Getting error in retriever Exception while running node 'retriever': 'numpy.ndarray' object has no attribute 'unsqueeze'

[Laxmi530]

I am getting error in retriever, also tried several techniques but unable to process further I am stuck in the retriever. I used the below code. Can some one please help me.

Thanks in advance.

from haystack.pipelines import Pipeline
from haystack.nodes import PromptNode
from haystack.nodes import EmbeddingRetriever
from haystack.document_stores import InMemoryDocumentStore
from haystack.utils import clean_wiki_text, convert_files_to_docs
import torch, numpy as np

document_store = InMemoryDocumentStore()
doc_dir = "./data"
docs = convert_files_to_docs(dir_path=doc_dir, clean_func=clean_wiki_text, split_paragraphs=True)
document_store.write_documents(docs)

class ModifiedEmbeddingRetriever(EmbeddingRetriever):
    def encode(self, texts):
        embeddings = super().encode(texts)
        return torch.Tensor(embeddings).unsqueeze(0)

retriever = ModifiedEmbeddingRetriever(document_store=document_store,
                                embedding_model="sentence-transformers/multi-qa-mpnet-base-dot-v1")
node = PromptNode("text-davinci-003", default_prompt_template="question-answering", 
                        api_key="xxxxxxxx") 
pipe = Pipeline()
pipe.add_node(component=retriever, name="retriever", inputs=["Query"])
pipe.add_node(component=node, name="node", inputs=["retriever"])

output = pipe.run(query="what is the name of the player", params={"retriever": {"top_k": 5}}, debug=True)
output["results"]

This is the error i am getting

AttributeError                            Traceback (most recent call last)
File ~\AppData\Roaming\Python\Python39\site-packages\haystack\pipelines\base.py:527, in Pipeline.run(self, query, file_paths, labels, documents, meta, params, debug)
    526     logger.debug("Running node '%s` with input: %s", node_id, node_input)
--> 527     node_output, stream_id = self._run_node(node_id, node_input)
    528 except Exception as e:
    529     # The input might be a really large object with thousands of embeddings.
    530     # If you really want to see it, raise the log level.

File ~\AppData\Roaming\Python\Python39\site-packages\haystack\pipelines\base.py:454, in Pipeline._run_node(self, node_id, node_input)
    453 def _run_node(self, node_id: str, node_input: Dict[str, Any]) -> Tuple[Dict, str]:
--> 454     return self.graph.nodes[node_id]["component"]._dispatch_run(**node_input)

File ~\AppData\Roaming\Python\Python39\site-packages\haystack\nodes\base.py:201, in BaseComponent._dispatch_run(self, **kwargs)
    197 """
    198 The Pipelines call this method when run() is executed. This method in turn executes the _dispatch_run_general()
    199 method with the correct run method.
    200 """
--> 201 return self._dispatch_run_general(self.run, **kwargs)

File ~\AppData\Roaming\Python\Python39\site-packages\haystack\nodes\base.py:245, in BaseComponent._dispatch_run_general(self, run_method, **kwargs)
    243         run_inputs[key] = value
--> 245 output, stream = run_method(**run_inputs, **run_params)
    247 # Collect debug information

File ~\AppData\Roaming\Python\Python39\site-packages\haystack\nodes\retriever\base.py:291, in BaseRetriever.run(self, root_node, query, filters, top_k, documents, index, headers, scale_score)
    290     run_query_timed = self.timing(self.run_query, "query_time")
--> 291     output, stream = run_query_timed(
    292         query=query, filters=filters, top_k=top_k, index=index, headers=headers, scale_score=scale_score
    293     )
    294 elif root_node == "File":

File ~\AppData\Roaming\Python\Python39\site-packages\haystack\nodes\retriever\base.py:111, in BaseRetriever.timing.<locals>.wrapper(*args, **kwargs)
    110 tic = perf_counter()
--> 111 ret = fn(*args, **kwargs)
    112 toc = perf_counter()

File ~\AppData\Roaming\Python\Python39\site-packages\haystack\nodes\retriever\base.py:345, in BaseRetriever.run_query(self, query, filters, top_k, index, headers, scale_score)
    336 def run_query(
    337     self,
    338     query: str,
   (...)
    343     scale_score: Optional[bool] = None,
    344 ):
--> 345     documents = self.retrieve(
...
    535 queue.pop(node_id)
    536 #

Exception: Exception while running node 'retriever': 'numpy.ndarray' object has no attribute 'unsqueeze'
Enable debug logging to see the data that was passed when the pipeline failed.

[bilgeyucel]

Hi @Laxmi530, I think the functionality you're looking for is Shaper rather than customizing the Retriever. Shaper comes with different functions to convert data in a form you would need for PromptNode to function correctly.
In this pipeline, Shaper with join_documents function takes a list of documents from retriever and merges them into a list containing a single document. 👇

shaper = Shaper(func="join_documents", inputs={"documents": "documents"}, outputs=["documents"])

retriever = EmbeddingRetriever(document_store=document_store, embedding_model="sentence-transformers/multi-qa-mpnet-base-dot-v1")
node = PromptNode("text-davinci-003", default_prompt_template="question-answering", api_key="xxxxxxxx")

pipe = Pipeline()
pipe.add_node(component=retriever, name="retriever", inputs=["Query"])
pipe.add_node(component=shaper, name="shaper", inputs=["retriever"])
pipe.add_node(component=node, name="prompt_node", inputs=["shaper"])

output = pipe.run(query="what is the name of the player", params={"retriever": {"top_k": 5}}, debug=True)
output["results"]