The visual is inspired by Context Free. I first played with the tool in ’07 and was impressed by the powerful outcome of combining  recursion and randomness. In this visualization each curve(represents a file or a folder) wanders freely in space and gradually fades out until it gets some attention. Each is colored by the last person who made changes. A curve branches when something derives from it (files added to folder; branching versions; etc.) The tree does get out of structure and less readable after some time. I’ll try some improved concept when I have the time. 

Full view (click to enlarge):

With my contribution highlighted:

Click here to play with the visualization. It should work with all decent browsers that support canvas. You can highlight the map by contributor or file. Hover over a node to see detail information. I did mess up the file names so (hopefully) I’m not violating any company policy. Please don’t get me in trouble. :-p

Tools: R, HTML5

Tools: Processing, Illustrator

Collaborator: Dominik Dahlem, Eric Baczuk
Editor: Madhumita Venkataramanan

Barabasi’s lab has published their disease networks generated by genetic similarity in 2007. In our first attempt, diseases/disorders are considered associated if a patient has got them at the same time or sequentially. The resulting network gives us new insight as to how closely connected some seemingly un-related health conditions might be. Such results force us to re-examine conventional categories of disease classification, as the boundaries between traditional disease categories are thoroughly blurred.

I made this interactive map for the general public to browse the data. You can switch between two layouts – network and circular. Dot sizes are proportional to the percentage of patients who sought medical attention in total population. Width of links shows the strength of connections. Hovering over a disease pops up detailed information. Clicking on a disease highlights its connections. It is also possible to filter the links by gender / category / keywords. Zooming and panning is supported. The aim is to let people locate their disease of interest quickly in a context.

Click here to view the video on YouTube.

It is a huge network. The first data files I got was more than 50M in size. It took me some time to get the loading time and real-time performance of the app to an acceptable level. I tried several schemes to filter the links while keeping the look and the structure. The force-directed layouts are pre-calculated. I ended up hard coding the network data instead of loading it from an external file, which helped a lot. The final package was ~1.5M, quite satisfying. And I do like the transition animation a lot.

The network vis was made with Flex and the visualization library Flare, and the user interface with Flash CS4. I’ve had some experience with Actionscript 3, but this is my first time learning Flex. I don’t know if it’s just my computer (Snow Leopard + Firefox 4 + Flash Builder 4), but debugging a Flex application was so much pain. The Flash plugin just crashes at every breaking point. Flare is a really well-done library, I’m truly grateful, but there is so much missing in its documentation. You have to dig into its source to discover the vast possibility of it.

The project was done around the time of my thesis “Seeing Differently: cartography for subjective maps based on dynamic urban data” (and the part of my life that I’m reluctant to look back to). I think the time I really worked on the core part of it was two weeks, then a lot of minor changes over a two months period. This is probably the most polished interactive vis among all projects I’ve posted. Thanks Eric and Dom!

Team: (Senseable) Carlo Ratti, Eric Baczuk, Dominik Dahlem, Xiaoji Chen
(General Electric) Camille Kubie, Aimee Atkinson

Tools used: Flash, Flex, Flare, R

At the first phase of the project, visualizations were intensely used to help the scientists in our team to explore and validate the data. Comparing to the British dataset, the telecommunication pattern in the States featured many more hubs and more entangled connections, which made it harder to be represented in one static image. In the following map I used reversed coloring (the arc color on the source end is the color of the target county and vise versa. The color gradient helps identify geographic regions) to show which regions a point is most strongly connected to.

Another attempt was an interactive map where the user can click on a county to see its connectivity with all the other counties of the country. The strength of connections is defined by quantiles of total call time. The interactive map is also available on our project website. The following screenshot shows the outgoing connections from San Diego:

The partitioning algorithm showed some very interesting results, such as the split of New Jersey and California, and some other states belonging together.  Some clusters extend through the state lines and claim how people form communities despite of administrative boundaries. Yet the communities still largely correspond to state borders. I think it has something to do with the carrier’s rate policies.

The project has been covered by TIME Magazine and New York Times.

Data source & sponsor: AT&T
Team / Senseable: Carlo Ratti, Franscesco Calabrese(IBM Research), Dominik Dahlem, Xiaoji Chen
Team / AT&T Labs: Alexandre Gerber, DeDe Paul, Christoper Rath, James Rowland

Tools used: R, Processing, Illustrator

The dominant influence factor is the climate. Winter is the most polluted season because of thermal inversion and less rainfall. Spring in northern China suffers from sandstorm. Still, you can easily identify the effect of government intervention, such as the significant improvement in Taiyuan. And look how amazingly Beijing performed in 2008 August through September for the Olympics. Click the image below to zoom in.

Tools used: R

Dedicated to my endearing home city.

[Looking into the Forbidden City from Jingshan - BJNews, March 21, 2011]

Update: Images now available as Flickr photostream

Last year I made my first attempt at isochronic map for the City of Paris, where the distance on map is proportional to travel time. Well, maps evolve.

Senseable City Lab is having this exciting exhibition Live Singapore! at Singapore Art Museum. We collaborated with Singapore government and companies of telecommunication, power, seaport, land transportation, etc. to create graphic visualizations that reveal Singapore’s urban dynamics.

In March I received the GPS location record of Singapore taxis of August, 2010. All taxis report their coordinates and availability status every few minutes during operation. Comparing to an animation of dense moving dots all over the map, I’m more interested in the underlying patterns of these activities and how they relate to the structure of land use and road layout.

Our brilliant scientist Chrisian Sommer built a network from this massive data and estimated the shortest travel time between every pair of places on an hourly basis. The data quality this time is far better than what I had for Paris (which was retrieved from Google Directions). It is dynamic, and it reflects real traffic condition. I used 290 control points over the city to distort the map. Selecting any of these points as origin, the other points will move away or towards it according to the travel time it takes to get there.

The final app runs on a big display controlled by a Magic Trackpad. Visitors can click anywhere on the map to see its animation through the month. This video demos the maps for the central business district and the airport. It is quite interesting to see the response to road density, the expansion of congested area and the travel time explosion when rush hour comes.

This isochronic map is one of the series of cognitive maps I’m developing – beyond objective projections, we are enabled to see what the city looks and feels to its residence. Maps may not be static anymore, but reflect the dynamic nature of contemporary cities. Also, maps can be dependent on the user (location-based in this case) – they are now about individuals.

Followed by an introduction teaser to the whole event. I also did the anthropogenic heating one. The other beautiful visualizations are credited to Aaron Siegel and Oliver Senn.

Tools used: R, Processing, Illustrator

Collaborators: Christian Sommer, Kristian Kloeckl

The game was a crowdsourcing experience, with participants asking their own questions to the crowd. Participants sent text messages to a short number through Ken’s awesome mSurvey system. Messages ending with “?” were recognized as questions and appeared on the screen with a sequential number. Messages started with “Q+number” followed by a space are recognized as answers to that question. Questions and answers were displayed in real-time, on a 90-foot screen.

My goal in designing this vis was to enable direct feedback to the users, invoke conversation among them and encourage participation. Considering the dimensions of the space, it was critical to keep the interface simple and learnable. Our team came up with this idea of “questions competing for answers” – each question moved from the left edge of screen to the right, gradually accelerating if nobody responded to it, till it was out of sight. When an answer came in, it was attached to the target question and therefore slowed it down. Each question left a trail behind it, whose width was related to speed. So the more popular questions would stay longer on the screen. It was a competition for both good questions and interesting answers.

Screenshot from the game:

Picture at the event:

I was a bit unhappy that the video staff insisted on adding moving backgrounds to the visualization. Then during the real run, we encountered some technical problem , causing the answers mismatched for a while (in fact a bug in the setup-at-the-last-minute message censorship system – censorship sucks, I knew it better than anyone). Anyway both the team and the guests had a lot of fun with the game. Some of us were sleepless for a few days to make this work, especially our great leader Ken, who’s been suffering from a fever since then – I hope he gets better now. ♥

Visualization: Xiaoji Chen, Yanni Loukissas
Backend: Kenfield Griffith, Reid Williams
And thanks to the rest of the team: Michael Berry, Kristyn Maiorca, Ella Peinovich who made this happen

I got curious how we can visualize how actually important the Chinese provinces are. I was reading Gastner & Newman’s paper Diffusion-based method for producing density-equalizing maps (PNAS 2004) at the time. So I set out to make my first density-equalizing map.

Newman’s code on his website deals with raster image only. I also tried to implement a diffusion simulator in Processing, but it was hard to preserve all the details of a vector map. Below is my first shot. It took (quite) some manual effort to remove the bad points. Still seeking solutions. Anyway, enjoy.

You can recognize in this map how unbalanced China is – the west is barely occupied due to challenging natural environment, and population keeps flowing from the middle towards the economic centers (Beijing and the southeast coast). They become both productivity and pressure for big cities.

What about looking at the provinces from a social network’s perspective? In the following graph I measured how often each two provinces appear in the same media coverage. It is clear now that Beijing is the absolute, mono-center of all China. The social power is not proportional to a province’s population. The south and the east coast get far more attention than inland provinces, which is a sad fact.

Tools used: Processing, Tulip, Illustrator

In fact I prefer to call it ‘the origami class’, which sounds more obscure to my friends. And complex, curving origami was indeed what I expected myself to do at the beginning. Well, the field turned out to be much broader and even more interesting (absolutely an understatement).

Anyway one day we were introduced to Kemp’s Universality Theorem, which says ‘there is a linkage that signs your name’. In proving that Kemp invented 3 gadgets: multiplicator, additor and translator, which perform arithmetic operations on any input angle. Then the idea came to me that if we can design gadgets that perform boolean operations, we can build a computer from just hinged bars.

I always had a thing for mechanical computers. The article that talks about the rope-and-pulley computer by Apraphulians was my all-time favorite of Scientific Americans. This project might be a little nerdy, but who knows — mechanical logic gates do make sense on nanoscale and in extreme environments like outer space.

Here is how I represent bits with bars. Note that each bar is constrained to rotate in half plane so I gain a nice ‘black box’ feature where the implementation of one gadget does not disrupt the rest of the machine.

Here are the logic gates: translator (moving a logic state across space), invertor (X -> not X), AND gate / OR gate (they are mirrored image of each other). Watch the video to see how they work.

Using the above gadgets I will be able to build a full adder (A, B, Cin as input and S, Cout as output). I’ve made a simple simulation with Processing. Click the input bars to switch states:

The models I constructed use wood sticks and rivets. I also proposed another way of building the gadgets — cut and fold them from one piece of flat material. It would be awesome to cut a whole mechanical computer out of one piece of thin metal, roll it and take it with you to a place with no electricity and do some crazy computation.

I’ll be continuing this project during the winter break. Many thanks to Erik & Martin Demaine and Tomohiro Tachi. It’s been so cool!

Tools used: AutoCAD, Processing

Do regional boundaries defined by governments respect the natural way that people interact across space? The URB team of SENSEable City Lab analyzed 12 billion anonymized landline calls in Great Britain to illustrate the true connections between places. The strength of connection is defined by the frequency and period of phone calls. It is revealed that people tend to communicate with those that are geographically close to them. Therefore, it is possible to identify clusters of connections as regional groups. It is fun to compare these new boundaries with existing ones and see how much people really love each other.

The visualization challenge here is the extra dense connections. An ideal vis solution should show clearer and finer pattern as data accumulates, not the opposite. Mauro Martino worked with the team from the beginning and derived the primary concept. I hopped on board later and finished with the final video to elaborate the whole idea. Processing is not able to handle this scale of objects (especially in animation) so a lot of pre-processing was done exclusively for each scene.

For those who cannot use YouTube, click this instead:

The research has also been covered by BBC and The Economist.