It's Alive! Machine Learning Writes Your Code! 🧠
summary
A machine learning pipeline that generates TypeScript function bodies from a signature, trained on 388,000 TypeScript functions pulled via BigQuery, cleans identifiers and strings into placeholders using AST and tsquery, one-hot encodes tokens, and uses a TensorFlow encoder/decoder LSTM on TPU for training. A small test set of ~5,000 functions trains faster on Colab then a TPU, producing occasionally syntactically valid code but often incorrect or amusing results, still clearly experimental.
AI-written summary · about
chapters (8)
- 0:00Introduction and inspiration
- 3:58Machine learning basics
- 8:10Data collection and cleaning
- 13:50Live coding with AST
- 22:36Preparing model samples
- 27:12Model architecture and training
- 30:03Model evaluation and results
- 32:10Takeaways and next steps
related talks
36mIt's Alive! Machine Learning Writes Your Code! 🧠🔮
35mIt's Alive! Machine Learning Writes Your Code!
37mMachine Learning on Source Code
36mLet Your Angular Code Write Itself
39mMachine Learning Magic for Your JavaScript Application
43mA Deep Dive into Advanced TypeScript: A Live Coding Expedition
43mMachine Learning for Developer Productivity
48mDeep Dive into Advanced TypeScript: A Live Coding Expedition