yujunjun/haystack
1
0
Fork 0
mirror of https://github.com/deepset-ai/haystack.git synced 2026-01-06 03:57:19 +00:00
Code Issues Packages Projects Releases Wiki Activity

Add File Upload Functionality in UI (#995)

Browse Source
This commit is contained in:
Bhadresh Savani 2021-04-30 14:16:30 +05:30 committed by GitHub
parent 056be3354b
commit 37a72d2f45
No known key found for this signature in database
GPG Key ID: 4AEE18F83AFDEB23
4 changed files with 144 additions and 79 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

6
rest_api/controller/file_upload.py
View File

@ -44,6 +44,10 @@ def file_upload(
file_path = Path(FILE_UPLOAD_PATH) / f"{uuid.uuid4().hex}_{file.filename}"
with file_path.open("wb") as buffer:
shutil.copyfileobj(file.file, buffer)
meta = json.loads(meta) or {}
meta["name"] = file.filename
INDEXING_PIPELINE.run(
file_path=file_path,
remove_numeric_tables=remove_numeric_tables,
@ -55,7 +59,7 @@ def file_upload(
split_length=split_length,
split_overlap=split_overlap,
split_respect_sentence_boundary=split_respect_sentence_boundary,
meta=json.loads(meta) or {},
meta=meta,
)
finally:
file.file.close()

3
rest_api/pipelines.yaml
View File

@ -20,6 +20,9 @@ components: # define all the building-blocks for Pipeline
type: PDFToTextConverter
- name: Preprocessor
type: PreProcessor
params:
split_by: word
split_length: 1000
- name: FileTypeClassifier
type: FileTypeClassifier

85
ui/utils.py
View File

@ -6,39 +6,58 @@ import streamlit as st
API_ENDPOINT = os.getenv("API_ENDPOINT", "http://localhost:8000")
DOC_REQUEST = "query"
DOC_FEEDBACK = "feedback"
DOC_UPLOAD = "file-upload"
@st.cache(show_spinner=False)
def retrieve_doc(query,filters=None,top_k_reader=5,top_k_retriever=5):
# Query Haystack API
url = f"{API_ENDPOINT}/{DOC_REQUEST}"
req = {"query": query, "filters": filters, "top_k_retriever": top_k_retriever, "top_k_reader": top_k_reader}
response_raw = requests.post(url,json=req).json()
# Format response
result = []
answers = response_raw["answers"]
for i in range(len(answers)):
answer = answers[i]['answer']
if answer:
context = '...' + answers[i]['context'] + '...'
meta_name = answers[i]['meta']['name']
relevance = round(answers[i]['probability']*100,2)
document_id = answers[i]['document_id']
offset_start_in_doc = answers[i]['offset_start_in_doc']
result.append({'context':context,'answer':answer,'source':meta_name,'relevance':relevance, 'document_id':document_id,'offset_start_in_doc':offset_start_in_doc})
return result, response_raw
def retrieve_doc(query, filters=None, top_k_reader=5, top_k_retriever=5):
# Query Haystack API
url = f"{API_ENDPOINT}/{DOC_REQUEST}"
req = {"query": query, "filters": filters, "top_k_retriever": top_k_retriever, "top_k_reader": top_k_reader}
response_raw = requests.post(url, json=req).json()
def feedback_doc(question,is_correct_answer,document_id,model_id,is_correct_document,answer,offset_start_in_doc):
# Feedback Haystack API
url = f"{API_ENDPOINT}/{DOC_FEEDBACK}"
req = {
"question": question,
"is_correct_answer": is_correct_answer,
"document_id": document_id,
"model_id": model_id,
"is_correct_document": is_correct_document,
"answer": answer,
"offset_start_in_doc": offset_start_in_doc
}
response_raw = requests.post(url,json=req).json()
return response_raw
# Format response
result = []
answers = response_raw["answers"]
for i in range(len(answers)):
answer = answers[i]["answer"]
if answer:
context = "..." + answers[i]["context"] + "..."
meta_name = answers[i]["meta"]["name"]
relevance = round(answers[i]["probability"] * 100, 2)
document_id = answers[i]["document_id"]
offset_start_in_doc = answers[i]["offset_start_in_doc"]
result.append(
{
"context": context,
"answer": answer,
"source": meta_name,
"relevance": relevance,
"document_id": document_id,
"offset_start_in_doc": offset_start_in_doc,
}
)
return result, response_raw
def feedback_doc(question, is_correct_answer, document_id, model_id, is_correct_document, answer, offset_start_in_doc):
# Feedback Haystack API
url = f"{API_ENDPOINT}/{DOC_FEEDBACK}"
req = {
"question": question,
"is_correct_answer": is_correct_answer,
"document_id": document_id,
"model_id": model_id,
"is_correct_document": is_correct_document,
"answer": answer,
"offset_start_in_doc": offset_start_in_doc,
}
response_raw = requests.post(url, json=req).json()
return response_raw
def upload_doc(file):
url = f"{API_ENDPOINT}/{DOC_UPLOAD}"
files = [("file", file)]
response_raw = requests.post(url, files=files).json()
return response_raw

129
ui/webapp.py
View File

@ -1,49 +1,74 @@
import os
import sys
import streamlit as st
from utils import retrieve_doc
from utils import feedback_doc
from annotated_text import annotated_text
import pandas as pd
# streamlit does not support any states out of the box. On every button click, streamlit reload the whole page
# and every value gets lost. To keep track of our feedback state we use the official streamlit gist mentioned
import streamlit as st
from annotated_text import annotated_text
# streamlit does not support any states out of the box. On every button click, streamlit reload the whole page
# and every value gets lost. To keep track of our feedback state we use the official streamlit gist mentioned
# here https://gist.github.com/tvst/036da038ab3e999a64497f42de966a92
import SessionState
from utils import feedback_doc
from utils import retrieve_doc
from utils import upload_doc
def annotate_answer(answer, context):
start_idx = context.find(answer)
end_idx = start_idx+len(answer)
annotated_text(context[:start_idx],(answer,"ANSWER","#8ef"),context[end_idx:])
end_idx = start_idx + len(answer)
annotated_text(context[:start_idx], (answer, "ANSWER", "#8ef"), context[end_idx:])
def random_questions(df):
random_row = df.sample(1)
random_question = random_row["Question Text"].values[0]
random_answer = random_row["Answer"].values[0]
return random_question, random_answer
random_row = df.sample(1)
random_question = random_row["Question Text"].values[0]
random_answer = random_row["Answer"].values[0]
return random_question, random_answer
# Define state
state_question = SessionState.get(random_question='Who is the father of Arya Starck?', random_answer='', next_question='false', run_query='false')
state_question = SessionState.get(
random_question="Who is the father of Arya Starck?", random_answer="", next_question="false", run_query="false"
)
# Initalize variables
# Initialize variables
eval_mode = False
random_question = "Who is the father of Arya Starck?"
eval_labels = os.getenv("EVAL_FILE", "eval_labels_example.csv")
# UI search bar and sidebar
# UI search bar and sidebar
st.write("# Haystack Demo")
st.sidebar.header("Options")
top_k_reader = st.sidebar.slider("Max. number of answers",min_value=1,max_value=10,value=3,step=1)
top_k_retriever = st.sidebar.slider("Max. number of documents from retriever",min_value=1,max_value=10,value=3,step=1)
top_k_reader = st.sidebar.slider("Max. number of answers", min_value=1, max_value=10, value=3, step=1)
top_k_retriever = st.sidebar.slider(
"Max. number of documents from retriever", min_value=1, max_value=10, value=3, step=1
)
eval_mode = st.sidebar.checkbox("Evalution mode")
debug = st.sidebar.checkbox("Show debug info")
st.sidebar.write("## File Upload:")
data_file = st.sidebar.file_uploader("", type=["pdf", "txt", "docx"])
# Upload file
if data_file:
raw_json = upload_doc(data_file)
st.write(raw_json)
if debug:
st.subheader("REST API JSON response")
st.write(raw_json)
# load csv into pandas dataframe
if eval_mode:
try:
df = pd.read_csv(eval_labels, sep=";")
except Exception:
sys.exit('The eval file was not found. Please check the README for more information.')
if state_question and hasattr(state_question, 'next_question') and hasattr(state_question, 'random_question') and state_question.next_question:
sys.exit("The eval file was not found. Please check the README for more information.")
if (
state_question
and hasattr(state_question, "next_question")
and hasattr(state_question, "random_question")
and state_question.next_question
):
random_question = state_question.random_question
random_answer = state_question.random_answer
else:
@ -55,16 +80,16 @@ if eval_mode:
if eval_mode:
next_question = st.button("Load new question")
if next_question:
random_question, random_answer = random_questions(df)
state_question.random_question = random_question
state_question.random_answer = random_answer
state_question.next_question = "true"
state_question.run_query = "false"
random_question, random_answer = random_questions(df)
state_question.random_question = random_question
state_question.random_answer = random_answer
state_question.next_question = "true"
state_question.run_query = "false"
else:
state_question.next_question = "false"
state_question.next_question = "false"
# Search bar
question = st.text_input("Please provide your query:",value=random_question)
question = st.text_input("Please provide your query:", value=random_question)
if state_question and state_question.run_query:
run_query = state_question.run_query
st.button("Run")
@ -76,38 +101,52 @@ raw_json_feedback = ""
# Get results for query
if run_query:
with st.spinner("Performing neural search on documents... 🧠 \n "
"Do you want to optimize speed or accuracy? \n"
"Check out the docs: https://haystack.deepset.ai/docs/latest/optimizationmd "):
results,raw_json = retrieve_doc(question,top_k_reader=top_k_reader,top_k_retriever=top_k_retriever)
with st.spinner(
"Performing neural search on documents... 🧠 \n "
"Do you want to optimize speed or accuracy? \n"
"Check out the docs: https://haystack.deepset.ai/docs/latest/optimizationmd "
):
results, raw_json = retrieve_doc(question, top_k_reader=top_k_reader, top_k_retriever=top_k_retriever)
# Show if we use a question of the given set
if question == random_question and eval_mode:
st.write("## Correct answers:")
random_answer
st.write("## Retrieved answers:")
# Make every button key unique
count = 0
for result in results:
annotate_answer(result['answer'],result['context'])
'**Relevance:** ', result['relevance'] , '**Source:** ' , result['source']
annotate_answer(result["answer"], result["context"])
"**Relevance:** ", result["relevance"], "**Source:** ", result["source"]
if eval_mode:
# Define columns for buttons
button_col1, button_col2, button_col3, button_col4 = st.beta_columns([1,1,1,6])
if button_col1.button("👍", key=(result['answer'] + str(count)), help="Correct answer"):
raw_json_feedback = feedback_doc(question,"true",result['document_id'],1,"true",result['answer'],result['offset_start_in_doc'])
st.success('Thanks for your feedback')
if button_col2.button("👎", key=(result['answer'] + str(count)), help="Wrong answer and wrong passage"):
raw_json_feedback = feedback_doc(question,"false",result['document_id'],1,"false",result['answer'],result['offset_start_in_doc'])
st.success('Thanks for your feedback!')
if button_col3.button("👎👍", key=(result['answer'] + str(count)), help="Wrong answer, but correct passage"):
raw_json_feedback = feedback_doc(question,"false",result['document_id'],1,"true",result['answer'],result['offset_start_in_doc'])
st.success('Thanks for your feedback!')
count+=1
button_col1, button_col2, button_col3, button_col4 = st.beta_columns([1, 1, 1, 6])
if button_col1.button("👍", key=(result["answer"] + str(count)), help="Correct answer"):
raw_json_feedback = feedback_doc(
question, "true", result["document_id"], 1, "true", result["answer"], result["offset_start_in_doc"]
)
st.success("Thanks for your feedback")
if button_col2.button("👎", key=(result["answer"] + str(count)), help="Wrong answer and wrong passage"):
raw_json_feedback = feedback_doc(
question,
"false",
result["document_id"],
1,
"false",
result["answer"],
result["offset_start_in_doc"],
)
st.success("Thanks for your feedback!")
if button_col3.button("👎👍", key=(result["answer"] + str(count)), help="Wrong answer, but correct passage"):
raw_json_feedback = feedback_doc(
question, "false", result["document_id"], 1, "true", result["answer"], result["offset_start_in_doc"]
)
st.success("Thanks for your feedback!")
count += 1
st.write("___")
if debug:
st.subheader("REST API JSON response")
st.write(raw_json)
st.write(raw_json)