jasondavies

Fixed. It was originally Amsterdam, but I was experimenting with resolving ambiguous capitals by using the de facto seat of government, based on other feedback here (e.g. South Africa originally had multiple capitals). I've switched back to the more familiar atlas capital in ambiguous cases.

Excellent point. I've removed Vatican City from the map for now, as it is entirely enclosed by another capital city (Rome) and so its Voronoi cell will be tiny.

I've updated the map to use only one de facto capital city per region.

Fixed!

Fixed!

Fixed!

Looks like they have samples on this page: https://cartesia.ai/sonic

Interesting to compare with a time-reversal mirror for water waves that uses a high-acceleration "jolt" to disrupt the wave propagation at all points in space simultaneously: https://pubs.aip.org/physicstoday/Online/11398/Reversing-tim...

I suspect this is more of a collaborative scientific effort rather than a commercial enterprise. However, they do mention in a tweet that they are supported with funding from the Flatiron Institute (Simons Foundation):

https://twitter.com/PolymathicAI/status/1711389556623450177

The press release on the Simons Foundation website might be a bit more informative: https://www.simonsfoundation.org/2023/10/09/scientists-begin...

Yes, it was announced on the (relatively brief) live stream.

For context, Rich Sutton is one of the founding fathers of reinforcement learning. He is also known for his 2019 essay, "The Bitter Lesson": http://incompleteideas.net/IncIdeas/BitterLesson.html

Thanks. Interpolating between map projections that have the same clip region is relatively straightforward. When projecting geometries from the surface of a sphere to a 2D plane, you're always going to have a discontinuity somewhere on the sphere, i.e. the clip region. The projections supported so far all have the same discontinuity along the antimeridian (the meridian at longitude ±180°).

Extending the interpolation to work for projections with different clip regions should be feasible but there are several ways to interpolate between arbitrary shapes in 2D, so I'd have to give it some thought.

Another way to transition between "polyhedral" projections like the Dymaxion map and the Waterman butterfly is to fold and unfold the maps in 3D to and from a closed polyhedral globe.

Interesting. The data is from 2014, which is when I created the visualisation. Perhaps I should auto-update it every so often!

This is probably the most frequently asked question! The data is from here: http://ourairports.com/data/

I had to limit the number of airports displayed for performance reasons. I filtered the airport data so that only those airports with scheduled services and which are denoted "large" or "medium" are included (according to OurAirports), bringing the number down to 2,980.

It's probably to avoid dead-code elimination, which would defeat the purpose of the benchmark. The rust benchmarks use test::black_box [0], which does the same thing, so there's no unfair advantage there.

[0]: http://doc.rust-lang.org/test/fn.black_box.html

I've just reduced the resampling slightly, using precision(.5) instead of precision(.1), not sure if that helps much.

The files used are from Mike Bostock’s World Atlas TopoJSON project: https://github.com/mbostock/world-atlas

They are in turn derived from Natural Earth data: http://www.naturalearthdata.com/downloads/110m-cultural-vect...

Thanks for the feedback!

I agree it’s not the most intuitive approach for all situations, particularly when you’re used to a North-South orientation for maps. The rolling is perhaps made more obvious with my use of a graticule. For this reason, I’d be interested in optionally restricting to two degrees of freedom when dragging on the globe, and the third angle could be modified when dragging outside the globe only. So the point being dragged on the globe could still remain under the cursor where possible using this approach.

Interestingly, Google Earth uses the same approach as I do, though perhaps it seems less obvious without a graticule.

Just added another example showing raster tiles reprojected using the satellite projection:

http://www.jasondavies.com/maps/raster/satellite/

Unfortunately zooming is not supported for this one due to issues with distortion, but it's something I'd like to fix.

And another (zoomable Mollweide watercolour):

http://www.jasondavies.com/maps/raster/mollweide/

The text is in fact processed in the client, and is quite fast even for large corpora such as the whole Bible: http://www.jasondavies.com/wordtree/?source=kjv.txt&pref...

It attempts to access URLs directly but this only works if the server sends the appropriate CORS headers (hardly ever).

Otherwise, it falls back to using a proxy, which means the client only sees what the proxy sees. However, you can also paste raw text on the main page.

I could imagine modifying the bookmarklet so it lifts the text directly from the browser instead of just copying the URL. This would solve the proxy issue neatly and would also work for local-only or intranet sites, for which the proxy also fails.

Scroll down and there's a bookmarklet that you can drag to your Bookmarks Bar, allowing you to turn the current URL into a word tree.

I tend to agree with Wattenberg and Viégas that it's interesting to treat all words and punctuation equally, but it's certainly a matter of opinion and it would be simple enough to tokenise the input data differently.

Yes. I will probably add the core wordtree layout as a plugin to https://github.com/d3/d3-plugins

The whole application ties together text processing, data retrieval, the wordtree and longscroll.js for fast rendering of the text view on the right-hand side: https://github.com/d3/d3-plugins/tree/master/longscroll

Try now? I was previously loading some datasets directly from Many Eyes, but I'm using my own copies now to relieve load on their servers.

Yes, e.g. http://cartography.oregonstate.edu/demos/CompositeMapProject...

Yes, there are two properties that are important for Google Maps:

1. Mercator is conformal, which means angles are preserved. So if you zoom into any small part of the map, it will look right. Equal-area maps distort angles.

2. North is always in the same direction—up. Together with the conformal property, it means that all directions are preserved: North, South, and everything else.

The primary use case for Google Maps is not to show the whole world at once, rather it is targeted at street-level mapping. Hence, Mercator is an ideal choice.

This is probably my fault; it turns out I've been incorrectly spelling it antemeridian for ages!

Antemeridian is an adjective meaning "of or relating to morning", and antimeridian is indeed the correct word. Thanks, Rob!

Site is down, but I think this is one of the images: http://imgur.com/07ukx

See also: http://www.jasondavies.com/factorisation-diagrams/