TreatSmartLiving: Maximizing your Halloween candies

A few years back Nextdoor started letting its user “announce” whether they are giving out candy or not for Halloween. That was exciting to us ! As real “nerds” we thought what if we can plan ahead the path that will give you the most candy without having to sweat too much!

Hence, we decided to do some legwork to prepare for the big day. We used the data on Nextdoor to identify the locations of houses giving out candies. Given that Next-door includes voluntary information only we also did some manual labor and drove around the area to see which houses at which streets where decorated – this increases the chance of them handing out candies…

Then the problem is “easy”. If you assume that every house that hands out candy provides the kid with a benefit, and assuming that all houses give out good candies so the benefit is the same for every such house, then what we have is essentially a constrained optimization problem. Maximize the “candy profit” by walking only a pre-defined maximum amount of distance!

Luckily we had readily available algorithms that we have developed with for urban tour recommendations (which we have used in the past to find paths full of trees) and hence, we crunched the data and…voil√†, we were ready to go get some candies. As for the evaluation, we will have to wait after Halloween to see how much candy we collected..! Hopefully, next year more people will have the information on Nextdoor so I will not have to do extra data collection…Just a note to Nextdoor folks, let the people say what candies they are handing out too, so we don’t end up with a lot of peanut butter ūüôā

Free parking and small business Saturday

Since 2010 the Saturday after Thanksgiving we are celebrating and supporting small businesses. This was a way to counter Black Friday and Cyber Monday that mainly feature big-box retailers and huge e-commerce stores (well, mainly Amazon). Many local governments promote and support (as they should do) small local businesses. In Pittsburgh (and in many other places across the country) one of the ways the local government promotes small business Saturday is by offering free curb parking across the city.

The intentions are certainly good; no one argues about that. The line of thought for offering free curb parking, is that by removing the “barrier” of curb parking pricing, more customers will decide to drive and come out to support local businesses by shopping. Sounds reasonable, but it is not the case. Sure people love free parking (even though nothing is free in life; we pay for it in other ways), but remember the objective here is to support local businesses. What do local businesses need? Infrastructure to support the – hopefully large – incoming flow of customers. Parking is one such infrastructure, that allows customers to temporarily store their cars, and run their errands, including shopping. Therefore, local businesses ideally want this infrastructure to be able to support as many cars as possible. Given that the number of curb parking spots are fixed, this means that the turnover of each parking spot should be high. The turnover of a spot is simply the number of cars parked in that spot during the day. A higher turnover, means that more cars parked in one spot, which can potentially translates to more customers.

What does free parking do? Free parking actually reduces the turnover of a parking spot! When parking is free, drivers do not have parking ticket anxiety anymore, and turnover plummets. Simply put, someone that might park for 30 minutes, run their errands and leave, might now stay 60 minutes and take their time in completing their errands. So this spot will be occupied for an additional 30 minutes, possibly not allowing another potential customer to park. Will this second person leave the area and not shop? Chances are not, but what will happen is that they will circle around to find a spot, increasing traffic, polution, and we all know what this does to our mood. Not to mention, that they might think of it twice before going back next time.

While we cannot know how exactly people will react when they cannot find a spot (i.e., leave or keep circling), we can get an estimate of the turnover of a spot through simulations. Assuming a simple Poisson model we can simulate our parking spot’s state. Starting with the – realistic for business areas – assumption that there is high demand for parking (which is actually another side-effect of free parking, that is, induced demand), we can simulate the turnover of a parking spot based on the changes in the average service rate of a parking spot. For example, if the average occupancy time of a parking spot is 30 minutes under normal pricing, we can expect this time to increase when we move to free parking. We do not know by how much but we can simulate a variety of cases ranging from conservative (e.g., a 5 minute increase on average) to more pessimistic ones (e.g., doubling the occupancy time). The following figure shows the simulated distributions for the turnover of a parking spot under different scenarios.

As we can see even under conservative assumption (average occupancy time increasing by 5 minutes), the spot turnover is reduced by approximately 14% (from an average turnover of 21 to an average turnover of 18). Businesses need spot turnover. Free parking achieves the opposite. So if we want to support local businesses, we should come up with other ways and not free parking. Borrowing from the core idea of PittSmartLiving of bringing together public transit, commuters and local businesses, free bus rides or free bike rides to specific business destinations could be a good start. This can help local businesses, can help dwellers particularly from underserved communities, while at the same time promoting public transit as a viable option for moving around the city. Equally important, is the signal sent by offering it for free during a small business celebration that public transit is a crucial link between people and businesses.

For anyone interested in the economics of parking (which are very different from what a superficial view of the topic might imply), the book “The High Cost of Free Parking” by Donald Shoup is a must-read.

These Days, Smart Living means Distant Living

At PittSmartLiving we have been trying to find ways to avoid over-crowded buses through win-win-win settings for all involved stakeholders (port authorities, riders and local businesses), improve riding conditions and in general ‚Äúflatten the curve‚ÄĚ for public transport.¬† As part of our efforts we have been analyzing crowding data in buses and businesses. The latter is important since it means that we could use the same methods to study changes in businesses’ foot traffic during the coronavirus pandemic, to understand how people react to recommendations for social distancing (which in the author’s personal opinion it should be termed as physical distancing, since we are still socializing using the technological advancements of our era).

We have been using Google’s Place API to collect crowding data. You might have interacted with this information available from this API when you tried to search for your favorite cafe and got back a bunch of information for it, including the bar chart below:

Screen Shot 2020-04-07 at 8.53.16 AM

This chart provides information on how “busy” the place is expected to be based on historical data from people’s cell phones that use Google’s services, as well as, how busy the place is right now (red-ish bar). These numbers are expressed as a fraction of the most popular time of the week for the venue. For example, if the red bar on the figure above corresponds to 40, and this place is expected to be at its weekly busiest on a Sunday at noon, then currently the place is 40% as busy as during a Sunday at noon. Now, of course, the current value can be larger than 100 as well, which means that the place is busier than the expected weekly busiest time. Google’s estimates for the expected levels of crowdedness are updated in a rolling window fashion (with details not being fully known to the public).

It should be evident that we can use these data to get an estimate of how people are distancing these days. We started collecting data for this purpose from a select number of venues in Pittsburgh on March 13th and we found some interesting patterns. People, in general, were following recommendations (the order for sheltering in place was announced on March 19th and enforced on March 23rd in Pennsylvania). During that week traffic in retail stores and malls was down (approximately 45%), traffic in restaurants was down (approximately 30%), traffic in transportation hubs/stations was down (approximately 65%). Following, are some representative time-series examples of venues that experienced a reduction in traffic.

Steel Plaza Station32020Starbucks32020

The Waterfront32520Phipps Conservatory and Botanical Gardens32020

One significant exception was bars during March 14th that were busier than normal, with Pittsburghers celebrating St. Patrick’s day as it can be seen by the following time-series:

Mario's South Side Saloon32020 (1)

Now the only type of business that did not see any significant decline during the first week of data collection in Pittsburgh was grocery stores. They saw a small decline of 4%, but since that week things have changed, with larger declines observed. For example, the following figure shows average daily changes from a busy grocery store:


These results seem to indicate that people stocked up and distanced themselves even from grocery stores. Among the 30 groceries that we have been monitoring currently, the last 10 days there has been an average reduction of 30% in the crowdedness levels in these businesses. Now it is always good to understand the data. Part of this decline can be policies put by various grocery stores on how many people can be within their premises at any given time. This certainly will have an impact on the volumes reported by Google and other providers. So it is always good to keep in mind these things when trying to understand the data and make conclusions.

Recently, I also came across another dataset from Foursquare that captures foot traffic in venues. Now I was a bit skeptical initially since people rarely check-in to places they go, but digging a bit deeper in the data, these are not based on check-ins but rather on passive sensing of user locations (i.e., similar to what Google does). I was particularly interested in residential venues (that we cannot get information about from the Google API) and how foot traffic has changed there. First I took a look at the US as a whole and following are some interesting figures:

We can see the natural progression here through the month of March across the whole country with foot traffic in residential places being significantly reduced by the end of the month (as compared to the month of February on a similar day)! Now again we have to understand what the data measure. Someone might be confused saying that this does not make sense since we are staying at home more. This is true, but these Foursquare data measure the foot traffic, i.e., how many people are in a building/venue. This means that by the end of March there were fewer people in a residential building than expected (as compared to a baseline from February). This points to people physically distancing from their close friends and family as well, staying home with their close/immediate family only. Simply put, they do not have people over. Following is the time series of these changes for Allegheny County, which tells a similar story.


Overall, people seem to be taking this seriously (as they should) but there is still more that we can do! Stay far from each other; it saves lives!

Note: Google published a similar analysis in the beginning of April that provides a similar analysis for several countries and you can access these reports here.

Navigating an urban environment beyond the shortest path

The first thing that comes to someone’s mind when the term Smart City is thrown around, is efficiency! Efficiency in energy consumption, efficiency in city government operations, efficiency in transportation and so on. Efficiency in transportation has become synonymous to fast/short transportation. But is this really making us, the city-dwellers, smart(er)? Isn’t this making us prisoners of time? Is this what we really want from our cities of the future? Efficiency? For sure efficiency in some (many) aspects is top priority (e.g., energy), but when it comes to navigating through the urban fabric efficiency¬†should not¬†be our top priority. Cities are living organisms and people are the nutrients that they need to survive and thrive. Consequently, following always the same efficient paths will lead to inadequate nutrition of specific parts of the city. Not to mention that this minimizes serendipity (and potentially your chance of finding love as Ariel Sabar describes in “Heart of the City: Nine Stories of Love and Serendipity on the Streets of New York“)

Daniele Quercia and his colleagues developed alternatives to shortest path routing, by considering routes that make people feel happy, routes that are filled with delightful smells (e.g., the smell of a bakery early in the morning) and routes that allow you to experience the city through its sounds. This was a breakthrough in urban way finding and inspired us to take it a step further. Why focus on a single objective for navigation? After all if we focus on a single objective most probably we are still minimizing serendipity, since the path that makes us the happiest will always do so! How about if we really have to be at school by 5pm but at the same time we want to increase our exposure to trees, or to street art? This is an example of multi-objective routing, where we want to find paths that optimize two (or potentially even more) objectives. There are several challenges associated with the problem of multi-objective routing with the two most important being:

  1. Many times the two objectives are conflicting, for the simple fact that a longer path will have more of everything (trees, street art, etc.).
  2. There are many many paths connecting two points in a city and each one of these provides different tradeoffs between the two objectives. However, we cannot show all these paths (possibly tenths or even hundreds) of paths to a dweller.

Luckily we have developed an algorithmic approach that is able identify a small set of paths that capture the different trade-offs that are possible given the structure of the road network. While one can see the technical paper for details, the main idea is to identify what we call non-dominated or Pareto optimal paths. These are paths for which there are no other paths that are better with respect to all the objectives of interest! To visualize that we can assume that every path is characterized by two values, x and y, that represent the performance of the path with regards to the two objectives. Let us assume that we are interested in maximizing the x-objective and minimizing the y-objective. If we plot the values for every pair of possible paths between our original and destination we will get a plot like the following:

Screen Shot 2018-03-19 at 8.34.14 PM.png

We are interested in paths that lay on the red line (called Pareto frontier or skyline, depending on the field literature you are reading). In this artificial example, there are only few paths on the frontier, but in a real network, there can be tenths or hundreds of paths. We have developed an algorithm for choosing a small number of them (7-10) that covers all the major tradeoffs. For example, the two points in the orange circle practically offer similar tradeoffs between the two objectives and hence, we can return to the user one of them.


We have used our algorithm to provide paths in the city of Pittsburgh that offer tradeoff between length and exposure to trees! For the latter, we used a nice dataset recently released that includes information for the trees¬†cared for and managed by the city’s ¬†Department of Public Works Forestry Division. ¬†Hence, our objective is to minimize the length of the path, while maximizing the exposure to trees, making for a relaxing path. For example, let us assume that we want to go from Oakland to Shadyside. There are many paths to follow and the following map shows 4 of them (the ones returned by our algorithm). ¬†The user can choose between the shortest path (the blue one), which is also the one with the smallest exposure to green, or the greenest path (the…green path), which is also the longest path! We also offer the user the choice of two other paths (red and black) that are neither the shortest nor the greenest, but they are non-dominated (i.e., would provide a good tradeoff)!

Screen Shot 2018-03-19 at 8.44.03 PM.png

One can think of several other objectives that can be included such as safety of biking/driving (e.g., due to a snowstorm), exposure to historic landmarks, exposure to places with personal significance to the user etc. The possibilities are only limited by the  data available to us!

This research is part of the PittSmartLiving project, which aims to put humans into the center of urban navigation and cyber-physical systems in general, and facilitate the design if systems that are truly smart – both technically but also socially!