Vanilla Neural Nets

To better understand neural networks, I’ve decided to implement several from scratch. Currently, this project contains a feedforward neural network with sigmoid activation functions, and both mean squared error and cross-entropy loss functions. Because everything is an object, adding future activation functions, loss functions, and optimization routines should be a breeze. In addition to being highly compose-able, … 

 

Single Neuron Gradient Descent

In my experience, the gap between a conceptual understanding of how a machine learning model “learns” and a concrete, “I can do this with a pencil and paper” understanding is large. This gap is further exacerbated by the nature of popular machine learning libraries which allow you to use powerful models without knowing how they … 

 

dotify: Recommending Spotify Music Through Country Arithmetic

Ever since the release of word2vec I’ve been fascinated with embedding things – words, places, people – into vector space. Though not a mathematical historian, I don’t believe this concept is at all new: matrix factorization methods like Singular Value Decomposition have given us this ability for years. This said, one of the most exciting revelations of word2vec is the remarkably … 

 

So You Want to Implement a Custom Loss Function?

It’s often not so hard. My venerable boss recently took a trip to Amsterdam. As we live in New York City, he needed to board a plane. Days before, we discussed the asymmetric risk of airport punctuality: “if I get to the gate early, it’s really not so bad; if I arrive too late and … 

 

Travel Recommendations with Jaccard Similarities

I recently finished building a web app that recommends travel destinations. You input a country, and it provides you with 5 other countries which you might also enjoy. The recommendations are generated via a basic application of collaborative filtering. In effect, you query for a country, and the engine suggests additional countries enjoyed by other users … 

 

Markovian Muse: Conjuring the Backpacker Within

I haven’t travel-blogged in a while – mainly because I haven’t been traveling. When I was, the writing flowed easily: I’d wander, hike, try and fail, and humbling lessons about the world and its creatures seemed around every corner. There was plenty about which to write. Now, living here in New York City, and spending much of my … 

 

Statistically Cycling Southeast Asia

I spent 6 months of my trip around the world traveling on a bicycle. First, I rode from Istanbul, Turkey to Bishkek, Kyrgyzstan, and then through small parts of China and Laos. While you may think I’m crazy, the other cyclists I met along the way would offer a very different perspective. Why? Because the … 

 

Text-Mining South American Constitutions

With the advent of social media, and the Babylonian deluge of tweets, YouTube comments, and wall posts traversing the internet daily, text-mining has become an increasingly important tool of the data scientist. Basically, all of these tweets, Facebook posts, etc. form a valuable and robust set of data, and text-mining helps us understand what it … 

 

Clustering Continued: A Gaucho on Vacation

In our previous post we chose to cluster South American airports into k = 3 distinct groups. Moving forward, we’ll take a closer look into what this really means. As mentioned previously, the k-means algorithm incorporates some element of mathematical randomness. On one k-means trial, the algorithm may assign 30% of our airports to Cluster … 

 

Clustering South American Airports

Greetings, all, and welcome to my new website! For those that know me, I am the increasingly lazy creator and curator of Will Travel Life – where I post stories, photos, and philosophical muse from a 2+ year backpacking and cycling trip around the world. For those that don’t, it’s a pleasure to have you at …