entire set, this graph is interactive and you can access it from our website. It's part of a blog post that's about the targeted editing series and while you can't read it very well, the reason why I put this up on the slide is because I want you to see how -- there's like a trend up and down. It's overall the series is something that people are interested in and interact with, just some posts are more popular than others. And the dotted line that's going across the chart is the time on page. Like how much time are people spending reading about the particular features that we're trying to introduce them to and some -- there are some surprises. As you would suspect, the highway tag post got more time on page, but then if you a few really jump out, like the fitness centers and the hospitals. People spend a lot of time reading those posts, which I'm very grateful for or it could mean that they're much more difficult to edit than I expected and there was a lot of back and forth trying to figure out like, what do I do for this hospital? So it's a really interesting way to look at the data. So now we have these posts, you have an idea of what the targeted editing series is about, how do we identify what to highlight in those maps? Well, it has a lot to do with the tags, and if we look at this one example for a travel post that we had, the first thing was, well, let's highlight hotels. And so you start on the OSM wiki and you just type in hotel and it will find the wiki page that describes how to tag these particular features. And that's the thing that we want to communicate to the users. Here is the way that the expected way to tag a hotel. So we find these tags. But the thing that you have to understand about OpenStreetMap is that the tagging scheme is very heterogeneous, there are many ways to tag things, and you'll see that at the bottom of any wiki page in OpenStreetMap, there's usually a link to tag info and you can click on that and you can find other variants, so in this case, I did that, and I thought, OK, amenity equals hospital, you really shouldn't tag things this way, but I know that it's very common, so let me just see how many examples there are. And I'm going through and I'm like, oh, look at this, there's one with 700 amenity equals no hotel, and maybe -- I put it in the query. I mean it's -- somebody might go there for a vacation, so I don't want to leave anybody out, but you have to go through tag info and find these different tag variations. >> And this is the adventure, it's like going to an aminesment park. It's ups and downs and the next thing you know, you have this pile of tags and you compile them all together and you're like, I'm ready. I think I have the whole collection of things that represent places you might stay when you go on vacation and what do I do with these now? So this is where the tiles come in play. I have to deliver mySQL to tiles and Mapzen team because the tiles at Mapzen only have a sub set of the OpenStreetMap data and if I'm going to highlight something, I want to make sure the features are there, so once I give queries to the tiles people they're great because they often identify if I miss something. And if I'm really lucky, I mean this almost never happens, I might identify something that they're not aware of and then' the hero for the day. So when that happens, it totally throws off arrased schedule. So they have all these different tile sets that I can choose from, and I cheese zinc. There's another one called refill and I love refill. It reminds me of stamen's toner if you're familiar with that map style. You know, it happened to be developed by the same cartographer, her name is Geraldine, she's fantastic, but at the time when I started the series, the refill style had less tag features in it than zinc, so that's why we went with zinc, but zinc is amazing, it has this great crisp gray gradient shades and it's really going to help me highlight the features that I want to highlight. So if I have these roads here and I want to highlight the segments that don't have names, I'm trying to think of what's a great color, obviously the first thing that comes to my mind is red. It's gray, the red, it really pops, it looks amazing, this is how I want to start my blog post series, I want to have these maps with the gray and the red. And the more I worked on it and the more development I did, I realized that this is not the message that I want to send, because I don't want to send the message that it's wrong and red communicates something negative or something wrong or stop, and that's not what's happening here, because yes, these segments don't have names, but that doesn't necessarily mean that they have names in real life. It could be a place where -P the population relies more on the locality as opposed to the actual street name or it could be that it's digitized as a dual carriageway where there's actually two representations of the road and one of them has the name on it and the other one doesn't. And so red is probably not a good choice to go with. So that's how we ended up switching to aqua. Now, aqua, as you can see, it's very similar to cyan, and a lot of people that start out in OpenStreetMap editing have a little bit of GIS background and cyan ask a very common color used to highlight features, so I figured this will be something people can relate to. It pops, it's bright, it matches well with the gray. OK, this is the way we're going to go so I'm really happy I took some time to think about it before I put out a bunch of maps with red highlighted features on them. So once we have the map in place and we're able to highlight the features based on the SQL, to make sure they are on the map I want you to be able to mouse over the features and get a little popup that will allow you to access an editing environment and so that's what you're seeing here, there is some wizardry between me and a few engineers at Mapzen, spread out over the west coast, east coast, it was a collaboration that we had running, one we wanted you to be able to edit in JOSM and also in iD, but also, what happens if you're look many looking at this map, you're looking at the top right map where we're trying to get you to add gyms or like fitness centers and you say hey, you're missing a gym that's next to the funky door, it's -- it's not there at all. What can you do? So we wanted you to be able to shift click anywhere on the map and still be able to add in missing features. And we have a lot of developers that use our tiles and those folks really care about the data itself within the tile. So this data here, it's a vector tile and the topo JSON is there. People want to see exactly what's happening in that particular tile, so if you option click the map, you see the topo JSON, so if you're into that type of representation of the data, we wanted to make that available, as well. So this is what will happen. You click on one of those links, and not only does it take you to iD, in this case that was the link I chose, it highlights that same segment that you hovered over in the previous map and this was something that was really important to me, because as you saw in the very first slide or the second slide, there were so many things to edit, it was overwhelming. I don't want to just send you to the coordinates in the map, because you'll spend a whole lot of time trying to reidentify the feature that didn't have the name. This way it's already highlighted for you and you can easily add the name and save and you've contributed to OpenStreetMap. We've seen a tremendous growth in each one of these features since each one of these posts were published and while we can't say how many of though edits have been attributed to the targeted editing series, we can see we're really excited to have been a part of the process of inspiring new editors and old editors by making it easier to see the unseen. So with that, I wanted to provide you with all of the links. If you go to the very top one, Indy Hurt.gitHub.io and scroll to the bottom of the page, I'll find a link to all the slides and in addition, there's a whole blog post, so for those of you who give talks and you're like, oh, what if I forget to say something, I'm so less worried about that because there's blog post and you can read all the details. And I'm Indymapper on Twitter. Please send me any details after the presentation if you don't have your question answered today. Thank you. >> OK, we have about two and a half minutes for questions. AUDIENCE MEMBER: Thank you. Good presentation. As I improve Rail infrastructure and transit routes, I found and discovered your very intriguing transit colors. Like in San Francisco, particularly on Market Street you've got BART subway at the bottom, you've got muni light rail still underground and on top of that. And on the surface level you've got tram routes. What's your router supposed to do when it gets multiple layers on top of each other? Do you render the most recent one that got changed? How is that supposed to work or look? >> So that's an excellent question. It's how do you display things that draw on top of each other. And right now we don't have any way to spread out those 6 lines on market so that you see the red line, the blue line, the green line and the black down all going down the same road, but the exciting thing is that we are working on that and we're working on it in two separate teams in Mapzen. We have Transitland, we have Meghan Hade who is actively working on a rendering engine to display those different colors so they don't draw on top of each other and we also have the tile zen group. It takes a little adjusting of the data like you need a cartographic representation on top of the real representation to be able to spread those lines out and provide a rendering alternative. AUDIENCE MEMBER: So would it be like a 3D view? >>Well, the 3D view would be quite interesting for the stations, the substations. I think we might be quite a ways off from that, but the ability to shift them so that they display as a row is something that we're actively working on. AUDIENCE MEMBER: Thank you. >> You're welcome. AUDIENCE MEMBER: I was going to ask, when you showed the statistics for New Delhi and the primary and the motorways, the percentages, did you just not include the percentages of residential roads that didn't have names or -- I mean that was -- you only had like two colors represented there, but, motor way and primary, I thought. >> Oh, no internet. >> For the slide, it only included just those two, and I clicked off the other ones in the graph. But if you go to the blog post, you'll be able to see all of the different road types or just the ones that I chose, which is residential, tertiary, secondary, primary, and motorway. And it's quite interesting. In all cases, I looked at 45 cities across the world, the percentage of unnamed segments, the total kilometers, was always highest in residential. And of course we care about the residential roads, but at least the highest-order roads are usually well represented with names. Yeah. >> OK, we're out of time. Thank you, Indy. [applause]