This is a fork of https://github.com/joeynmt/joeynmt/
My BA thesis. See BA.pdf
Check RAM.md for current problems, ideas and questions
Issue: Some knowledgebases are empty Solution: Add default dummy entry to all knowledgebases. Would a sensible alternative be to just not forward the kvr_attention for these batches?
Goals were set:
Reimplement the kvr model, use same dataset, achieve/compare/criticize results
Adapt the model for transformer
Depending on if necessary, enrich dataset, pretrain model OOD etc; possibly look at chatbots (probably not)
Other datasets: