Spaces:
Running
Running
Update app.py
#4
by
gigiliu12
- opened
app.py
CHANGED
@@ -1,104 +1,156 @@
|
|
1 |
-
|
|
|
|
|
2 |
import streamlit as st
|
3 |
-
import
|
4 |
-
|
5 |
-
|
6 |
-
|
7 |
-
#
|
8 |
-
|
9 |
-
|
10 |
-
|
11 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
12 |
DB_PASSWORD = os.getenv("DB_PASSWORD")
|
13 |
|
14 |
@st.cache_data(ttl=600)
|
15 |
-
def
|
16 |
-
|
17 |
-
|
18 |
-
|
19 |
-
|
20 |
-
|
21 |
-
|
22 |
-
|
23 |
-
|
24 |
-
|
25 |
-
|
26 |
-
|
27 |
-
|
28 |
-
|
29 |
-
|
30 |
-
|
31 |
-
|
32 |
-
|
33 |
-
|
34 |
-
#
|
35 |
-
|
36 |
-
|
37 |
-
|
38 |
-
|
39 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
40 |
st.sidebar.header("π Filter Questions")
|
41 |
|
42 |
-
|
43 |
-
|
44 |
-
year_options = sorted(df["year"].dropna().unique())
|
45 |
|
46 |
-
|
47 |
-
|
48 |
-
keyword
|
49 |
-
"Keyword Search (Question
|
50 |
-
)
|
51 |
-
|
52 |
|
53 |
-
# Apply filters
|
54 |
filtered = df[
|
55 |
-
(df["country"].isin(
|
56 |
-
(df["year"].isin(
|
57 |
(
|
58 |
df["question_text"].str.contains(keyword, case=False, na=False) |
|
59 |
-
df["answer_text"].str.contains(keyword, case=False, na=False)
|
60 |
-
df["question_code"].astype(str).str.contains(keyword, case=False, na=False)
|
61 |
)
|
62 |
]
|
63 |
|
64 |
-
#
|
65 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
66 |
st.subheader("π Grouped by Question Text")
|
67 |
-
|
68 |
grouped = (
|
69 |
filtered.groupby("question_text")
|
70 |
.agg({
|
71 |
"country": lambda x: sorted(set(x)),
|
72 |
-
"year":
|
73 |
-
"answer_text": lambda x: list(x)[:3]
|
74 |
})
|
75 |
.reset_index()
|
76 |
.rename(columns={
|
77 |
"country": "Countries",
|
78 |
-
"year":
|
79 |
"answer_text": "Sample Answers"
|
80 |
})
|
81 |
)
|
82 |
-
|
83 |
st.dataframe(grouped)
|
84 |
-
|
85 |
if grouped.empty:
|
86 |
st.info("No questions found with current filters.")
|
87 |
-
|
88 |
else:
|
89 |
-
#
|
90 |
-
|
91 |
-
if
|
92 |
-
|
93 |
-
if
|
94 |
-
heading_parts.append("Years: " + ", ".join(map(str, selected_years)))
|
95 |
-
if heading_parts:
|
96 |
-
st.markdown("### Results for " + " | ".join(heading_parts))
|
97 |
-
else:
|
98 |
-
st.markdown("### Results for All Countries and Years")
|
99 |
-
|
100 |
st.dataframe(filtered[["country", "year", "question_text", "answer_text"]])
|
101 |
-
|
102 |
if filtered.empty:
|
103 |
st.info("No matching questions found.")
|
104 |
-
|
|
|
1 |
+
#!/usr/bin/env python3
|
2 |
+
import os, io, json, gc
|
3 |
+
import boto3, psycopg2, pandas as pd, torch
|
4 |
import streamlit as st
|
5 |
+
from sentence_transformers import SentenceTransformer, util
|
6 |
+
|
7 |
+
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
8 |
+
# 0) Hugging Face secrets β env vars (already set inside Spaces)
|
9 |
+
# DB_HOST / DB_PORT / DB_NAME / DB_USER / DB_PASSWORD
|
10 |
+
# AWS creds must be in aws_creds.json pushed with the app repo
|
11 |
+
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
12 |
+
with open("aws_creds.json") as f:
|
13 |
+
creds = json.load(f)
|
14 |
+
os.environ["AWS_ACCESS_KEY_ID"] = creds["AccessKey"]
|
15 |
+
os.environ["AWS_SECRET_ACCESS_KEY"] = creds["SecretAccessKey"]
|
16 |
+
os.environ["AWS_DEFAULT_REGION"] = "us-east-2"
|
17 |
+
|
18 |
+
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
19 |
+
# 1) DB β DataFrame (cached 10 min) |
|
20 |
+
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
21 |
+
DB_HOST = os.getenv("DB_HOST")
|
22 |
+
DB_PORT = os.getenv("DB_PORT", "5432")
|
23 |
+
DB_NAME = os.getenv("DB_NAME")
|
24 |
+
DB_USER = os.getenv("DB_USER")
|
25 |
DB_PASSWORD = os.getenv("DB_PASSWORD")
|
26 |
|
27 |
@st.cache_data(ttl=600)
|
28 |
+
def load_survey_dataframe() -> pd.DataFrame:
|
29 |
+
conn = psycopg2.connect(
|
30 |
+
host=DB_HOST, port=DB_PORT,
|
31 |
+
dbname=DB_NAME, user=DB_USER, password=DB_PASSWORD,
|
32 |
+
sslmode="require",
|
33 |
+
)
|
34 |
+
df = pd.read_sql_query(
|
35 |
+
"""SELECT id, country, year, section,
|
36 |
+
question_code, question_text,
|
37 |
+
answer_code, answer_text
|
38 |
+
FROM survey_info
|
39 |
+
""",
|
40 |
+
conn,
|
41 |
+
)
|
42 |
+
conn.close()
|
43 |
+
return df
|
44 |
+
|
45 |
+
df = load_survey_dataframe()
|
46 |
+
|
47 |
+
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
48 |
+
# 2) S3 β ids + embeddings (cached for session) |
|
49 |
+
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
50 |
+
@st.cache_resource
|
51 |
+
def load_embeddings():
|
52 |
+
BUCKET = "cgd-embeddings-bucket"
|
53 |
+
KEY = "survey_info_embeddings.pt" # contains {'ids', 'embeddings'}
|
54 |
+
bio = io.BytesIO()
|
55 |
+
boto3.client("s3").download_fileobj(BUCKET, KEY, bio)
|
56 |
+
bio.seek(0)
|
57 |
+
ckpt = torch.load(bio, map_location="cpu")
|
58 |
+
bio.close(); gc.collect()
|
59 |
+
if not (isinstance(ckpt, dict) and {"ids","embeddings"} <= ckpt.keys()):
|
60 |
+
st.error("Bad checkpoint format"); st.stop()
|
61 |
+
return ckpt["ids"], ckpt["embeddings"]
|
62 |
+
|
63 |
+
ids_list, emb_tensor = load_embeddings()
|
64 |
+
|
65 |
+
# build quick lookup from id β row index in DataFrame
|
66 |
+
row_lookup = {row_id: i for i, row_id in enumerate(df["id"])}
|
67 |
+
|
68 |
+
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
69 |
+
# 3) Streamlit UI |
|
70 |
+
# ββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββββ
|
71 |
+
st.title("π CGD Survey Explorer (Live DB + Semantic Search)")
|
72 |
+
|
73 |
+
# ββ 3a) Sidebar filters (original UI) ββββββββββββοΏ½οΏ½οΏ½ββββββββββββββββββββββ
|
74 |
st.sidebar.header("π Filter Questions")
|
75 |
|
76 |
+
country_opts = sorted(df["country"].dropna().unique())
|
77 |
+
year_opts = sorted(df["year"].dropna().unique())
|
|
|
78 |
|
79 |
+
sel_countries = st.sidebar.multiselect("Select Country/Countries", country_opts)
|
80 |
+
sel_years = st.sidebar.multiselect("Select Year(s)", year_opts)
|
81 |
+
keyword = st.sidebar.text_input(
|
82 |
+
"Keyword Search (Question / Answer / Code)", ""
|
83 |
+
)
|
84 |
+
group_by_q = st.sidebar.checkbox("Group by Question Text")
|
85 |
|
86 |
+
# Apply keyword & dropdown filters
|
87 |
filtered = df[
|
88 |
+
(df["country"].isin(sel_countries) if sel_countries else True) &
|
89 |
+
(df["year"].isin(sel_years) if sel_years else True) &
|
90 |
(
|
91 |
df["question_text"].str.contains(keyword, case=False, na=False) |
|
92 |
+
df["answer_text"] .str.contains(keyword, case=False, na=False) |
|
93 |
+
df["question_code"].astype(str).str.contains(keyword, case=False, na=False)
|
94 |
)
|
95 |
]
|
96 |
|
97 |
+
# ββ 3b) Semantic-search panel βββββββββββββββββββββββββββββββββββββββββββ
|
98 |
+
st.sidebar.markdown("---")
|
99 |
+
st.sidebar.subheader("π§ Semantic Search")
|
100 |
+
sem_query = st.sidebar.text_input("Enter a natural-language query")
|
101 |
+
if st.sidebar.button("Search", disabled=not sem_query.strip()):
|
102 |
+
with st.spinner("Embedding & searchingβ¦"):
|
103 |
+
model = SentenceTransformer("sentence-transformers/all-MiniLM-L6-v2")
|
104 |
+
q_vec = model.encode(sem_query.strip(), convert_to_tensor=True).cpu()
|
105 |
+
scores = util.cos_sim(q_vec, emb_tensor)[0]
|
106 |
+
top_vals, top_idx = torch.topk(scores, k=10)
|
107 |
+
results = []
|
108 |
+
for score, emb_row in zip(top_vals.tolist(), top_idx.tolist()):
|
109 |
+
db_id = ids_list[emb_row]
|
110 |
+
if db_id in row_lookup:
|
111 |
+
row = df.iloc[row_lookup[db_id]]
|
112 |
+
results.append({
|
113 |
+
"score": f"{score:.3f}",
|
114 |
+
"country": row["country"],
|
115 |
+
"year": row["year"],
|
116 |
+
"question": row["question_text"],
|
117 |
+
"answer": row["answer_text"],
|
118 |
+
})
|
119 |
+
if results:
|
120 |
+
st.subheader("π Semantic Results")
|
121 |
+
st.write(f"Showing top {len(results)} for **{sem_query}**")
|
122 |
+
st.dataframe(pd.DataFrame(results))
|
123 |
+
else:
|
124 |
+
st.info("No semantic matches found.")
|
125 |
+
|
126 |
+
st.markdown("---")
|
127 |
+
|
128 |
+
# ββ 3c) Original results table / grouped view βββββββββββββββββββββββββββ
|
129 |
+
if group_by_q:
|
130 |
st.subheader("π Grouped by Question Text")
|
|
|
131 |
grouped = (
|
132 |
filtered.groupby("question_text")
|
133 |
.agg({
|
134 |
"country": lambda x: sorted(set(x)),
|
135 |
+
"year": lambda x: sorted(set(x)),
|
136 |
+
"answer_text": lambda x: list(x)[:3]
|
137 |
})
|
138 |
.reset_index()
|
139 |
.rename(columns={
|
140 |
"country": "Countries",
|
141 |
+
"year": "Years",
|
142 |
"answer_text": "Sample Answers"
|
143 |
})
|
144 |
)
|
|
|
145 |
st.dataframe(grouped)
|
|
|
146 |
if grouped.empty:
|
147 |
st.info("No questions found with current filters.")
|
|
|
148 |
else:
|
149 |
+
# contextual heading
|
150 |
+
hdr = []
|
151 |
+
if sel_countries: hdr.append("Countries: " + ", ".join(sel_countries))
|
152 |
+
if sel_years: hdr.append("Years: " + ", ".join(map(str, sel_years)))
|
153 |
+
st.markdown("### Results for " + (" | ".join(hdr) if hdr else "All Countries and Years"))
|
|
|
|
|
|
|
|
|
|
|
|
|
154 |
st.dataframe(filtered[["country", "year", "question_text", "answer_text"]])
|
|
|
155 |
if filtered.empty:
|
156 |
st.info("No matching questions found.")
|
|