Ducky's InfoVis Project Proposal

Participants

Kaitlin Duck Sherwood, ducky@csDELETEthisTEXT.ubc.ca

Experience

• I did a project integrating maps, floorplans, and a virtual walkthrough about ten years ago at UIUC.
• I have been very interested in a related visualization of census data for 20 years. I haven't ever gotten past the exploratory stages, but this means I am highly motivated to learn how to unravel US Census Bureau datasets.
• I have never written anything in Javascript before, nor any dynamic client-side code of any kind, but would like to learn.
• I am not a graphics whiz. I am not sure what algorithms to use to select what data is likely to be visible. OTOH, I have Google and a copy of Foley and Van Dam within easy reach.
• Roughly eighteen years of industrial experience makes me anticipate and plan for unanticipated problems.

What does this thing do?

Domain, task, and dataset

This project is designed to help people visualize US population data (from the 2000 US Census) by providing them with interactive zomming/panning controls and the context of a familiar map.

Scenario

Billy Rubin, a fifth-grade student at Rancho Cowabunga Middle School, is exploring population density as part of his urban housing presentation. He goes to Webfoot's Information Visualization site and is presented with a what looks like a standard Google map of Palo Alto, CA. Palo Alto not being Rancho Cowabunga, he zooms out until he finds Rancho Cowabunga, then zooms back in.

Once he has zoomed in to the area he's interested in, he clicks the "Show population" checkbox. The map changes to have a translucent overlay over it. Areas with more people have more of a yellow tinge; areas with fewer people have a bluer tinge.

Billy uses the standard Google controls to move around, and the population overlay moves with the map. Billy is able to click-to-recenter, pan by dragging, pan by clicking on the directional controls, and zoom by clicking on the zoom controls. At every stage, the overlay pans and/or resizes to match the map.

Illustration

The following is a fast and dirty mockup of what the UI will look like:

Note that the labels on the bar will need to be adjusted once I figure out what reasonable values are.

Also note that there will be areas with no people (e.g. big parks, the Stanford Industrial Park, commercial districts, the Bay, etc). I have left those clear.

How will I make it?

Proposed solution

I will show maps with the hue of polygons on the map representing the number of people living in that polygon as counted by the 2000 US Census. I will provide context to the users by making the polygons translucent, leaving the underlying map data discernible.

This project does not pretend to push the frontiers of research adequately to eject a publication. This project is designed to do something cool and useful.

I am more interested in providing (and more worried about) snappy performance than I am in broad geographical coverage. There are numerous opportunities for the performance to be inadequate. Determining the right data to display for a given clipping region, retrieving the information, aggregating information (when zoomed out), rendering the PNG image, and serving the image all take time (in addition to the time Google takes, which I won't be able to control).

I expect that I will need to trade disk space for speed, pre-processing information and caching some information on disk. As my last name isn't Google, I expect that I will not have enough disk space to handle all of the United States. I am willing to restrict the geographic range in order to ensure optimal performance. If I need to restrict the area of interest, my final paper will discuss what resources would be needed for the entire US.

Implementation approach

I plan to use Javascript and the Google Maps API to serve maps, capture pan/zoom events, and to overlay area data on top of the Google Maps.

I plan to use the gd library, probably with a perl wrapper, to generate a transparent PNG image to overlay on top of the maps.

I plan to use C language Shapefile C library to parse a variant of the TIGER/Line data -- the ESRI shapefile data -- which includes population demographic information.

Milestones

• 4 November: Proposal finished
• 13 November: One PNG file generated with one polygon extracted from dataset, requiring:
  • extracting polygons with the Shapefile C library
  • extracting the appropriate demographic data from the dataset with the Shapefile C library and the data files
  • associating polygon <->demographic data
  • understanding the gd library well enough to generate a PNG with a transparent polygon
• 20 November: Functional panning, perhaps limited to one US state and perhaps slow, requiring:
  • determining which polygons will be visible given lat/long bounds
  • iterating through visible polygons, drawing each
  • plugging overlays into the Google Maps API
• 4 December: Functional zooming
  • determining which polygons will be visible given lat/long bounds and zoom level
  • aggregating lower-level data (note that I do not think this is trivial)
• 11 December: Performance tuning and bug fixes done, possibly requiring:
  • pre-processing census bureau data into a form that is easier to extract quickly
  • prefetching of neighboring overlays
  • caching previously-used overlays
• 16 December: Status update due
• 19 December: Final due date

Future work

There are all kinds of interesting and useful things that could be done on top of the base project.

• I believe that once I am able to display total population density, then it is relatively easy to extend the code to allow displaying one of many different flavors of demographic data, e.g. the non-white population. I would like to do that, think I can probably do it, but hesitate to promise it.
• Allowing the user to select information from one of several different censuses is also clearly within the realm of imagination. I believe that this is slightly too ambitious (in part due to disk space limitations) for the scope of the class.