WebRetriver bug in Google API

[HGamalElDin]

Hello guys! I'm trying to run a WebRetriever RAG pipeline, but the webretriever with google doesn't work anyway!

Here's a code snippet I'm trying,

pn = PromptNode(
    "gpt-3.5-turbo",
    api_key="my_Key",
    max_length=256,
    default_prompt_template="question-answering-with-document-scores",
)
web_retriever = WebRetriever(search_engine_provider="GoogleAPI", api_key="my_key", top_search_results=5, top_k=3, search_engine_kwargs={"engine_id": "engineID"})
pipeline = WebQAPipeline(retriever=web_retriever, prompt_node=pn)

I got the error: TypeError: WebRetriever.init() got an unexpected keyword argument 'search_engine_kwargs'

When I remove this argument and try to run the pipeline, I get Exception: Exception while running node 'Retriever': You need to provide an engine ID for the Google API. See https://developers.google.com/custom-search/v1/overview
Enable debug logging to see the data that was passed when the pipeline failed.

I tried this snippet in both haystack versions, 1.18.1, and 1.21 and the same results showing.
Any help!

[anakin87]

I think this should work.
(Unfortunately, it is undocumented and convoluted.)

from haystack.nodes.search_engine.providers import GoogleAPI

# see https://github.com/deepset-ai/haystack/blob/main/haystack/nodes/search_engine/providers.py#L305
search_engine = GoogleAPI (api_key="my_key", top_k=5, engine_id="engineID")

web_retriever = WebRetriever(api_key="my_key", search_engine_provider=search_engine, top_k=3)

pn = PromptNode(
    "gpt-3.5-turbo",
    api_key="my_Key",
    max_length=256,
    default_prompt_template="question-answering-with-document-scores",
)

pipeline = WebQAPipeline(retriever=web_retriever, prompt_node=pn)

[HGamalElDin]

I think this should work. (Unfortunately, it is undocumented and convoluted.)

from haystack.nodes.search_engine.providers import GoogleAPI

# see https://github.com/deepset-ai/haystack/blob/main/haystack/nodes/search_engine/providers.py#L305
search_engine = GoogleAPI (api_key="my_key", top_k=5, engine_id="engineID")

web_retriever = WebRetriever(api_key="my_key", search_engine_provider=search_engine, top_k=3)

pn = PromptNode(
    "gpt-3.5-turbo",
    api_key="my_Key",
    max_length=256,
    default_prompt_template="question-answering-with-document-scores",
)

pipeline = WebQAPipeline(retriever=web_retriever, prompt_node=pn)

Thank You so much @anakin87, Yes unfortunately it's not well documented. Your solution worked for me only if I set the the mode to "preprocessed_documents", but If I left it's default value "snippet" It throws this error that I also cannot find and ref for:
Exception: Exception while running node 'Shaper': 'score'
Enable debug logging to see the data that was passed when the pipeline failed.

[anakin87]

My intuition is that the prompt template question-answering-with-document-scores expects the Document scores.
If they are missing, it fails.

You can use a different prompt template.
This blog article may help you.

(@dfokina as you can understand from this issue, the example about GoogleAPI in our docs is wrong.)

[HGamalElDin]

My intuition is that the prompt template question-answering-with-document-scores expects the Document scores. If they are missing, it fails.

You can use a different prompt template. This blog article may help you.

(@dfokina as you can understand from this issue, the example about GoogleAPI in our docs is wrong.)

Honestly no! I tried a custom prompt template and it throws same error! It seams the issue from the WebQAPipeline itself. It somehow includes a default shaper.

The question is, what is the key difference between WebRetriever and WebSearch classes?

[bilgeyucel]

The main issue is solved on the main branch and will be released with Haystack 1.22.0 👍

However, I can reproduce the issue about Shaper on main:
Exception: Exception while running node 'Shaper': 'score'
Enable debug logging to see the data that was passed when the pipeline failed.

This is a problem when snippets mode is used. "raw_documents" and "preprocessed_documents" are working fine. If GoogleAPI doesn't support the snippets mode, we need to update the documentation.

Here's the colab notebook to reproduce it: https://colab.research.google.com/drive/1otbIATmR_dh2RtTry7sYgF-JhasHS61B?usp=sharing