Planet DCS@UofT

May 24, 2012

Serendipity

Imagine Toronto as a Better City

There’s a fascinating piece up this week on The Grid on how to make Toronto a better city. They asked a whole bunch of prominent people for ideas, each to be no longer than 200 words. The ideas didn’t necessarily have to be practical, but would be things to make us think. Some of them are wacky, some are brilliant, and some are both. My favourites are:

  • Give people alternative ways to pay their dues, e.g. instead of taxes, struggling artists donate public art, etc. (Seema Jethalal);
  • Hold a blackout holiday twice a year, to mimic the sense of connectness we all got when the power grid went down in 2003 (Carlyle Jansen)
  • Use ranked ballots for all municipal elections (Dave Meslin)
  • Banish all outdoor commercial ads (Sean Martindale)
  • Ban parking on all main streets (Chris Hume)
  • Build a free wireless internet via decentralized network sharing (Jesse Hirsh)
  • Make the TTC (our public transit) free (David Mirvish)

Better yet, they asked for more suggestions from readers. Here are mine:

Safe bike routes to schools. Every school should be connected to a network of safe bike paths for kids. Unlike the city’s current bike network, these bike baths should avoid main roads as much as possible: bike lanes on main roads are not safe for kids. Instead they should go via residential streets, parks, and marginal spaces, and physically separate the bikes from all vehicle traffic. These routes should provide uninterrupted links from sheltered bike parking at each school all the way through the  residential neighbourhoods that each school serves. They should also provide a larger network, linking each school with neighbouring schools, for families where the kids go to different local schools, and where kids use services (e.g. pools) in other local schools.

Advantages: kids get exercise biking to school, gain some independence from parents, and become better connected with their environment. Traffic congestion and pollution at school drop-off and pickup times would drop. To build such a network, we would have to sacrifice some on-street parking in residential streets. However, a complete network of such bike paths could become a safer alternative to the current bike lanes on main streets, thus freeing up space on main streets.

and:

Car-free blocks on streetcar routes. On each streetcar route through the city, select individual blocks (i.e. stretches between adjacent cross-streets) at several points along each route, and close these stretches to all other motorized vehicle traffic. Such blocks would only allow pedestrians, bikes and streetcars. The sidewalks would then be extended for use as patios by cafes and restaurants. Delivery vehicles would still be permitted, perhaps only at certain times of day.

The aim is to discourage other traffic from using the streets that our streetcars run on as major commuting corridors through the city, and to speed up the flow of streetcars. The blocks selected to pedestrianize would be those where there is already a lively street life, with existing cafes, etc. Such blocks would become desirable destinations for shoppers, diners and tourists.

by Steve at May 24, 2012 07:57 PM

The Third Bit

Everything You Need to Know About Standardized Testing

If poor inner-city children consistently outscored children from wealthy suburban homes on standardized tests, is anyone naive enough to believe that we would still insist on using these tests as indicators of success?
— Kenneth Wesson, in Littky and Grabelle’s The Big Picture

by Greg Wilson at May 24, 2012 04:16 PM

Software Modeling Blog

UML is now an ISO standard (and also OCL, KDM,…)

Antonio Vallecillo (one of the best information sources for this portal as you may have already noticed ) alerts me of the publication of several OMG specifications as 2012 ISO/IEC Standards. In particular these OMG specifications are now ISO standards: ISO/IEC 19507:2012 (OCL 2.3.1) http://www.iso.org/iso/iso_catalogue/catalogue_tc/catalogue_detail.htm?csnumber=57306 ISO/IEC 19506:2012 (ADM/KDM 1.3) http://www.iso.org/iso/iso_catalogue/catalogue_tc/catalogue_detail.htm?csnumber=32625 ISO/IEC 19505-1:2012 (UML 2.4.1 Infrastructure)

by jordi at May 24, 2012 04:39 AM

May 23, 2012

This Number Crunching Life

Testing Intuitions about Markov Chain Monte Carlo: Do I have a bug?

For one project I've been working on recently, I'm using a Markov Chain Monte Carlo (MCMC) method known as slice sampling. There are some good tutorials, examples, and implementations out there (e.g., by Iain Murray or Radford Neal), but for various reasons, I wanted to implement my own version.

Now, debugging MCMC algorithms is somewhat troublesome, due to their random nature and the fact that chains just sometimes mix slowly, but there are some good ways to be pretty sure that you get things right. For example, the Geweke method is highly regarded as _the_ method to make sure you're getting it right. So this exercise is not actually really about debugging. It's more about testing intuitions about the behavior of a sampler.

With that out of the way, on to the question:
I implemented my sampler, initialized it with small random numbers for the parameters, and set it running on a simple test model (which I'm intentionally not describing in detail). One high level statistic that is relevant to look at is the (log) probability of samples versus iteration of the sampler, so I made that plot. It looks like this:
This plot looks a bit surprising. Upon initialization, the sampler moves directly to regions of space that have very low probability (remember, this is a _log_ probability*), and it appears to just keep going to exponentially less and less probable regions. The point of a sampler is that it should spend an amount of time in a state in proportion to the state's probability. And this sampler is making a beeline to a state that is e^-600 times less probable than where it started.

So here's the question: do I have a bug? In other words, if you were my supervisor and I came to you with this plot, would you dismiss this plot and send me back to debugging. If not, explain how this possibly could make sense.

I'll post my answer sometime in the next couple days.

* I'm leaving out constants, so the graph would be shifted down (but wouldn't change shape or scale) if I was including all the constants.

Update: This is long overdue, but Iain Murray nailed it in the comments:
No there isn’t (necessarily) a bug. This type of plot is very easily produced with valid code: e.g. by slice sampling a unit spherical Gaussian distribution in D=5000 dimensions and initializing at an atypical point of high-probability (much closer to the origin than sqrt(D) away). Simple Metropolis and slice samplers can’t change the log-prob by more than ≈1 per iteration, so large log-prob changes are slow and painful. Intelligent proposals, reparameterizations, or auxiliary variable methods can improve matters. This is a nice illustration that initializing with an optimizer (without some form of early stopping) can be a bad idea!
In fact, I was using Matlab's built-in slice sampler, along with a zero-mean, spherical, many thousand-dimensional Gaussian distribution for the likelihood, and initializing near the mode (0).


by Danny Tarlow (noreply@blogger.com) at May 23, 2012 05:19 PM

May 22, 2012

The Third Bit

May 18, 2012

Serendipity

Sustainability Visioning Symposium

I’ve been working for the past couple of months with the Cities Centre and the U of T Sustainability Office to put together a symposium on sustainability, where we pose the question “What role should the University of Toronto play in the broader challenge of building a sustainable city?”. We now finally have all the details in place:

  • An Evening Lecture, on the evening of June 13, 6pm to 9pm, at Innis Town Hall, featuring Bob Willard, author of “The Sustainability Advantage”, Tanzeel Merchant, of the Ontario Growth Secretariat and Heritage Toronto, and John Robinson, Director of the UBC Sustainability initiative and the Royal Canadian Geographical Society’s Canadian Environmental Scientist of the Year.
  • A full day visioning workshop on June 14, 9am to 5pm, at the Debates Room, Hart House. With a mix of speakers and working group sessions, the goal will be to map out a vision for sustainability at U of T, that brings together research, teaching and operations at the University, and explores how we can use the University as a “living lab” to investigate challenges in urban sustainability.

And it’s free. Register here!

by Steve at May 18, 2012 08:48 PM

debeakered

An almost-blog post

A few months ago, Steve suggested I write a paper for a particular conference. He even gave me a topic. Brat that I am, I thought about for a few minutes and then flat out said no to that topic but thought of another one suitable for the conference. I decided to write on the ...

Continue reading An almost-blog post

by Jonathan Lung at May 18, 2012 04:00 PM