def tutorial4_faq_style_qa():
## "FAQ-Style QA": Utilizing existing FAQs for Question Answering
# While *extractive Question Answering* works on pure texts and is therefore more generalizable, there's also a common alternative that utilizes existing FAQ data.
#
# Pros:
# - Very fast at inference time
# - Utilize existing FAQ data
# - Quite good control over answers
#
# Cons:
# - Generalizability: We can only answer questions that are similar to existing ones in FAQ
#
# In some use cases, a combination of extractive QA and FAQ-style can also be an interesting option.
launch_es()
### Init the DocumentStore
# In contrast to Tutorial 1 (extractive QA), we:
#
# * specify the name of our `text_field` in Elasticsearch that we want to return as an answer
# * specify the name of our `embedding_field` in Elasticsearch where we'll store the embedding of our question and that is used later for calculating our similarity to the incoming user question
# * set `excluded_meta_data=["question_emb"]` so that we don't return the huge embedding vectors in our search results
document_store = ElasticsearchDocumentStore(
host="localhost",
username="",
password="",
index="document",
embedding_field="question_emb",
embedding_dim=768,
excluded_meta_data=["question_emb"],
similarity="cosine")
### Create a Retriever using embeddings
# Instead of retrieving via Elasticsearch's plain BM25, we want to use vector similarity of the questions (user question vs. FAQ ones).
# We can use the `EmbeddingRetriever` for this purpose and specify a model that we use for the embeddings.
#
retriever = EmbeddingRetriever(document_store=document_store,
embedding_model="deepset/sentence_bert",
use_gpu=True)
# Download a csv containing some FAQ data
# Here: Some question-answer pairs related to COVID-19
temp = requests.get(
"https://raw.githubusercontent.com/deepset-ai/COVID-QA/master/data/faqs/faq_covidbert.csv"
)
open('small_faq_covid.csv', 'wb').write(temp.content)
# Get dataframe with columns "question", "answer" and some custom metadata
df = pd.read_csv("small_faq_covid.csv")
# Minimal cleaning
df.fillna(value="", inplace=True)
df["question"] = df["question"].apply(lambda x: x.strip())
print(df.head())
# Get embeddings for our questions from the FAQs
questions = list(df["question"].values)
df["question_emb"] = retriever.embed_queries(texts=questions)
df = df.rename(columns={"question": "text"})
# Convert Dataframe to list of dicts and index them in our DocumentStore
docs_to_index = df.to_dict(orient="records")
document_store.write_documents(docs_to_index)
# Initialize a Pipeline (this time without a reader) and ask questions
from haystack.pipeline import FAQPipeline
pipe = FAQPipeline(retriever=retriever)
prediction = pipe.run(query="How is the virus spreading?",
top_k_retriever=10)
print_answers(prediction, details="all")
from haystack.question_generator import QuestionGenerator
from haystack.utils import launch_es
from haystack.document_store import ElasticsearchDocumentStore
from haystack.retriever import ElasticsearchRetriever
from pprint import pprint
from haystack.reader import FARMReader
from tqdm import tqdm
from haystack.pipeline import QuestionGenerationPipeline, RetrieverQuestionGenerationPipeline, QuestionAnswerGenerationPipeline
"""
This is a bare bones tutorial showing what is possible with the QuestionGenerator Node which automatically generates
questions which the model thinks can be answered by a given document.
"""
# Start Elasticsearch service via Docker
launch_es()
text1 = "Python is an interpreted, high-level, general-purpose programming language. Created by Guido van Rossum and first released in 1991, Python's design philosophy emphasizes code readability with its notable use of significant whitespace."
text2 = "Princess Arya Stark is the third child and second daughter of Lord Eddard Stark and his wife, Lady Catelyn Stark. She is the sister of the incumbent Westerosi monarchs, Sansa, Queen in the North, and Brandon, King of the Andals and the First Men. After narrowly escaping the persecution of House Stark by House Lannister, Arya is trained as a Faceless Man at the House of Black and White in Braavos, using her abilities to avenge her family. Upon her return to Westeros, she exacts retribution for the Red Wedding by exterminating the Frey male line."
text3 = "Dry Cleaning are an English post-punk band who formed in South London in 2018.[3] The band is composed of vocalist Florence Shaw, guitarist Tom Dowse, bassist Lewis Maynard and drummer Nick Buxton. They are noted for their use of spoken word primarily in lieu of sung vocals, as well as their unconventional lyrics. Their musical stylings have been compared to Wire, Magazine and Joy Division.[4] The band released their debut single, 'Magic of Meghan' in 2019. Shaw wrote the song after going through a break-up and moving out of her former partner's apartment the same day that Meghan Markle and Prince Harry announced they were engaged.[5] This was followed by the release of two EPs that year: Sweet Princess in August and Boundary Road Snacks and Drinks in October. The band were included as part of the NME 100 of 2020,[6] as well as DIY magazine's Class of 2020.[7] The band signed to 4AD in late 2020 and shared a new single, 'Scratchcard Lanyard'.[8] In February 2021, the band shared details of their debut studio album, New Long Leg. They also shared the single 'Strong Feelings'.[9] The album, which was produced by John Parish, was released on 2 April 2021.[10]"
docs = [{"text": text1}, {"text": text2}, {"text": text3}]
# Initialize document store and write in the documents
document_store = ElasticsearchDocumentStore()
document_store.write_documents(docs)
# Initialize Question Generator
question_generator = QuestionGenerator()
"""
The most basic version of a question generator pipeline takes a document as input and outputs generated questions
which the the document can answer.
def tutorial5_evaluation():
##############################################
# Settings
##############################################
# Choose from Evaluation style from ['retriever_closed', 'reader_closed', 'retriever_reader_open']
# 'retriever_closed' - evaluates only the retriever, based on whether the gold_label document is retrieved.
# 'reader_closed' - evaluates only the reader in a closed domain fashion i.e. the reader is given one query
# and one document and metrics are calculated on whether the right position in this text is selected by
# the model as the answer span (i.e. SQuAD style)
# 'retriever_reader_open' - evaluates retriever and reader in open domain fashion i.e. a document is considered
# correctly retrieved if it contains the answer string within it. The reader is evaluated based purely on the
# predicted string, regardless of which document this came from and the position of the extracted span.
style = "retriever_reader_open"
# make sure these indices do not collide with existing ones, the indices will be wiped clean before data is inserted
doc_index = "tutorial5_docs"
label_index = "tutorial5_labels"
##############################################
# Code
##############################################
launch_es()
device, n_gpu = initialize_device_settings(use_cuda=True)
# Download evaluation data, which is a subset of Natural Questions development set containing 50 documents
doc_dir = "../data/nq"
s3_url = "https://s3.eu-central-1.amazonaws.com/deepset.ai-farm-qa/datasets/nq_dev_subset_v2.json.zip"
fetch_archive_from_http(url=s3_url, output_dir=doc_dir)
# Connect to Elasticsearch
document_store = ElasticsearchDocumentStore(host="localhost",
username="",
password="",
index="document",
create_index=False,
embedding_field="emb",
embedding_dim=768,
excluded_meta_data=["emb"])
# Add evaluation data to Elasticsearch document store
# We first delete the custom tutorial indices to not have duplicate elements
# and also split our documents into shorter passages using the PreProcessor
preprocessor = PreProcessor(split_by="word",
split_length=500,
split_overlap=0,
split_respect_sentence_boundary=False,
clean_empty_lines=False,
clean_whitespace=False)
document_store.delete_all_documents(index=doc_index)
document_store.delete_all_documents(index=label_index)
document_store.add_eval_data(filename="../data/nq/nq_dev_subset_v2.json",
doc_index=doc_index,
label_index=label_index,
preprocessor=preprocessor)
# Let's prepare the labels that we need for the retriever and the reader
labels = document_store.get_all_labels_aggregated(index=label_index)
# Initialize Retriever
retriever = ElasticsearchRetriever(document_store=document_store)
# Alternative: Evaluate DensePassageRetriever
# Note, that DPR works best when you index short passages < 512 tokens as only those tokens will be used for the embedding.
# Here, for nq_dev_subset_v2.json we have avg. num of tokens = 5220(!).
# DPR still outperforms Elastic's BM25 by a small margin here.
# 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)
# document_store.update_embeddings(retriever, index=doc_index)
# Initialize Reader
reader = FARMReader(model_name_or_path="deepset/roberta-base-squad2",
top_k=4,
return_no_answer=True)
# Here we initialize the nodes that perform evaluation
eval_retriever = EvalDocuments()
eval_reader = EvalAnswers(
sas_model="sentence-transformers/paraphrase-multilingual-mpnet-base-v2"
)
## Evaluate Retriever on its own in closed domain fashion
if style == "retriever_closed":
retriever_eval_results = retriever.eval(top_k=10,
label_index=label_index,
doc_index=doc_index)
## Retriever Recall is the proportion of questions for which the correct document containing the answer is
## among the correct documents
print("Retriever Recall:", retriever_eval_results["recall"])
## Retriever Mean Avg Precision rewards retrievers that give relevant documents a higher rank
print("Retriever Mean Avg Precision:", retriever_eval_results["map"])
# Evaluate Reader on its own in closed domain fashion (i.e. SQuAD style)
elif style == "reader_closed":
reader_eval_results = reader.eval(document_store=document_store,
device=device,
label_index=label_index,
doc_index=doc_index)
# Evaluation of Reader can also be done directly on a SQuAD-formatted file without passing the data to Elasticsearch
#reader_eval_results = reader.eval_on_file("../data/nq", "nq_dev_subset_v2.json", device=device)
## Reader Top-N-Accuracy is the proportion of predicted answers that match with their corresponding correct answer
print("Reader Top-N-Accuracy:", reader_eval_results["top_n_accuracy"])
## Reader Exact Match is the proportion of questions where the predicted answer is exactly the same as the correct answer
print("Reader Exact Match:", reader_eval_results["EM"])
## Reader F1-Score is the average overlap between the predicted answers and the correct answers
print("Reader F1-Score:", reader_eval_results["f1"])
# Evaluate combination of Reader and Retriever in open domain fashion
elif style == "retriever_reader_open":
# Here is the pipeline definition
p = Pipeline()
p.add_node(component=retriever, name="ESRetriever", inputs=["Query"])
p.add_node(component=eval_retriever,
name="EvalDocuments",
inputs=["ESRetriever"])
p.add_node(component=reader, name="QAReader", inputs=["EvalDocuments"])
p.add_node(component=eval_reader,
name="EvalAnswers",
inputs=["QAReader"])
results = []
for l in labels:
res = p.run(
query=l.question,
top_k_retriever=10,
labels=l,
top_k_reader=10,
index=doc_index,
)
results.append(res)
eval_retriever.print()
print()
retriever.print_time()
print()
eval_reader.print(mode="reader")
print()
reader.print_time()
print()
eval_reader.print(mode="pipeline")
else:
raise ValueError(
f'style={style} is not a valid option. Choose from retriever_closed, reader_closed, retriever_reader_open'
)
def get_document_store(document_store_type,
similarity='dot_product',
index="document"):
""" TODO This method is taken from test/conftest.py but maybe should be within Haystack.
Perhaps a class method of DocStore that just takes string for type of DocStore"""
if document_store_type == "sql":
if os.path.exists("haystack_test.db"):
os.remove("haystack_test.db")
document_store = SQLDocumentStore(url="sqlite:///haystack_test.db")
assert document_store.get_document_count() == 0
elif document_store_type == "memory":
document_store = InMemoryDocumentStore()
elif document_store_type == "elasticsearch":
launch_es()
# make sure we start from a fresh index
client = Elasticsearch()
client.indices.delete(index='haystack_test*', ignore=[404])
document_store = ElasticsearchDocumentStore(index="eval_document",
similarity=similarity,
timeout=3000)
elif document_store_type in ("milvus_flat", "milvus_hnsw"):
launch_milvus()
if document_store_type == "milvus_flat":
index_type = IndexType.FLAT
index_param = None
search_param = None
elif document_store_type == "milvus_hnsw":
index_type = IndexType.HNSW
index_param = {"M": 64, "efConstruction": 80}
search_param = {"ef": 20}
document_store = MilvusDocumentStore(similarity=similarity,
index_type=index_type,
index_param=index_param,
search_param=search_param,
index=index)
assert document_store.get_document_count(index="eval_document") == 0
elif document_store_type in ("faiss_flat", "faiss_hnsw"):
if document_store_type == "faiss_flat":
index_type = "Flat"
elif document_store_type == "faiss_hnsw":
index_type = "HNSW"
status = subprocess.run(['docker rm -f haystack-postgres'], shell=True)
time.sleep(1)
status = subprocess.run([
'docker run --name haystack-postgres -p 5432:5432 -e POSTGRES_PASSWORD=password -d postgres'
],
shell=True)
time.sleep(6)
status = subprocess.run([
'docker exec haystack-postgres psql -U postgres -c "CREATE DATABASE haystack;"'
],
shell=True)
time.sleep(1)
document_store = FAISSDocumentStore(
sql_url="postgresql://*****:*****@localhost:5432/haystack",
faiss_index_factory_str=index_type,
similarity=similarity,
index=index)
assert document_store.get_document_count() == 0
elif document_store_type in ("opensearch_flat", "opensearch_hnsw"):
launch_opensearch()
if document_store_type == "opensearch_flat":
index_type = "flat"
elif document_store_type == "opensearch_hnsw":
index_type = "hnsw"
document_store = OpenSearchDocumentStore(index_type=index_type,
timeout=3000)
else:
raise Exception(
f"No document store fixture for '{document_store_type}'")
return document_store
def tutorial14_query_classifier():
#Download and prepare data - 517 Wikipedia articles for Game of Thrones
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
got_dicts = convert_files_to_dicts(
dir_path=doc_dir,
clean_func=clean_wiki_text,
split_paragraphs=True
)
# Initialize DocumentStore and index documents
launch_es()
document_store = ElasticsearchDocumentStore()
document_store.delete_all_documents()
document_store.write_documents(got_dicts)
# Initialize Sparse retriever
es_retriever = ElasticsearchRetriever(document_store=document_store)
# Initialize dense retriever
dpr_retriever = DensePassageRetriever(document_store)
document_store.update_embeddings(dpr_retriever, update_existing_embeddings=False)
reader = FARMReader(model_name_or_path="deepset/roberta-base-squad2")
# Here we build the pipeline
sklearn_keyword_classifier = Pipeline()
sklearn_keyword_classifier.add_node(component=SklearnQueryClassifier(), name="QueryClassifier", inputs=["Query"])
sklearn_keyword_classifier.add_node(component=dpr_retriever, name="DPRRetriever", inputs=["QueryClassifier.output_1"])
sklearn_keyword_classifier.add_node(component=es_retriever, name="ESRetriever", inputs=["QueryClassifier.output_2"])
sklearn_keyword_classifier.add_node(component=reader, name="QAReader", inputs=["ESRetriever", "DPRRetriever"])
sklearn_keyword_classifier.draw("pipeline_classifier.png")
# Run only the dense retriever on the full sentence query
res_1 = sklearn_keyword_classifier.run(
query="Who is the father of Arya Stark?",
top_k_retriever=10
)
print("DPR Results" + "\n" + "="*15)
print_answers(res_1)
# Run only the sparse retriever on a keyword based query
res_2 = sklearn_keyword_classifier.run(
query="arya stark father",
top_k_retriever=10
)
print("ES Results" + "\n" + "="*15)
print_answers(res_2)
# Run only the dense retriever on the full sentence query
res_3 = sklearn_keyword_classifier.run(
query="which country was jon snow filmed ?",
top_k_retriever=10
)
print("DPR Results" + "\n" + "="*15)
print_answers(res_3)
# Run only the sparse retriever on a keyword based query
res_4 = sklearn_keyword_classifier.run(
query="jon snow country",
top_k_retriever=10
)
print("ES Results" + "\n" + "="*15)
print_answers(res_4)
# Run only the dense retriever on the full sentence query
res_5 = sklearn_keyword_classifier.run(
query="who are the younger brothers of arya stark ?",
top_k_retriever=10
)
print("DPR Results" + "\n" + "="*15)
print_answers(res_5)
# Run only the sparse retriever on a keyword based query
res_6 = sklearn_keyword_classifier.run(
query="arya stark younger brothers",
top_k_retriever=10
)
print("ES Results" + "\n" + "="*15)
print_answers(res_6)
# Here we build the pipeline
transformer_keyword_classifier = Pipeline()
transformer_keyword_classifier.add_node(component=TransformersQueryClassifier(), name="QueryClassifier", inputs=["Query"])
transformer_keyword_classifier.add_node(component=dpr_retriever, name="DPRRetriever", inputs=["QueryClassifier.output_1"])
transformer_keyword_classifier.add_node(component=es_retriever, name="ESRetriever", inputs=["QueryClassifier.output_2"])
transformer_keyword_classifier.add_node(component=reader, name="QAReader", inputs=["ESRetriever", "DPRRetriever"])
transformer_keyword_classifier.draw("pipeline_classifier.png")
# Run only the dense retriever on the full sentence query
res_1 = transformer_keyword_classifier.run(
query="Who is the father of Arya Stark?",
top_k_retriever=10
)
print("DPR Results" + "\n" + "="*15)
print_answers(res_1)
# Run only the sparse retriever on a keyword based query
res_2 = transformer_keyword_classifier.run(
query="arya stark father",
top_k_retriever=10
)
print("ES Results" + "\n" + "="*15)
print_answers(res_2)
# Run only the dense retriever on the full sentence query
res_3 = transformer_keyword_classifier.run(
query="which country was jon snow filmed ?",
top_k_retriever=10
)
print("DPR Results" + "\n" + "="*15)
print_answers(res_3)
# Run only the sparse retriever on a keyword based query
res_4 = transformer_keyword_classifier.run(
query="jon snow country",
top_k_retriever=10
)
print("ES Results" + "\n" + "="*15)
print_answers(res_4)
# Run only the dense retriever on the full sentence query
res_5 = transformer_keyword_classifier.run(
query="who are the younger brothers of arya stark ?",
top_k_retriever=10
)
print("DPR Results" + "\n" + "="*15)
print_answers(res_5)
# Run only the sparse retriever on a keyword based query
res_6 = transformer_keyword_classifier.run(
query="arya stark younger brothers",
top_k_retriever=10
)
print("ES Results" + "\n" + "="*15)
print_answers(res_6)
# Here we build the pipeline
transformer_question_classifier = Pipeline()
transformer_question_classifier.add_node(component=dpr_retriever, name="DPRRetriever", inputs=["Query"])
transformer_question_classifier.add_node(component=TransformersQueryClassifier(model_name_or_path="shahrukhx01/question-vs-statement-classifier"), name="QueryClassifier", inputs=["DPRRetriever"])
transformer_question_classifier.add_node(component=reader, name="QAReader", inputs=["QueryClassifier.output_1"])
transformer_question_classifier.draw("question_classifier.png")
# Run only the QA reader on the question query
res_1 = transformer_question_classifier.run(
query="Who is the father of Arya Stark?",
top_k_retriever=10
)
print("DPR Results" + "\n" + "="*15)
print_answers(res_1)
# Show only DPR results
res_2 = transformer_question_classifier.run(
query="Arya Stark was the daughter of a Lord.",
top_k_retriever=10
)
print("ES Results" + "\n" + "="*15)
res_2
# Here we create the keyword vs question/statement query classifier
queries = ["arya stark father","jon snow country",
"who is the father of arya stark","which country was jon snow filmed?"]
keyword_classifier = TransformersQueryClassifier()
for query in queries:
result = keyword_classifier.run(query=query)
if result[1] == "output_1":
category = "question/statement"
else:
category = "keyword"
print(f"Query: {query}, raw_output: {result}, class: {category}")
# Here we create the question vs statement query classifier
queries = ["Lord Eddard was the father of Arya Stark.","Jon Snow was filmed in United Kingdom.",
"who is the father of arya stark?","Which country was jon snow filmed in?"]
question_classifier = TransformersQueryClassifier(model_name_or_path="shahrukhx01/question-vs-statement-classifier")
for query in queries:
result = question_classifier.run(query=query)
if result[1] == "output_1":
category = "question"
else:
category = "statement"
print(f"Query: {query}, raw_output: {result}, class: {category}")
def tutorial1_basic_qa_pipeline():
logger = logging.getLogger(__name__)
# ## Document Store
#
# Haystack finds answers to queries within the documents stored in a `DocumentStore`. The current implementations of
# `DocumentStore` include `ElasticsearchDocumentStore`, `FAISSDocumentStore`, `SQLDocumentStore`, and `InMemoryDocumentStore`.
#
# **Here:** We recommended Elasticsearch as it comes preloaded with features like full-text queries, BM25 retrieval,
# and vector storage for text embeddings.
# **Alternatives:** If you are unable to setup an Elasticsearch instance, then follow the Tutorial 3
# for using SQL/InMemory document stores.
# **Hint**:
# This tutorial creates a new document store instance with Wikipedia articles on Game of Thrones. However, you can
# configure Haystack to work with your existing document stores.
#
# Start an Elasticsearch server
# You can start Elasticsearch on your local machine instance using Docker. If Docker is not readily available in
# your environment (eg., in Colab notebooks), then you can manually download and execute Elasticsearch from source.
launch_es()
# Connect to Elasticsearch
document_store = ElasticsearchDocumentStore(host="localhost",
username="",
password="",
index="document")
# ## Preprocessing of documents
#
# Haystack provides a customizable pipeline for:
# - converting files into texts
# - cleaning texts
# - splitting texts
# - writing them to a Document Store
# In this tutorial, we download Wikipedia articles about Game of Thrones, apply a basic cleaning function, and add
# them in Elasticsearch.
# Let's first fetch some documents that we want to query
# Here: 517 Wikipedia articles for Game of Thrones
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)
# You can optionally supply a cleaning function that is applied to each doc (e.g. to remove footers)
# It must take a str as input, and return a str.
# Now, let's write the docs to our DB.
document_store.write_documents(dicts)
# ## Initalize Retriever, Reader, & Finder
#
# ### Retriever
#
# Retrievers help narrowing down the scope for the Reader to smaller units of text where a given question
# could be answered.
#
# They use some simple but fast algorithm.
# **Here:** We use Elasticsearch's default BM25 algorithm
# **Alternatives:**
# - Customize the `ElasticsearchRetriever`with custom queries (e.g. boosting) and filters
# - Use `EmbeddingRetriever` to find candidate documents based on the similarity of
# embeddings (e.g. created via Sentence-BERT)
# - Use `TfidfRetriever` in combination with a SQL or InMemory Document store for simple prototyping and debugging
retriever = ElasticsearchRetriever(document_store=document_store)
# Alternative: An in-memory TfidfRetriever based on Pandas dataframes for building quick-prototypes
# with SQLite document store.
#
# from haystack.retriever.tfidf import TfidfRetriever
# retriever = TfidfRetriever(document_store=document_store)
# ### Reader
#
# A Reader scans the texts returned by retrievers in detail and extracts the k best answers. They are based
# on powerful, but slower deep learning models.
#
# Haystack currently supports Readers based on the frameworks FARM and Transformers.
# With both you can either load a local model or one from Hugging Face's model hub (https://huggingface.co/models).
# **Here:** a medium sized RoBERTa QA model using a Reader based on
# FARM (https://huggingface.co/deepset/roberta-base-squad2)
# **Alternatives (Reader):** TransformersReader (leveraging the `pipeline` of the Transformers package)
# **Alternatives (Models):** e.g. "distilbert-base-uncased-distilled-squad" (fast) or
# "deepset/bert-large-uncased-whole-word-masking-squad2" (good accuracy)
# **Hint:** You can adjust the model to return "no answer possible" with the no_ans_boost. Higher values mean
# the model prefers "no answer possible"
#
# #### FARMReader
# 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)
# #### TransformersReader
# Alternative:
# reader = TransformersReader(
# model_name_or_path="distilbert-base-uncased-distilled-squad", tokenizer="distilbert-base-uncased", use_gpu=-1)
# ### Pipeline
#
# With a Haystack `Pipeline` you can stick together your building blocks to a search pipeline.
# Under the hood, `Pipelines` are Directed Acyclic Graphs (DAGs) that you can easily customize for your own use cases.
# To speed things up, Haystack also comes with a few predefined Pipelines. One of them is the `ExtractiveQAPipeline` that combines a retriever and a reader to answer our questions.
# You can learn more about `Pipelines` in the [docs](https://haystack.deepset.ai/docs/latest/pipelinesmd).
from haystack.pipeline import ExtractiveQAPipeline
pipe = ExtractiveQAPipeline(reader, retriever)
## Voilà! Ask a question!
prediction = pipe.run(query="Who is the father of Arya Stark?",
top_k_retriever=10,
top_k_reader=5)
# prediction = pipe.run(query="Who created the Dothraki vocabulary?", top_k_reader=5)
# prediction = pipe.run(query="Who is the sister of Sansa?", top_k_reader=5)
print_answers(prediction, details="minimal")
def tutorial11_pipelines():
#Download and prepare data - 517 Wikipedia articles for Game of Thrones
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
got_dicts = convert_files_to_dicts(dir_path=doc_dir,
clean_func=clean_wiki_text,
split_paragraphs=True)
# Initialize DocumentStore and index documents
launch_es()
document_store = ElasticsearchDocumentStore()
document_store.delete_all_documents()
document_store.write_documents(got_dicts)
# Initialize Sparse retriever
es_retriever = ElasticsearchRetriever(document_store=document_store)
# Initialize dense retriever
dpr_retriever = DensePassageRetriever(document_store)
document_store.update_embeddings(dpr_retriever,
update_existing_embeddings=False)
reader = FARMReader(model_name_or_path="deepset/roberta-base-squad2")
######################
# Prebuilt Pipelines #
######################
# Extractive QA Pipeline
########################
p_extractive_premade = ExtractiveQAPipeline(reader=reader,
retriever=es_retriever)
res = p_extractive_premade.run(query="Who is the father of Arya Stark?",
top_k_retriever=10,
top_k_reader=5)
print_answers(res, details="minimal")
# Document Search Pipeline
##########################
p_retrieval = DocumentSearchPipeline(es_retriever)
res = p_retrieval.run(query="Who is the father of Arya Stark?",
top_k_retriever=10)
print_documents(res, max_text_len=200)
# Generator Pipeline
##########################
# We set this to True so that the document store returns document embeddings
# with each document, this is needed by the Generator
document_store.return_embedding = True
# Initialize generator
rag_generator = RAGenerator()
# Generative QA
p_generator = GenerativeQAPipeline(generator=rag_generator,
retriever=dpr_retriever)
res = p_generator.run(query="Who is the father of Arya Stark?",
top_k_retriever=10)
print_answers(res, details="minimal")
# We are setting this to False so that in later pipelines,
# we get a cleaner printout
document_store.return_embedding = False
##############################
# Creating Pipeline Diagrams #
##############################
p_extractive_premade.draw("pipeline_extractive_premade.png")
p_retrieval.draw("pipeline_retrieval.png")
p_generator.draw("pipeline_generator.png")
####################
# Custom Pipelines #
####################
# Extractive QA Pipeline
########################
# Custom built extractive QA pipeline
p_extractive = Pipeline()
p_extractive.add_node(component=es_retriever,
name="Retriever",
inputs=["Query"])
p_extractive.add_node(component=reader,
name="Reader",
inputs=["Retriever"])
# Now we can run it
res = p_extractive.run(query="Who is the father of Arya Stark?",
top_k_retriever=10,
top_k_reader=5)
print_answers(res, details="minimal")
p_extractive.draw("pipeline_extractive.png")
# Ensembled Retriever Pipeline
##############################
# Create ensembled pipeline
p_ensemble = Pipeline()
p_ensemble.add_node(component=es_retriever,
name="ESRetriever",
inputs=["Query"])
p_ensemble.add_node(component=dpr_retriever,
name="DPRRetriever",
inputs=["Query"])
p_ensemble.add_node(component=JoinDocuments(join_mode="concatenate"),
name="JoinResults",
inputs=["ESRetriever", "DPRRetriever"])
p_ensemble.add_node(component=reader,
name="Reader",
inputs=["JoinResults"])
p_ensemble.draw("pipeline_ensemble.png")
# Run pipeline
res = p_ensemble.run(
query="Who is the father of Arya Stark?",
top_k_retriever=5 #This is top_k per retriever
)
print_answers(res, details="minimal")
# Query Classification Pipeline
###############################
# Decision Nodes help you route your data so that only certain branches of your `Pipeline` are run.
# Though this looks very similar to the ensembled pipeline shown above,
# the key difference is that only one of the retrievers is run for each request.
# By contrast both retrievers are always run in the ensembled approach.
class QueryClassifier():
outgoing_edges = 2
def run(self, **kwargs):
if "?" in kwargs["query"]:
return (kwargs, "output_2")
else:
return (kwargs, "output_1")
# Here we build the pipeline
p_classifier = Pipeline()
p_classifier.add_node(component=QueryClassifier(),
name="QueryClassifier",
inputs=["Query"])
p_classifier.add_node(component=es_retriever,
name="ESRetriever",
inputs=["QueryClassifier.output_1"])
p_classifier.add_node(component=dpr_retriever,
name="DPRRetriever",
inputs=["QueryClassifier.output_2"])
p_classifier.add_node(component=reader,
name="QAReader",
inputs=["ESRetriever", "DPRRetriever"])
p_classifier.draw("pipeline_classifier.png")
# Run only the dense retriever on the full sentence query
res_1 = p_classifier.run(query="Who is the father of Arya Stark?",
top_k_retriever=10)
print("DPR Results" + "\n" + "=" * 15)
print_answers(res_1)
# Run only the sparse retriever on a keyword based query
res_2 = p_classifier.run(query="Arya Stark father", top_k_retriever=10)
print("ES Results" + "\n" + "=" * 15)
print_answers(res_2)