Spaces:
Runtime error
Runtime error
Create new file
Browse files
app.py
ADDED
@@ -0,0 +1,98 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
1 |
+
import streamlit as st
|
2 |
+
import pandas as pd
|
3 |
+
|
4 |
+
from haystack.schema import Answer
|
5 |
+
from haystack.document_stores import InMemoryDocumentStore
|
6 |
+
from haystack.pipeline import FAQPipeline
|
7 |
+
from haystack.retriever.dense import EmbeddingRetriever
|
8 |
+
import logging
|
9 |
+
|
10 |
+
#Haystack function calls - streamlit structure from Tuana GoT QA Haystack demo
|
11 |
+
@st.cache(hash_funcs={"builtins.SwigPyObject": lambda _: None},allow_output_mutation=True) # use streamlit cache
|
12 |
+
|
13 |
+
def start_haystack():
|
14 |
+
document_store = InMemoryDocumentStore(index="document", embedding_field='embedding', embedding_dim=384, similarity='cosine')
|
15 |
+
retriever = EmbeddingRetriever(document_store=document_store, embedding_model='sentence-transformers/all-MiniLM-L6-v2', use_gpu=False, top_k=1)
|
16 |
+
load_data_to_store(document_store,retriever)
|
17 |
+
pipeline = FAQPipeline(retriever=retriever)
|
18 |
+
return pipeline
|
19 |
+
|
20 |
+
def load_data_to_store(document_store, retriever):
|
21 |
+
df = pd.read_csv('monopoly_qa-v1.csv')
|
22 |
+
questions = list(df.Question)
|
23 |
+
df['embedding'] = retriever.embed_queries(texts=questions)
|
24 |
+
df = df.rename(columns={"Question":"content","Answer":"answer"})
|
25 |
+
df.drop('link to source (to prevent duplicate sources)',axis=1, inplace=True)
|
26 |
+
|
27 |
+
dicts = df.to_dict(orient="records")
|
28 |
+
document_store.write_documents(dicts)
|
29 |
+
|
30 |
+
pipeline = start_haystack()
|
31 |
+
|
32 |
+
# Streamlit App section
|
33 |
+
|
34 |
+
def set_state_if_absent(key, value):
|
35 |
+
if key not in st.session_state:
|
36 |
+
st.session_state[key] = value
|
37 |
+
|
38 |
+
set_state_if_absent("question", "how much money should each player have at the beginning?")
|
39 |
+
set_state_if_absent("results", None)
|
40 |
+
|
41 |
+
|
42 |
+
st.markdown( """
|
43 |
+
Haystack FAQ Semantic Search Pipeline
|
44 |
+
""", unsafe_allow_html=True)
|
45 |
+
|
46 |
+
question = st.text_input("", value=st.session_state.question, max_chars=100, on_change=reset_results)
|
47 |
+
|
48 |
+
def ask_question(question):
|
49 |
+
prediction = pipeline.run(query=question, params={"Retriever": {"top_k": 10}, "Reader": {"top_k": 5}})
|
50 |
+
results = []
|
51 |
+
for answer in prediction["answers"]:
|
52 |
+
answer = answer.to_dict()
|
53 |
+
if answer["answer"]:
|
54 |
+
results.append(
|
55 |
+
{
|
56 |
+
"context": "..." + answer["context"] + "...",
|
57 |
+
"answer": answer["answer"],
|
58 |
+
"relevance": round(answer["score"] * 100, 2),
|
59 |
+
"offset_start_in_doc": answer["offsets_in_document"][0]["start"],
|
60 |
+
}
|
61 |
+
)
|
62 |
+
else:
|
63 |
+
results.append(
|
64 |
+
{
|
65 |
+
"context": None,
|
66 |
+
"answer": None,
|
67 |
+
"relevance": round(answer["score"] * 100, 2),
|
68 |
+
}
|
69 |
+
)
|
70 |
+
return results
|
71 |
+
|
72 |
+
if question:
|
73 |
+
with st.spinner("π Performing semantic search on royal scripts..."):
|
74 |
+
try:
|
75 |
+
msg = 'Asked ' + question
|
76 |
+
logging.info(msg)
|
77 |
+
st.session_state.results = ask_question(question)
|
78 |
+
except Exception as e:
|
79 |
+
logging.exception(e)
|
80 |
+
|
81 |
+
|
82 |
+
|
83 |
+
if st.session_state.results:
|
84 |
+
st.write('## Top Results')
|
85 |
+
for count, result in enumerate(st.session_state.results):
|
86 |
+
if result["answer"]:
|
87 |
+
answer, context = result["answer"], result["context"]
|
88 |
+
start_idx = context.find(answer)
|
89 |
+
end_idx = start_idx + len(answer)
|
90 |
+
st.write(
|
91 |
+
markdown(context[:start_idx] + str(annotation(body=answer, label="ANSWER", background="#964448", color='#ffffff')) + context[end_idx:]),
|
92 |
+
unsafe_allow_html=True,
|
93 |
+
)
|
94 |
+
st.markdown(f"**Relevance:** {result['relevance']}")
|
95 |
+
else:
|
96 |
+
st.info(
|
97 |
+
"π€ Haystack is unsure whether any of the documents contain an answer to your question. Try to reformulate it!"
|
98 |
+
)
|