How to Cool Down a City? | Issue #89
and some other updates...
Hi there!
Next is back in its original form: five stories, four packages, three jargons, two tweets, and a meme. When I started that two months ago, I said I am putting my energy into some other projects. While they’re mostly baked, some garnishing is still needed. Dea and I created it together, and we will do a post together on it.
Quick update: I'm switching my newsletter from weekly to monthly. My PhD research and side projects are ramping up, and I need to manage my time better. After careful consideration, this is the change I'm making.
Let’s dive in!
Five Stories
How to Cool Down a City
In this really cool data story, New York Times illustrates how to combat urban heat in today’s cities (which are essentially concrete jungle). Singapore has warmed at twice the global average over the past six decades, and the city-state is taking innovative steps to cool down its urban areas.
One of the most effective strategies is planting more trees, which not only provide shade but also naturally cool the air. Singapore is also integrating greenery into buildings, offering financial incentives for rooftop gardens and vertical green facades. Another approach is using light-colored reflective paints on building roofs to absorb less heat. The city is also exploring district cooling systems that are more efficient than individual air conditioning units.
This is particularly noteworthy because Singapore emits less than 0.1% of global carbon emissions, yet it's taking substantial steps to mitigate climate change impacts locally.
Information Camouflage
Have you tried searching for a specific keyword on Google and found it frustrating that it doesn’t “understand” you? Try searching for Apple. You will see all about apple.com but not about the fruit. In fact, Hillel Wayne has a better example:
I wanted to find a museum about zoos. Famous zookeepers, zookeeping tools, how habitats are designed, etc. All searches for a "museum of zoos" (MOZ) instead gets you animal museums or combination zoo/museums. Way more people are interested in combination zoo/museums, which camouflages the more niche MOZ.
What is Information Camouflage?
It is a situation where a less popular topic (A) is overshadowed in search results by a more popular but unrelated topic (B) because they share similar names.
What's intriguing is Wayne's multi-pronged approach to tackle information camouflage. He suggests starting with terms closely related to the camouflaged term but not the camouflager. If that fails, he recommends a more methodical top-down approach, like going through museum directories or professional organizations. He also mentions the power of "Knowing a Guy," stating that personal connections can often provide the most direct answers.
My Homecooked Apps
Spencer Chang is a really creative engineer. He has built so much stuff that I kinda get lost in his website every time I open it.
The article "my homecooked apps" by Spencer Chang discusses his journey in creating various "home-cooked" apps and websites for personal use. Chang emphasizes that most of his creations aim to archive continuous rituals or practices in his life. Here are some interesting ones:
Living Notes: Lazy notes, now and here
Bulletin: Random collection of stories and ideas
Fits: trying new outfits
net-shots: Automatic website screenshotter
expense-tracker: Script to track expenses using APIs and Coda
Tat will tell: Tattoos and time preferences
The paper "Tat will tell: Tattoos and time preferences" explores the relationship between having tattoos and short-sighted time preferences. The study finds that individuals with tattoos, especially visible ones, tend to be more impulsive and short-sighted compared to those without tattoos.
Almost nothing mitigates these results, neither the motive for the tattoo, the time contemplated before getting tattooed nor the time elapsed since the last tattoo. Even the expressed intention to get a(nother) tattoo predicts increased short-sightedness and helps establish the direction of causality between tattoos and short-sightedness.
(Maybe, because I have a tattoo, I do so much random stuff. 🧞♂️)
What This Graph of a Dinosaur Can Teach Us about Doing Better Science
The article is good reminder on why we need to visualize data before jumping to conclusions from summary statistics.
It introduces "Anscombe’s quartet" and the "datasaurus dozen" as examples that demonstrate why relying solely on summary statistics can be misleading. Francis Anscombe, a statistician, created Anscombe's quartet to show that four different datasets could have the same summary statistics but look entirely different when plotted.
The article also mentions a project by Justin Matejka and George Fitzmaurice, who extended Anscombe's concept by transforming any dataset into any shape while preserving its summary statistics, resulting in the datasaurus dozen.
You can visualize them in R using datasauRus.
Four Packages
paletteer is a comprehensive collection of colour packages in R. There are a gazillion packages for colour pallets today, and this makes the experience tidy. Vignette. Github. All pallets.
datasauRus packages a dozen datasets from research by Justin Matejka and George Fitzmaurice which have the same summary statistics but completely different visualizations. Vignette.
rio is Swiss-Army knife for input and output in R. Whatever be the format, import() will read it in, without needing you to specify the format. Github.
streamlit is fast, easy and free way to share Python apps. You do not need any frontend development experience and it can be hosted for free! (Kinda better than R Shiny.) Website.
Three Jargons
Attention Mechanism: In deep learning, this technique allows models to focus on certain parts of the input, rather than using the entire fixed context at each step, often used in NLP tasks.
Generative Adversarial Networks (GANs): In machine learning, these are a class of artificial intelligence algorithms used in unsupervised learning, consisting of two neural networks contesting with each other.
Transfer Learning: A machine learning technique where a model trained on one task is adapted for a second related task. This is especially useful in deep learning where training from scratch requires substantial resources.
Two Tweets
https://twitter.com/zevross/status/1668659173276495885
https://twitter.com/kyle_e_walker/status/1668635195539591171
One Meme
That’s a wrap!
It felt great to be back to the 5-4-3-2-1 format. The next Next will hit your inbox on October 25, 2023. Before that, Dea and I will write to you once explaining about our newest project. Until then, sayonara!