This feature enables training a Lightwood predictor without any mixers or subsequent analysis if the embedding_only flag is enabled in the problem definition. The resulting output is the concatenation of encoded representations learnt by each encoder, a.k.a. the "embedding".
Alternatively, normally trained predictors can also return these "embeddings" if the return_embedding argument is set to True.
This feature enables training a Lightwood predictor without any mixers or subsequent analysis if the
embedding_onlyflag is enabled in the problem definition. The resulting output is the concatenation of encoded representations learnt by each encoder, a.k.a. the "embedding".Alternatively, normally trained predictors can also return these "embeddings" if the
return_embeddingargument is set toTrue.