Visualize the lyrical structure of pop songs using self-similarity matrices. (Best experienced on a screen bigger than a phone.)
I trained some restricted Boltzmann machines on short texts and got them to generate new ones. Here are some web apps that let you sample outputs from models trained on a few different domains:
The Hidden Unit Zoo shows visualizations of the hidden units of an RBM model trained on GitHub repos.
Visualizing char-CNN filters
Here are some visualizations of convolutional filters from a neural language model released by Google.
Sentences through the eyes of a language model
See how a language model (the same one as above) assigns probabilities to each word in a sentence. Explore random sentences from:
- The Billion Word Benchmark - a heldout fold of the same corpus the model was trained on
- Brown Corpus - news - the same domain as the training corpus, but from 1961.
- Brown Corpus - romance novels
Tour of Heroes
Tour of Heroes is an incremental game implemented using Angular2. It’s in a rough beta-ish state. I’ll get around to finishing it eventually, maybe.
A visual essay commissioned by The Pudding, exploring repetition in song lyrics with the help of a lossless compression algorithm.
Also, check out this standalone demo of applying the Lempel-Ziv compression algorithm to the lyrics of some pop songs.