Generating embeddings for Yu-Gi-Oh Cards with NumPy
summary
Builds embeddings from 13,000 cards and >100,000 scraped decks into a 13,000×13,000 cooccurrence matrix (about 5% nonzero), decomposes it with svds SVD to 50 dimensions using NumPy/SciPy, scales and standardizes vectors, stores them in an Annoy index and runs nearest‑neighbor queries. Result: compact card vectors that mostly surface sensible recommendations but sometimes fail (Exodia parts), dataset bias toward competitive staples, code and Kaggle dataset provided.
AI-written summary · about
chapters (11)
- 0:00Yu-Gi-Oh embeddings intro
- 1:53What embeddings are
- 8:34Card co-occurrence embeddings
- 11:48Processing card data
- 14:37Dense vectors with SVD
- 18:11NumPy embeddings implementation
- 20:56Finding similar cards
- 22:42Live card recommendations
- 26:19Embedding quality evaluation
- 27:44Dataset biases and future work
- 28:47Key takeaways
related talks
48mHow I found my Pokémon cards thanks to Postgres: an AI journey
32mDistributed Vector Databases - What, Why, and How
50mA fun and absurd introduction to Vector Databases
50mVector Ops: run vector embedding-powered apps in prod
45mBehind the hype: managing billion-scale embeddings in Elasticsearch and OpenSearch
28m"Embeddings: How Computers Learned to Read"
35mUsing Embedding Layers to Manage High Cardinality Categorical Data
16mDense Vectors in Elasticsearch