Retriever only finds first ever inserted document?

[sorenwacker]

I was following this https://docs.pinecone.io/docs/haystack#initializing-the-pineconedocumentstore instructions, only that I added a single almost empty doc in the beginning. I then added more docs, step by step to the index.

Then retriever only ever finds the first document, not the ones added on a later stage. Am I missing something?

#!/usr/bin/env python
# coding: utf-8

# In[1]:


import os
import logging

OPENAI_API_KEY = os.getenv("OPENAI_API_KEY")
PINECONE_API_KEY = os.getenv("PINECONE_API_KEY")

len(OPENAI_API_KEY), len(PINECONE_API_KEY)


# # Pinecone vector store with haystack

# In[2]:


from haystack.document_stores import PineconeDocumentStore

document_store = PineconeDocumentStore(
    api_key=PINECONE_API_KEY,
    index='haystack-lfqa',
    similarity="cosine",
    embedding_dim=384,
    environment = "asia-southeast1-gcp-free"
)


# In[3]:


document_store.get_document_count()


# In[4]:


basic = """
facts 1
"""


# In[19]:


basics_2 = """facts 2"""


# In[21]:


from haystack import Document
docs = []

doc = Document(
    content=basic,
    meta={
        "title": "Introduction of races I",
        'context': 'context'
    }
)

docs.append(doc)

doc = Document(
    content=basics_2,
    meta={
        "title": "Introduction of races I",
        'context': 'context'
    }
)

docs.append(doc)


# In[22]:


docs


# In[23]:


document_store.write_documents(docs)


# In[24]:


from haystack.nodes import EmbeddingRetriever

retriever = EmbeddingRetriever(
    document_store=document_store,
    embedding_model="multi-qa-MiniLM-L6-cos-v1",
    model_format="sentence_transformers"
)


# In[25]:


document_store.update_embeddings(
    retriever,
    batch_size=16
)


# In[26]:


from haystack.nodes import FARMReader

reader = FARMReader(
    model_name_or_path='deepset/electra-base-squad2', 
    use_gpu=True
)


# In[27]:


from haystack.pipelines import ExtractiveQAPipeline

pipe = ExtractiveQAPipeline(reader, retriever)


# In[28]:


from haystack.utils import print_answers

query = "What races are in aventuria?"
# get the answer
answer = pipe.run(
    query=query,
    params={
        "Retriever": {"top_k": 1},
    }
)
# print the answer(s)
print_answers(answer)


# In[29]:


query = "What is aventuria?"
# get the answer
answer = pipe.run(
    query=query,
    params={
        "Retriever": {"top_k": 10},
    }
)
# print the answer(s)
print_answers(answer)

'Query: What is aventuria?'
'Answers:'
[   <Answer {'answer': 'content', 'type': 'extractive', 'score': 1.604350472916849e-05, 'context': 'content', 'offsets_in_document': [{'start': 0, 'end': 7}], 'offsets_in_context': [{'start': 0, 'end': 7}], 'document_ids': ['8643a38b84ae409d5a8763734349fda0'], 'meta': {'context': 'context', 'title': 'title'}}>]

'content' was the content of the first item that i had created.

[vblagoje]

What happens if you increase top_k?

[sorenwacker]

Not sure, what happened here. It seems after waiting a bit, the instances are found. Maybe its just a delay on the pinecone end.