Chapter 7: Train Your Own Embeddings
The reason Flair sequence taggers yield such outstanding performance can mainly be attributed to its secret sauce – Flair embeddings. Their contextual design, the fact that they are character-based, and the way they can be used in the backward-forward configuration make them a perfect fit for use in sequence labeling tasks. But, up until this point in this book, we haven't focused much on how these embeddings are actually trained.
In the previous chapter, where we covered model training and learned about word embeddings, we simply used the pre-trained embeddings that were available as part of the Flair Python package. But there are many Natural Language Processing (NLP) problems we may stumble upon where the pre-trained embeddings will not be sufficient.
When working with flair, you may find yourself dealing with a language that isn't covered by Flair's pre-trained embeddings yet, or you may simply need embeddings that...