Playground
Try the ML models
Each demo runs one selected model rebuilt from a larger project that runs within your browser. These are parts of the projects. No text/data is saved.
Feedforward neural network
Kanji Radical Match
Type an English word and see the kanji radicals the neural net links it to.
This runs the trained feedforward network with predictions precomputed.
How it works — The word is looked up in the network's output layer, read back as a probability over kanji radicals, and charted from most to least confident.
Collaborative filtering
Tabelog Recommender
Pick a restaurant and get calculated ratings from item-item collaborative filtering across six categories.
This demo runs item-item collaborative filtering only.
How it works — Each restaurant's category ratings are filled in with item-item cosine similarity, then plotted against the observed means.
Logistic regression
Social Media Sentiment Analysis
Score any text for how inflammatory it reads, live in your browser.
This demo runs on data from the Logistic Regression model.
How it works — The text is tokenized and stemmed. Every known token adds its learned weight, and a sigmoid turns the running total into a 0 to 1 score.