show and tell for hackers in DC.

Round 63: Eponymous Documentation

11 Dec 2018

Notes copied from the hackpad!

Joey Shepard

Current hobby: building homebrew calculators.

Why make calculators? TI-83 is perhaps the most popular calculator, but only has a 6Mhz processor and 32k of RAM. The TI-83 is still being sold after 20+ years but technology is now WAY better.

So what kind of calculator could Joey build with modern tech?

MSP430 microcontroller, has processor, RAM, and flash like any microcontroller, but there’s no way to expand that RAM.

There’s an external memory chip that has much larger capacity but It’s slower than the internal memory.

Needed to create a memory converter program to be able to pre-process the external memory accesses, so could write software in normal format and have converter generate functions to access external memory.

Built the casing from copper-coated circuit board substrate, sourced in Kyrgyzstan, hand-cut and soldered together.

Jordan Nelson, @howesjnelson

Does Florida really have the craziest news?

Disclaimer: Not statistically significant!


But there are so many instances of Florida Man! Can we put some data to this?

How wild is Florida’s news compared to other states?

NPR is considered “news”. Scraped 22,000 posts from the Onion back to 1998 and a ton of articles from NPR as well. Built a model to predict whether local news is going to be closer to NPR or closer to the Onion.

Data is super dirty! Tried to remove location slugs; filled nulls with “no_summary”; satire has more profanity in it. Targets: 1: news, 0: satire

Used Logistic regression. 2.9% of Onion articles were misclassified as being real news; 4.9% of NPR articles were misclassified as being satirical

Scraped the second-largest news source and Florida didn’t have the wildest news – it was more in the middle! Illinois was worse!

Belén Sánchez, @BelenSanchezH

My project predicts household poverty level in order to allocate social welfare benefits in Costa Rica and is a clear example of the importance of data science in public policy.

Using data from the Inter-American Development Bank on households in Costa Rica, I developed a model that predicts the poverty levels of households with 94% of accuracy.

Social transfer has a direct impact on reducing poverty; about 10% of Costa Ricans live on less than $5 per day.

Data comes from Inter-American Development Bank, released data on households and individuals

Exploratory data analysis - 34% of households are between extreme poverty and vulnerable; 71% of households are in urban area. Overcrowding is higher in extreme poverty; most people in the dataset has access to mobile phones but fewer had access to television.

Education levels tend to be lower with extreme poverty.

Feature engineering: Rent per capita, demographic information per household.

Tried several different models; best model was Random Forest, accuracy 94% in predicting poverty levels of households.

Most important (predictive) features: years of education; dependency of children living in the household; overcrowding; and children per adult.

Resulting policy recommendations: Focus support on welfare transfers with educational support to guarantee that the children will be going to school.

Most surprising result was how clear the data showed the link between (lack of) education and poverty

Eric Haengel, @EricHaengel

I got the opportunity to design some custom hardware for a soon to be installed public art light installation in an underground walkway connecting to the Alexandria King St metro station. I’ll be bringing in the prototype hardware for a demo.

There’s a very dreary hallway that needed some color and music, so Eric build a light show that turns sound into light.

Part of an art collective that won the proposal; turns sound into light.

Weatherproof LEDs and VERY bright!

Sister chip to the MSP430.

Currently plays sound off of his laptop but installation will be going up in early January. There are surprises and you need to go to see it yourself!

Don’t know the name of the exhibit yet but it’ll be tweeted up on the next DC Hack and Tell.

Shannon Turner, @svthmc

What’s new in Metro Map Maker version 2! Lots of major improvements, new features, and performance increases. Plus a whole lot of beautiful maps!

December 2018! Version 2. Recap, inspired by art, speculative metro maps. “What would it look like if we really really invested in public transit”. In 2017, launched the original. The original was very MS paint inspired. Draw, click, draw, click, was too much. Whoa, maps are 9x larger.

More people making maps → now 10X more maps

Remix any map

World metro map:

Erik Green

Convolutional Neural Net to detect skin cancer from images. 20% of Americans will develop skin cancer, and it’s one of the more treatable forms of cancer, especially if it’s detected early.

Medical imagery is being taken over by machines.

HAM10000 dataset - made up of 10015 images of skin lesions; these lesions were initially classified through expert medical consensus to ensure their accuracy.

Needed seperate GPU to process imagery.

Seven types of lesions in the dataset; some benign, some malignant.

Ran a convolutional neural net with binary (malignant or not) and multiclass (which type of lesion is this)?

This model performed better than doctors on binary classification; on multiclass, Erik’s model was just slightly worse than a german team

Jared Nielsen, @jarednielsen

Machine Learning app that detects whether a given string is more like Samuel Beckett or Yogi Berra. All in the browser!

Live demo:

Machine learning is being ported over to Javascript, meaning we can do it in-browser; meaning we can push it off to the client to deal with all of the heavy lifting!

Brain.js is a lightweight implementation of neural nets in Javascript

Take any given string and compare the text vs Samuel Beckett and Yogi Berra

Veni Vidi Vici -> Yogi Berra He will come -> Yogi Berra

In the console, outputting a log of what’s going on, up to 20,000 training lifecycles.

Dataset is pretty small, so he just trains it each time. If it were larger, it would make sense to cache it.

How many quotes from each did you use? About 20 quotes for each. A very small dataset, but increasing it would also take longer to run and eventually max out the browser.

Hammad Malik, @tomatohammado

Multiply Blend Background CSS Hack

Hero image, but the designer used a multiply-blend – and see how things go wrong:

Multiply cares about what’s in the foreground and the background and multiplies them together.

Parent container, two sibling containers

Uses CSS and Javascript to fix a problem where the source image

Jay Kay

You can get notifications of motion sent to your phone from anywhere within the service range of a mobile hotspot. No electrical outlet required! Monitor your boat, your picnic, your car!

A camera called Nubo, for a baker who was selling bread via vending machine but using the Nubo camera, he was able to catch the vandals.

Buying the Nubo would be 349 Euro + 13 Euro per month ($700 / 2 years)

But tonight! The Hack and Tell special! You can get it all for $540 / 2 years!

Has an SD card in the camera; uses a Verizon Mifi for the hotspot. This battery is much better (12 hours!) than the Nubo battery (2-3 hours)

You can use it to make sure that your skittish kitty is actually using the robotic litterbox before getting rid of the old one!

Travis Hoppe, @metasemantic

Prosody of US names and an iambic poem

Like most projects where you get into the data, the data takes you in different directions than you anticipated.

There’s so much cultural history and all that wrapped up in names – let’s ignore all that!

Prosody - rhythm and sound used in poetry.

Olivia - four syllables; over 18,000 Social Security numbers registered to Olivia in 2017.

Prosodic breaks up the name to get syllablization and gives you the IPA

Only captures the stress of the Anglo/European side

Something happened in the 60s – more syllables! Less “Will” and more “Alejandro”

Dav-id vs. Da-vid

Almost all two-syllable names are trochees; four syllable names

Iambic pentameter names


Create Iambic pentameter names: “Lucille Milan Loretta-Juliet” or “Irene Kareem Sophia-Adelaide”