Charting the English Language...in pure Clojure
summary
Reproduces Turian’s word map of Collobert and Weston from the Wikipedia corpus using SENNA embeddings, a Clojure reimplementation of t-SNE and pure data plumbing into SVG; uses pure Clojure with core.matrix, deals with numpy/JSON interop and Python idiosyncrasies via a generative test harness that fuzz-compares Clojure results to Python, implements ULP-based fuzzy equality for doubles, renders ~2,500–3,000 words reliably, scales to ~36,502 words but is slow (~2 hours) and not as fast as C++.
AI-written summary · about
chapters (9)
- 0:00Introduction
- 3:11Machine Learning
- 6:56What went wrong
- 14:43Python interop
- 15:45Python bash
- 17:22Generator
- 18:14Fuzzy equals
- 20:25Closure
- 22:50Data Visualization
related talks
43mA Case Study of Natural Language Processing in Clojure
38mScientific Clojure, a bird's eye view
30mWolframite: Bring computational intelligence to Clojure
33mBuilding Conversational Speech Annotation Tool in Clojure
33mScalably Internationalizing Millions of Latent Semantic Labels
38mBuilding a Grammar for Statistical Graphics in Clojure
44mSome tentative first steps towards a Star Trek universal communicator
39mCrafting Artisanal Vector Graphics