Technical description for mtg-similarity-1.0
We use google's universal sentence encoder to create an embedding for the provided text. We then compute the cosine simliarity between this embedding and the embeddings for all effects currently in print, and return the top 5 closest embeddings.
We used the AllCards dataset available from mtgjson.com
Data is the same as for mtg-colors.
Training is the same as for mtg-colors.