Update app.py
Browse files
app.py
CHANGED
@@ -3,34 +3,52 @@ import pandas as pd
|
|
3 |
import numpy as np
|
4 |
from sentence_transformers import SentenceTransformer
|
5 |
from sklearn.metrics.pairwise import cosine_similarity
|
|
|
|
|
6 |
import os
|
|
|
|
|
7 |
from datetime import datetime
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
8 |
from datasets import load_dataset
|
9 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
10 |
# Initialize session state
|
11 |
-
|
12 |
-
st.session_state
|
13 |
-
|
14 |
-
st.session_state['search_columns'] = []
|
15 |
-
if 'dataset_loaded' not in st.session_state:
|
16 |
-
st.session_state['dataset_loaded'] = False
|
17 |
-
if 'current_page' not in st.session_state:
|
18 |
-
st.session_state['current_page'] = 0
|
19 |
-
if 'data_cache' not in st.session_state:
|
20 |
-
st.session_state['data_cache'] = None
|
21 |
-
if 'dataset_info' not in st.session_state:
|
22 |
-
st.session_state['dataset_info'] = None
|
23 |
-
|
24 |
-
ROWS_PER_PAGE = 100 # Number of rows to load at a time
|
25 |
|
26 |
@st.cache_resource
|
27 |
def get_model():
|
28 |
-
"""Cache the model loading"""
|
29 |
return SentenceTransformer('all-MiniLM-L6-v2')
|
30 |
|
31 |
@st.cache_data
|
32 |
def load_dataset_page(dataset_id, token, page, rows_per_page):
|
33 |
-
"""Load and cache a specific page of data"""
|
34 |
try:
|
35 |
start_idx = page * rows_per_page
|
36 |
end_idx = start_idx + rows_per_page
|
@@ -47,113 +65,408 @@ def load_dataset_page(dataset_id, token, page, rows_per_page):
|
|
47 |
|
48 |
@st.cache_data
|
49 |
def get_dataset_info(dataset_id, token):
|
50 |
-
"""Load and cache dataset information"""
|
51 |
try:
|
52 |
-
dataset = load_dataset(
|
53 |
-
dataset_id,
|
54 |
-
token=token,
|
55 |
-
streaming=True
|
56 |
-
)
|
57 |
return dataset['train'].info
|
58 |
except Exception as e:
|
59 |
st.error(f"Error loading dataset info: {str(e)}")
|
60 |
return None
|
61 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
62 |
class FastDatasetSearcher:
|
63 |
def __init__(self, dataset_id="tomg-group-umd/cinepile"):
|
64 |
self.dataset_id = dataset_id
|
65 |
self.text_model = get_model()
|
66 |
self.token = os.environ.get('DATASET_KEY')
|
67 |
if not self.token:
|
68 |
-
st.error("Please set the DATASET_KEY environment variable
|
69 |
st.stop()
|
70 |
|
71 |
-
# Initialize numpy for model inputs
|
72 |
-
self.np = np
|
73 |
-
|
74 |
-
# Load dataset info if not already loaded
|
75 |
if st.session_state['dataset_info'] is None:
|
76 |
st.session_state['dataset_info'] = get_dataset_info(self.dataset_id, self.token)
|
77 |
|
78 |
def load_page(self, page=0):
|
79 |
-
"""Load a specific page of data using cached function"""
|
80 |
return load_dataset_page(self.dataset_id, self.token, page, ROWS_PER_PAGE)
|
81 |
|
82 |
def quick_search(self, query, df):
|
83 |
-
"""
|
84 |
-
if df.empty:
|
85 |
return df
|
86 |
|
87 |
try:
|
88 |
-
#
|
|
|
|
|
|
|
|
|
89 |
searchable_cols = []
|
90 |
for col in df.columns:
|
91 |
sample_val = df[col].iloc[0]
|
92 |
if not isinstance(sample_val, (np.ndarray, bytes)):
|
93 |
searchable_cols.append(col)
|
94 |
|
95 |
-
# Prepare query
|
96 |
query_lower = query.lower()
|
|
|
97 |
query_embedding = self.text_model.encode([query], show_progress_bar=False)[0]
|
|
|
98 |
scores = []
|
|
|
99 |
|
100 |
-
# Process each row
|
101 |
for _, row in df.iterrows():
|
102 |
-
# Combine text from searchable columns
|
103 |
text_parts = []
|
|
|
|
|
|
|
104 |
for col in searchable_cols:
|
105 |
val = row[col]
|
106 |
if val is not None:
|
107 |
-
|
108 |
-
|
109 |
-
|
110 |
-
|
111 |
|
112 |
text = ' '.join(text_parts)
|
113 |
|
114 |
-
# Calculate scores
|
115 |
if text.strip():
|
116 |
-
#
|
117 |
-
|
|
|
|
|
118 |
|
119 |
-
#
|
120 |
text_embedding = self.text_model.encode([text], show_progress_bar=False)[0]
|
121 |
semantic_score = float(cosine_similarity([query_embedding], [text_embedding])[0][0])
|
122 |
|
123 |
-
#
|
124 |
-
combined_score = 0.
|
|
|
|
|
|
|
|
|
125 |
else:
|
126 |
combined_score = 0.0
|
|
|
127 |
|
128 |
scores.append(combined_score)
|
|
|
129 |
|
130 |
-
# Get top results
|
131 |
results_df = df.copy()
|
132 |
results_df['score'] = scores
|
133 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
134 |
|
135 |
except Exception as e:
|
136 |
st.error(f"Search error: {str(e)}")
|
137 |
return df
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
138 |
|
139 |
-
|
140 |
-
|
141 |
-
|
142 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
143 |
|
144 |
def render_result(result):
|
145 |
-
|
146 |
-
|
147 |
-
|
148 |
-
result_filtered = result.drop('score') if 'score' in result else result
|
149 |
|
150 |
-
# Display video if available
|
151 |
if 'youtube_id' in result:
|
152 |
-
st.video(
|
153 |
-
f"https://youtube.com/watch?v={result['youtube_id']}&t={result.get('start_time', 0)}"
|
154 |
-
)
|
155 |
|
156 |
-
# Display other fields
|
157 |
cols = st.columns([2, 1])
|
158 |
with cols[0]:
|
159 |
for key, value in result_filtered.items():
|
@@ -164,61 +477,114 @@ def render_result(result):
|
|
164 |
st.metric("Relevance Score", f"{score:.2%}")
|
165 |
|
166 |
def main():
|
167 |
-
st.title("π₯
|
168 |
|
169 |
-
# Initialize search
|
170 |
-
|
171 |
|
172 |
-
#
|
173 |
-
|
174 |
-
|
175 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
176 |
|
177 |
-
|
178 |
-
|
179 |
-
|
180 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
181 |
|
182 |
-
#
|
183 |
-
with
|
184 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
185 |
|
186 |
-
|
187 |
-
|
188 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
189 |
|
190 |
-
#
|
191 |
-
|
192 |
-
|
193 |
-
query = st.text_input("Search in current page:",
|
194 |
-
help="Searches within currently loaded data")
|
195 |
-
with col2:
|
196 |
-
max_results = st.slider("Max results", 1, ROWS_PER_PAGE, 10)
|
197 |
|
198 |
-
|
199 |
-
|
200 |
-
|
201 |
-
|
202 |
-
|
203 |
-
st.
|
204 |
-
|
205 |
-
|
206 |
-
|
207 |
-
|
208 |
-
|
209 |
-
|
210 |
-
|
211 |
-
|
212 |
-
|
213 |
-
|
214 |
-
|
215 |
-
|
216 |
-
|
217 |
-
st.rerun()
|
218 |
-
with cols[1]:
|
219 |
-
if st.button("Next Page β‘οΈ"):
|
220 |
-
st.session_state['current_page'] = current_page + 1
|
221 |
-
st.rerun()
|
222 |
|
223 |
if __name__ == "__main__":
|
224 |
main()
|
|
|
3 |
import numpy as np
|
4 |
from sentence_transformers import SentenceTransformer
|
5 |
from sklearn.metrics.pairwise import cosine_similarity
|
6 |
+
import torch
|
7 |
+
import json
|
8 |
import os
|
9 |
+
import glob
|
10 |
+
from pathlib import Path
|
11 |
from datetime import datetime
|
12 |
+
import edge_tts
|
13 |
+
import asyncio
|
14 |
+
import requests
|
15 |
+
from collections import defaultdict
|
16 |
+
from audio_recorder_streamlit import audio_recorder
|
17 |
+
import streamlit.components.v1 as components
|
18 |
+
from urllib.parse import quote
|
19 |
+
from xml.etree import ElementTree as ET
|
20 |
from datasets import load_dataset
|
21 |
|
22 |
+
# π§ Initialize session state variables
|
23 |
+
SESSION_VARS = {
|
24 |
+
'search_history': [], # Track search history
|
25 |
+
'last_voice_input': "", # Last voice input
|
26 |
+
'transcript_history': [], # Conversation history
|
27 |
+
'should_rerun': False, # Trigger for UI updates
|
28 |
+
'search_columns': [], # Available search columns
|
29 |
+
'initial_search_done': False, # First search flag
|
30 |
+
'tts_voice': "en-US-AriaNeural", # Default voice
|
31 |
+
'arxiv_last_query': "", # Last ArXiv search
|
32 |
+
'dataset_loaded': False, # Dataset load status
|
33 |
+
'current_page': 0, # Current data page
|
34 |
+
'data_cache': None, # Data cache
|
35 |
+
'dataset_info': None # Dataset metadata
|
36 |
+
}
|
37 |
+
|
38 |
+
# Constants
|
39 |
+
ROWS_PER_PAGE = 100
|
40 |
+
|
41 |
# Initialize session state
|
42 |
+
for var, default in SESSION_VARS.items():
|
43 |
+
if var not in st.session_state:
|
44 |
+
st.session_state[var] = default
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
45 |
|
46 |
@st.cache_resource
|
47 |
def get_model():
|
|
|
48 |
return SentenceTransformer('all-MiniLM-L6-v2')
|
49 |
|
50 |
@st.cache_data
|
51 |
def load_dataset_page(dataset_id, token, page, rows_per_page):
|
|
|
52 |
try:
|
53 |
start_idx = page * rows_per_page
|
54 |
end_idx = start_idx + rows_per_page
|
|
|
65 |
|
66 |
@st.cache_data
|
67 |
def get_dataset_info(dataset_id, token):
|
|
|
68 |
try:
|
69 |
+
dataset = load_dataset(dataset_id, token=token, streaming=True)
|
|
|
|
|
|
|
|
|
70 |
return dataset['train'].info
|
71 |
except Exception as e:
|
72 |
st.error(f"Error loading dataset info: {str(e)}")
|
73 |
return None
|
74 |
|
75 |
+
def fetch_dataset_info(dataset_id):
|
76 |
+
info_url = f"https://huggingface.co/api/datasets/{dataset_id}"
|
77 |
+
try:
|
78 |
+
response = requests.get(info_url, timeout=30)
|
79 |
+
if response.status_code == 200:
|
80 |
+
return response.json()
|
81 |
+
except Exception as e:
|
82 |
+
st.warning(f"Error fetching dataset info: {e}")
|
83 |
+
return None
|
84 |
+
|
85 |
+
def fetch_dataset_rows(dataset_id, config="default", split="train", max_rows=100):
|
86 |
+
url = f"https://datasets-server.huggingface.co/first-rows?dataset={dataset_id}&config={config}&split={split}"
|
87 |
+
try:
|
88 |
+
response = requests.get(url, timeout=30)
|
89 |
+
if response.status_code == 200:
|
90 |
+
data = response.json()
|
91 |
+
if 'rows' in data:
|
92 |
+
processed_rows = []
|
93 |
+
for row_data in data['rows']:
|
94 |
+
row = row_data.get('row', row_data)
|
95 |
+
# Process embeddings if present
|
96 |
+
for key in row:
|
97 |
+
if any(term in key.lower() for term in ['embed', 'vector', 'encoding']):
|
98 |
+
if isinstance(row[key], str):
|
99 |
+
try:
|
100 |
+
row[key] = [float(x.strip()) for x in row[key].strip('[]').split(',') if x.strip()]
|
101 |
+
except:
|
102 |
+
continue
|
103 |
+
row['_config'] = config
|
104 |
+
row['_split'] = split
|
105 |
+
processed_rows.append(row)
|
106 |
+
return processed_rows
|
107 |
+
except Exception as e:
|
108 |
+
st.warning(f"Error fetching rows: {e}")
|
109 |
+
return []
|
110 |
+
|
111 |
class FastDatasetSearcher:
|
112 |
def __init__(self, dataset_id="tomg-group-umd/cinepile"):
|
113 |
self.dataset_id = dataset_id
|
114 |
self.text_model = get_model()
|
115 |
self.token = os.environ.get('DATASET_KEY')
|
116 |
if not self.token:
|
117 |
+
st.error("Please set the DATASET_KEY environment variable")
|
118 |
st.stop()
|
119 |
|
|
|
|
|
|
|
|
|
120 |
if st.session_state['dataset_info'] is None:
|
121 |
st.session_state['dataset_info'] = get_dataset_info(self.dataset_id, self.token)
|
122 |
|
123 |
def load_page(self, page=0):
|
|
|
124 |
return load_dataset_page(self.dataset_id, self.token, page, ROWS_PER_PAGE)
|
125 |
|
126 |
def quick_search(self, query, df):
|
127 |
+
"""Enhanced search with improved relevance filtering"""
|
128 |
+
if df.empty or not query.strip():
|
129 |
return df
|
130 |
|
131 |
try:
|
132 |
+
# Define relevance thresholds
|
133 |
+
MIN_KEYWORD_MATCHES = 0.1
|
134 |
+
MIN_SEMANTIC_SCORE = 0.3
|
135 |
+
|
136 |
+
# Get searchable columns
|
137 |
searchable_cols = []
|
138 |
for col in df.columns:
|
139 |
sample_val = df[col].iloc[0]
|
140 |
if not isinstance(sample_val, (np.ndarray, bytes)):
|
141 |
searchable_cols.append(col)
|
142 |
|
|
|
143 |
query_lower = query.lower()
|
144 |
+
query_terms = set(query_lower.split())
|
145 |
query_embedding = self.text_model.encode([query], show_progress_bar=False)[0]
|
146 |
+
|
147 |
scores = []
|
148 |
+
matched_any = []
|
149 |
|
|
|
150 |
for _, row in df.iterrows():
|
|
|
151 |
text_parts = []
|
152 |
+
row_matched = False
|
153 |
+
|
154 |
+
# Check for direct matches
|
155 |
for col in searchable_cols:
|
156 |
val = row[col]
|
157 |
if val is not None:
|
158 |
+
val_str = str(val).lower()
|
159 |
+
if any(term in val_str for term in query_terms):
|
160 |
+
row_matched = True
|
161 |
+
text_parts.append(str(val))
|
162 |
|
163 |
text = ' '.join(text_parts)
|
164 |
|
|
|
165 |
if text.strip():
|
166 |
+
# Calculate term-based keyword score
|
167 |
+
text_terms = set(text.lower().split())
|
168 |
+
matching_terms = query_terms.intersection(text_terms)
|
169 |
+
keyword_score = len(matching_terms) / len(query_terms)
|
170 |
|
171 |
+
# Calculate semantic score
|
172 |
text_embedding = self.text_model.encode([text], show_progress_bar=False)[0]
|
173 |
semantic_score = float(cosine_similarity([query_embedding], [text_embedding])[0][0])
|
174 |
|
175 |
+
# Weighted combination
|
176 |
+
combined_score = 0.7 * keyword_score + 0.3 * semantic_score
|
177 |
+
|
178 |
+
# Boost exact matches
|
179 |
+
if row_matched:
|
180 |
+
combined_score *= 1.5
|
181 |
else:
|
182 |
combined_score = 0.0
|
183 |
+
row_matched = False
|
184 |
|
185 |
scores.append(combined_score)
|
186 |
+
matched_any.append(row_matched)
|
187 |
|
|
|
188 |
results_df = df.copy()
|
189 |
results_df['score'] = scores
|
190 |
+
results_df['matched'] = matched_any
|
191 |
+
|
192 |
+
# Filter relevant results
|
193 |
+
filtered_df = results_df[
|
194 |
+
(results_df['matched']) | # Include direct matches
|
195 |
+
(results_df['score'] > MIN_KEYWORD_MATCHES) # Or high relevance
|
196 |
+
]
|
197 |
+
|
198 |
+
return filtered_df.sort_values('score', ascending=False)
|
199 |
|
200 |
except Exception as e:
|
201 |
st.error(f"Search error: {str(e)}")
|
202 |
return df
|
203 |
+
|
204 |
+
class VideoSearch:
|
205 |
+
def __init__(self):
|
206 |
+
self.text_model = SentenceTransformer('all-MiniLM-L6-v2')
|
207 |
+
self.dataset_id = "omegalabsinc/omega-multimodal"
|
208 |
+
self.load_dataset()
|
209 |
+
|
210 |
+
def fetch_dataset_rows(self):
|
211 |
+
try:
|
212 |
+
df, configs, splits = search_dataset(
|
213 |
+
self.dataset_id,
|
214 |
+
"",
|
215 |
+
include_configs=None,
|
216 |
+
include_splits=None
|
217 |
+
)
|
218 |
+
|
219 |
+
if not df.empty:
|
220 |
+
st.session_state['search_columns'] = [col for col in df.columns
|
221 |
+
if col not in ['video_embed', 'description_embed', 'audio_embed']
|
222 |
+
and not col.startswith('_')]
|
223 |
+
return df
|
224 |
+
|
225 |
+
return self.load_example_data()
|
226 |
+
|
227 |
+
except Exception as e:
|
228 |
+
st.warning(f"Error loading videos: {e}")
|
229 |
+
return self.load_example_data()
|
230 |
+
|
231 |
+
def load_example_data(self):
|
232 |
+
example_data = [{
|
233 |
+
"video_id": "sample-123",
|
234 |
+
"youtube_id": "dQw4w9WgXcQ",
|
235 |
+
"description": "An example video",
|
236 |
+
"views": 12345,
|
237 |
+
"start_time": 0,
|
238 |
+
"end_time": 60
|
239 |
+
}]
|
240 |
+
return pd.DataFrame(example_data)
|
241 |
+
|
242 |
+
def load_dataset(self):
|
243 |
+
self.dataset = self.fetch_dataset_rows()
|
244 |
+
self.prepare_features()
|
245 |
+
|
246 |
+
def prepare_features(self):
|
247 |
+
try:
|
248 |
+
embed_cols = [col for col in self.dataset.columns
|
249 |
+
if any(term in col.lower() for term in ['embed', 'vector', 'encoding'])]
|
250 |
+
|
251 |
+
embeddings = {}
|
252 |
+
for col in embed_cols:
|
253 |
+
try:
|
254 |
+
data = []
|
255 |
+
for row in self.dataset[col]:
|
256 |
+
if isinstance(row, str):
|
257 |
+
values = [float(x.strip()) for x in row.strip('[]').split(',') if x.strip()]
|
258 |
+
elif isinstance(row, list):
|
259 |
+
values = row
|
260 |
+
else:
|
261 |
+
continue
|
262 |
+
data.append(values)
|
263 |
+
|
264 |
+
if data:
|
265 |
+
embeddings[col] = np.array(data)
|
266 |
+
except:
|
267 |
+
continue
|
268 |
+
|
269 |
+
self.video_embeds = embeddings.get('video_embed', next(iter(embeddings.values())) if embeddings else None)
|
270 |
+
self.text_embeds = embeddings.get('description_embed', self.video_embeds)
|
271 |
+
|
272 |
+
except:
|
273 |
+
num_rows = len(self.dataset)
|
274 |
+
self.video_embeds = np.random.randn(num_rows, 384)
|
275 |
+
self.text_embeds = np.random.randn(num_rows, 384)
|
276 |
+
|
277 |
+
def search(self, query, column=None, top_k=20):
|
278 |
+
"""Enhanced search with better relevance scoring"""
|
279 |
+
MIN_RELEVANCE = 0.3 # Minimum relevance threshold
|
280 |
+
|
281 |
+
query_embedding = self.text_model.encode([query])[0]
|
282 |
+
video_sims = cosine_similarity([query_embedding], self.video_embeds)[0]
|
283 |
+
text_sims = cosine_similarity([query_embedding], self.text_embeds)[0]
|
284 |
+
combined_sims = 0.7 * text_sims + 0.3 * video_sims # Favor text matches
|
285 |
+
|
286 |
+
if column and column in self.dataset.columns and column != "All Fields":
|
287 |
+
# Direct matches in specified column
|
288 |
+
matches = self.dataset[column].astype(str).str.contains(query, case=False)
|
289 |
+
combined_sims[matches] *= 1.5 # Boost exact matches
|
290 |
+
|
291 |
+
# Filter by minimum relevance
|
292 |
+
relevant_indices = np.where(combined_sims >= MIN_RELEVANCE)[0]
|
293 |
+
if len(relevant_indices) == 0:
|
294 |
+
return []
|
295 |
+
|
296 |
+
top_k = min(top_k, len(relevant_indices))
|
297 |
+
top_indices = relevant_indices[np.argsort(combined_sims[relevant_indices])[-top_k:][::-1]]
|
298 |
+
|
299 |
+
results = []
|
300 |
+
for idx in top_indices:
|
301 |
+
result = {'relevance_score': float(combined_sims[idx])}
|
302 |
+
for col in self.dataset.columns:
|
303 |
+
if col not in ['video_embed', 'description_embed', 'audio_embed']:
|
304 |
+
result[col] = self.dataset.iloc[idx][col]
|
305 |
+
results.append(result)
|
306 |
|
307 |
+
return results
|
308 |
+
|
309 |
+
def search_dataset(dataset_id, search_text, include_configs=None, include_splits=None):
|
310 |
+
dataset_info = fetch_dataset_info(dataset_id)
|
311 |
+
if not dataset_info:
|
312 |
+
return pd.DataFrame(), [], []
|
313 |
+
|
314 |
+
configs = include_configs if include_configs else dataset_info.get('config_names', ['default'])
|
315 |
+
all_rows = []
|
316 |
+
available_splits = set()
|
317 |
+
|
318 |
+
for config in configs:
|
319 |
+
try:
|
320 |
+
splits_url = f"https://datasets-server.huggingface.co/splits?dataset={dataset_id}&config={config}"
|
321 |
+
splits_response = requests.get(splits_url, timeout=30)
|
322 |
+
if splits_response.status_code == 200:
|
323 |
+
splits_data = splits_response.json()
|
324 |
+
splits = [split['split'] for split in splits_data.get('splits', [])]
|
325 |
+
if not splits:
|
326 |
+
splits = ['train']
|
327 |
+
|
328 |
+
if include_splits:
|
329 |
+
splits = [s for s in splits if s in include_splits]
|
330 |
+
|
331 |
+
available_splits.update(splits)
|
332 |
+
|
333 |
+
for split in splits:
|
334 |
+
rows = fetch_dataset_rows(dataset_id, config, split)
|
335 |
+
for row in rows:
|
336 |
+
text_content = ' '.join(str(v) for v in row.values()
|
337 |
+
if isinstance(v, (str, int, float)))
|
338 |
+
if search_text.lower() in text_content.lower():
|
339 |
+
row['_matched_text'] = text_content
|
340 |
+
row['_relevance_score'] = text_content.lower().count(search_text.lower())
|
341 |
+
all_rows.append(row)
|
342 |
+
except Exception as e:
|
343 |
+
st.warning(f"Error processing config {config}: {e}")
|
344 |
+
continue
|
345 |
+
|
346 |
+
if all_rows:
|
347 |
+
df = pd.DataFrame(all_rows)
|
348 |
+
df = df.sort_values('_relevance_score', ascending=False)
|
349 |
+
return df, configs, list(available_splits)
|
350 |
+
|
351 |
+
return pd.DataFrame(), configs, list(available_splits)
|
352 |
+
|
353 |
+
@st.cache_resource
|
354 |
+
def get_speech_model():
|
355 |
+
return edge_tts.Communicate
|
356 |
+
|
357 |
+
async def generate_speech(text, voice=None):
|
358 |
+
if not text.strip():
|
359 |
+
return None
|
360 |
+
if not voice:
|
361 |
+
voice = st.session_state['tts_voice']
|
362 |
+
try:
|
363 |
+
communicate = get_speech_model()(text, voice)
|
364 |
+
audio_file = f"speech_{datetime.now().strftime('%Y%m%d_%H%M%S')}.mp3"
|
365 |
+
await communicate.save(audio_file)
|
366 |
+
return audio_file
|
367 |
+
except Exception as e:
|
368 |
+
st.error(f"Error generating speech: {e}")
|
369 |
+
return None
|
370 |
+
|
371 |
+
def transcribe_audio(audio_path):
|
372 |
+
"""Placeholder for ASR implementation"""
|
373 |
+
return "ASR not implemented. Add your preferred speech recognition here!"
|
374 |
+
|
375 |
+
def arxiv_search(query, max_results=5):
|
376 |
+
base_url = "http://export.arxiv.org/api/query?"
|
377 |
+
search_url = base_url + f"search_query={quote(query)}&start=0&max_results={max_results}"
|
378 |
+
try:
|
379 |
+
r = requests.get(search_url)
|
380 |
+
if r.status_code == 200:
|
381 |
+
root = ET.fromstring(r.text)
|
382 |
+
ns = {'atom': 'http://www.w3.org/2005/Atom'}
|
383 |
+
entries = root.findall('atom:entry', ns)
|
384 |
+
results = []
|
385 |
+
for entry in entries:
|
386 |
+
title = entry.find('atom:title', ns).text.strip()
|
387 |
+
summary = entry.find('atom:summary', ns).text.strip()
|
388 |
+
link = next((l.get('href') for l in entry.findall('atom:link', ns)
|
389 |
+
if l.get('type') == 'text/html'), None)
|
390 |
+
results.append((title, summary, link))
|
391 |
+
return results
|
392 |
+
except Exception as e:
|
393 |
+
st.error(f"ArXiv search error: {e}")
|
394 |
+
return []
|
395 |
+
|
396 |
+
def show_file_manager():
|
397 |
+
st.subheader("π File Manager")
|
398 |
+
col1, col2 = st.columns(2)
|
399 |
+
|
400 |
+
with col1:
|
401 |
+
uploaded_file = st.file_uploader("Upload File", type=['txt', 'md', 'mp3'])
|
402 |
+
if uploaded_file:
|
403 |
+
with open(uploaded_file.name, "wb") as f:
|
404 |
+
f.write(uploaded_file.getvalue())
|
405 |
+
st.success(f"Uploaded: {uploaded_file.name}")
|
406 |
+
st.experimental_rerun()
|
407 |
+
|
408 |
+
with col2:
|
409 |
+
if st.button("π Clear Files"):
|
410 |
+
for f in glob.glob("*.txt") + glob.glob("*.md") + glob.glob("*.mp3"):
|
411 |
+
os.remove(f)
|
412 |
+
st.success("All files cleared!")
|
413 |
+
st.experimental_rerun()
|
414 |
+
|
415 |
+
files = glob.glob("*.txt") + glob.glob("*.md") + glob.glob("*.mp3")
|
416 |
+
if files:
|
417 |
+
st.write("### Existing Files")
|
418 |
+
for f in files:
|
419 |
+
with st.expander(f"π {os.path.basename(f)}"):
|
420 |
+
if f.endswith('.mp3'):
|
421 |
+
st.audio(f)
|
422 |
+
else:
|
423 |
+
with open(f, 'r', encoding='utf-8') as file:
|
424 |
+
st.text_area("Content", file.read(), height=100)
|
425 |
+
if st.button(f"Delete {os.path.basename(f)}", key=f"del_{f}"):
|
426 |
+
os.remove(f)
|
427 |
+
st.experimental_rerun()
|
428 |
+
|
429 |
+
def perform_arxiv_lookup(query, vocal_summary=True, titles_summary=True, full_audio=False):
|
430 |
+
results = arxiv_search(query, max_results=5)
|
431 |
+
if not results:
|
432 |
+
st.write("No results found.")
|
433 |
+
return
|
434 |
+
|
435 |
+
st.markdown(f"**ArXiv Results for '{query}':**")
|
436 |
+
for i, (title, summary, link) in enumerate(results, start=1):
|
437 |
+
st.markdown(f"**{i}. {title}**")
|
438 |
+
st.write(summary)
|
439 |
+
if link:
|
440 |
+
st.markdown(f"[View Paper]({link})")
|
441 |
+
|
442 |
+
if vocal_summary:
|
443 |
+
spoken_text = f"Here are ArXiv results for {query}. "
|
444 |
+
if titles_summary:
|
445 |
+
spoken_text += " Titles: " + ", ".join([res[0] for res in results])
|
446 |
+
else:
|
447 |
+
spoken_text += " " + results[0][1][:200]
|
448 |
+
|
449 |
+
audio_file = asyncio.run(generate_speech(spoken_text))
|
450 |
+
if audio_file:
|
451 |
+
st.audio(audio_file)
|
452 |
+
|
453 |
+
if full_audio:
|
454 |
+
full_text = ""
|
455 |
+
for i, (title, summary, _) in enumerate(results, start=1):
|
456 |
+
full_text += f"Result {i}: {title}. {summary} "
|
457 |
+
audio_file_full = asyncio.run(generate_speech(full_text))
|
458 |
+
if audio_file_full:
|
459 |
+
st.write("### Full Audio Summary")
|
460 |
+
st.audio(audio_file_full)
|
461 |
|
462 |
def render_result(result):
|
463 |
+
score = result.get('relevance_score', 0)
|
464 |
+
result_filtered = {k: v for k, v in result.items()
|
465 |
+
if k not in ['relevance_score', 'video_embed', 'description_embed', 'audio_embed']}
|
|
|
466 |
|
|
|
467 |
if 'youtube_id' in result:
|
468 |
+
st.video(f"https://youtube.com/watch?v={result['youtube_id']}&t={result.get('start_time', 0)}")
|
|
|
|
|
469 |
|
|
|
470 |
cols = st.columns([2, 1])
|
471 |
with cols[0]:
|
472 |
for key, value in result_filtered.items():
|
|
|
477 |
st.metric("Relevance Score", f"{score:.2%}")
|
478 |
|
479 |
def main():
|
480 |
+
st.title("π₯ Advanced Video & Dataset Search with Voice")
|
481 |
|
482 |
+
# Initialize search
|
483 |
+
search = VideoSearch()
|
484 |
|
485 |
+
# Create tabs
|
486 |
+
tab1, tab2, tab3, tab4 = st.tabs([
|
487 |
+
"π Search", "ποΈ Voice Input", "π ArXiv", "π Files"
|
488 |
+
])
|
489 |
+
|
490 |
+
# Search Tab
|
491 |
+
with tab1:
|
492 |
+
st.subheader("Search Videos")
|
493 |
+
col1, col2 = st.columns([3, 1])
|
494 |
+
with col1:
|
495 |
+
query = st.text_input("Enter search query:",
|
496 |
+
value="" if st.session_state['initial_search_done'] else "aliens")
|
497 |
+
with col2:
|
498 |
+
search_column = st.selectbox("Search in:",
|
499 |
+
["All Fields"] + st.session_state['search_columns'])
|
500 |
+
|
501 |
+
col3, col4 = st.columns(2)
|
502 |
+
with col3:
|
503 |
+
num_results = st.slider("Max results:", 1, 100, 20)
|
504 |
+
with col4:
|
505 |
+
search_button = st.button("π Search")
|
506 |
|
507 |
+
if (search_button or not st.session_state['initial_search_done']) and query:
|
508 |
+
st.session_state['initial_search_done'] = True
|
509 |
+
selected_column = None if search_column == "All Fields" else search_column
|
510 |
+
|
511 |
+
with st.spinner("Searching..."):
|
512 |
+
results = search.search(query, selected_column, num_results)
|
513 |
+
|
514 |
+
if results:
|
515 |
+
st.session_state['search_history'].append({
|
516 |
+
'query': query,
|
517 |
+
'timestamp': datetime.now().strftime("%Y-%m-%d %H:%M:%S"),
|
518 |
+
'results': results[:5]
|
519 |
+
})
|
520 |
+
|
521 |
+
st.write(f"Found {len(results)} results:")
|
522 |
+
for i, result in enumerate(results, 1):
|
523 |
+
with st.expander(f"Result {i}", expanded=(i==1)):
|
524 |
+
render_result(result)
|
525 |
+
else:
|
526 |
+
st.warning("No matching results found.")
|
527 |
|
528 |
+
# Voice Input Tab
|
529 |
+
with tab2:
|
530 |
+
st.subheader("Voice Search")
|
531 |
+
st.write("ποΈ Record your query:")
|
532 |
+
audio_bytes = audio_recorder()
|
533 |
+
if audio_bytes:
|
534 |
+
with st.spinner("Processing audio..."):
|
535 |
+
audio_path = f"temp_audio_{datetime.now().strftime('%Y%m%d_%H%M%S')}.wav"
|
536 |
+
with open(audio_path, "wb") as f:
|
537 |
+
f.write(audio_bytes)
|
538 |
+
|
539 |
+
voice_query = transcribe_audio(audio_path)
|
540 |
+
st.markdown("**Transcribed Text:**")
|
541 |
+
st.write(voice_query)
|
542 |
+
st.session_state['last_voice_input'] = voice_query
|
543 |
+
|
544 |
+
if st.button("π Search from Voice"):
|
545 |
+
results = search.search(voice_query, None, 20)
|
546 |
+
for i, result in enumerate(results, 1):
|
547 |
+
with st.expander(f"Result {i}", expanded=(i==1)):
|
548 |
+
render_result(result)
|
549 |
+
|
550 |
+
if os.path.exists(audio_path):
|
551 |
+
os.remove(audio_path)
|
552 |
|
553 |
+
# ArXiv Tab
|
554 |
+
with tab3:
|
555 |
+
st.subheader("ArXiv Search")
|
556 |
+
arxiv_query = st.text_input("Search ArXiv:", value=st.session_state['arxiv_last_query'])
|
557 |
+
vocal_summary = st.checkbox("π Quick Audio Summary", value=True)
|
558 |
+
titles_summary = st.checkbox("π Titles Only", value=True)
|
559 |
+
full_audio = st.checkbox("π Full Audio Summary", value=False)
|
560 |
+
|
561 |
+
if st.button("π Search ArXiv"):
|
562 |
+
st.session_state['arxiv_last_query'] = arxiv_query
|
563 |
+
perform_arxiv_lookup(arxiv_query, vocal_summary, titles_summary, full_audio)
|
564 |
|
565 |
+
# File Manager Tab
|
566 |
+
with tab4:
|
567 |
+
show_file_manager()
|
|
|
|
|
|
|
|
|
568 |
|
569 |
+
# Sidebar
|
570 |
+
with st.sidebar:
|
571 |
+
st.subheader("βοΈ Settings & History")
|
572 |
+
if st.button("ποΈ Clear History"):
|
573 |
+
st.session_state['search_history'] = []
|
574 |
+
st.experimental_rerun()
|
575 |
+
|
576 |
+
st.markdown("### Recent Searches")
|
577 |
+
for entry in reversed(st.session_state['search_history'][-5:]):
|
578 |
+
with st.expander(f"{entry['timestamp']}: {entry['query']}"):
|
579 |
+
for i, result in enumerate(entry['results'], 1):
|
580 |
+
st.write(f"{i}. {result.get('description', '')[:100]}...")
|
581 |
+
|
582 |
+
st.markdown("### Voice Settings")
|
583 |
+
st.selectbox("TTS Voice:", [
|
584 |
+
"en-US-AriaNeural",
|
585 |
+
"en-US-GuyNeural",
|
586 |
+
"en-GB-SoniaNeural"
|
587 |
+
], key="tts_voice")
|
|
|
|
|
|
|
|
|
|
|
588 |
|
589 |
if __name__ == "__main__":
|
590 |
main()
|