haystack/tutorials/Tutorial6_Better_Retrieval_via_DPR.py
kolk f2b6cc761b
Refactor DPR from FB to Transformers codebase (#308)
* change_HFBertEncoder to transformers DPREncoder

* Removed BertTensorizer

* model download relative path

* Refactor model load

* Tutorial5 DPR updated

* fix print_eval_results typo

* copy transformers DPR modules in dpr_utils and test

* transformer v3.0.2 import errors fixed

* remove dependency of DPRConfig on attribute use_return_tuple

* Adjust transformers 302 locally to work with dpr

* projection layer removed from DPR encoders

* fixed mypy errors

* transformers DPR compatible code added

* transformers DPR compatibility added

* bug fix in tutorial 6 notebook

* Docstring update and variable naming issues fix

* tutorial modified to reflect DPR variable naming change

* title addition to passage use-cases handled

* modified handling untitled batch

* resolved mypy errors

* typos in docstrings and comments fixed

* cleaned DPR code and added new test cases

* warnings added for non-bert model [SEP] token removal

* changed warning to logger warning

* title mask creation refactored

* bug fix on cuda issues

* tutorial 6 instantiates modified DPR

* tutorial 5 modified

* tutorial 5 ipython notebook modified: DPR instantiation

* batch_size added to DPR instantiation

* tutorial 5 jupyter notebook typos fixed

* improved docstrings, fixed typos

* Update docstring

Co-authored-by: Timo Moeller <timo.moeller@deepset.ai>
Co-authored-by: Malte Pietsch <malte.pietsch@deepset.ai>
2020-08-25 20:16:00 +05:30

62 lines
3.0 KiB
Python
Executable File

from haystack import Finder
from haystack.database.faiss import FAISSDocumentStore
from haystack.indexing.cleaning import clean_wiki_text
from haystack.indexing.utils import convert_files_to_dicts, fetch_archive_from_http
from haystack.reader.farm import FARMReader
from haystack.utils import print_answers
from haystack.retriever.dense import DensePassageRetriever
# FAISS is a library for efficient similarity search on a cluster of dense vectors.
# The FAISSDocumentStore uses a SQL(SQLite in-memory be default) database under-the-hood
# to store the document text and other meta data. The vector embeddings of the text are
# indexed on a FAISS Index that later is queried for searching answers.
document_store = FAISSDocumentStore()
# ## Cleaning & indexing documents
# Let's first get some documents that we want to query
doc_dir = "data/article_txt_got"
s3_url = "https://s3.eu-central-1.amazonaws.com/deepset.ai-farm-qa/datasets/documents/wiki_gameofthrones_txt.zip"
fetch_archive_from_http(url=s3_url, output_dir=doc_dir)
# convert files to dicts containing documents that can be indexed to our datastore
dicts = convert_files_to_dicts(dir_path=doc_dir, clean_func=clean_wiki_text, split_paragraphs=True)
# Now, let's write the docs to our DB.
document_store.write_documents(dicts)
### Retriever
retriever = DensePassageRetriever(document_store=document_store,
query_embedding_model="facebook/dpr-question_encoder-single-nq-base",
passage_embedding_model="facebook/dpr-ctx_encoder-single-nq-base",
use_gpu=True,
embed_title=True,
remove_sep_tok_from_untitled_passages=True)
# Important:
# Now that after we have the DPR initialized, we need to call update_embeddings() to iterate over all
# previously indexed documents and update their embedding representation.
# While this can be a time consuming operation (depending on corpus size), it only needs to be done once.
# At query time, we only need to embed the query and compare it the existing doc embeddings which is very fast.
document_store.update_embeddings(retriever)
### Reader
# Load a local model or any of the QA models on
# Hugging Face's model hub (https://huggingface.co/models)
reader = FARMReader(model_name_or_path="deepset/roberta-base-squad2", use_gpu=True)
### Finder
# The Finder sticks together reader and retriever in a pipeline to answer our actual questions.
finder = Finder(reader, retriever)
### Voilà! Ask a question!
# You can configure how many candidates the reader and retriever shall return
# The higher top_k_retriever, the better (but also the slower) your answers.
prediction = finder.get_answers(question="Who is the father of Arya Stark?", top_k_retriever=10, top_k_reader=5)
# prediction = finder.get_answers(question="Who created the Dothraki vocabulary?", top_k_reader=5)
# prediction = finder.get_answers(question="Who is the sister of Sansa?", top_k_reader=5)
print_answers(prediction, details="minimal")