Error in adding sentence embeddings to existing index

[anetschka]

Description
I am trying to use haystack with a sentence embedding model from the sentence-transformers library. Calling indexer.update_embeddings(embedding_retriever) results in an AttributeError.

Error message
line 318, in update_embeddings
"doc": {self.embedding_field: emb.tolist()},
AttributeError: 'list' object has no attribute 'tolist'

So something does not work in the handling of the embeddings given by the model.

System:

• Haystack version (commit or version number): cloned yesterday
• DocumentStore: ElasticsearchDocumentStore(host="localhost", username="", password="", index="document", embedding_field="embedding", embedding_dim=768)
• Retriever: EmbeddingRetriever(indexer, model_format='sentence_transformers', embedding_model='distiluse-base-multilingual-cased')

[tholor]

Thanks for reporting @anetschka ! I am able to reproduce this and already found the bug. While the DensePassageRetriever returns a List[ndarray], the EmbeddingRetriever returns List[List[float]]. I will add a fix for this

[anetschka]

Prima! Allerbesten Dank!

[tholor]

Should be fixed with #245. Can you please try to pull & install from latest master?

[anetschka]

I confirm that the bug is fixed, closing the ticket,