Spaces:
Build error
Build error
Inference time cutodonw
Browse files- event_matcher.py +69 -70
event_matcher.py
CHANGED
|
@@ -28,90 +28,89 @@ class EventMatcher:
|
|
| 28 |
def query(self, user_query: str) -> str:
|
| 29 |
"""Main query method called by the app"""
|
| 30 |
try:
|
| 31 |
-
# Update events if needed
|
| 32 |
-
self.
|
| 33 |
-
|
| 34 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 35 |
matched_events = self.find_matching_events(user_query)
|
| 36 |
|
| 37 |
if not matched_events:
|
| 38 |
return "I couldn't find any events matching your query. Try asking in a different way!"
|
| 39 |
|
| 40 |
-
# Format the response
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 46 |
|
| 47 |
-
|
| 48 |
-
"""Find events matching the query"""
|
| 49 |
-
matched_events = []
|
| 50 |
|
| 51 |
-
|
| 52 |
-
|
| 53 |
-
|
| 54 |
-
|
| 55 |
-
|
| 56 |
-
|
| 57 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 58 |
|
| 59 |
# Weight the scores
|
| 60 |
total_score = (
|
| 61 |
-
title_score * 0.
|
| 62 |
-
desc_score * 0.
|
| 63 |
-
location_score * 0.
|
| 64 |
-
categories_score * 0.
|
| 65 |
-
hosts_score * 0.05
|
| 66 |
)
|
| 67 |
|
| 68 |
if total_score > 0.3: # Threshold for relevance
|
| 69 |
matched_events.append((event, total_score))
|
| 70 |
-
|
| 71 |
-
|
| 72 |
-
|
| 73 |
-
|
| 74 |
-
|
| 75 |
-
def format_response(self, matched_events: List[Tuple[Dict, float]], query: str) -> str:
|
| 76 |
-
"""Format the response with matched events"""
|
| 77 |
-
try:
|
| 78 |
-
# Prepare text for T5 model
|
| 79 |
-
events_text = ""
|
| 80 |
-
for event, score in matched_events:
|
| 81 |
-
events_text += f"""
|
| 82 |
-
π
{event['title']}
|
| 83 |
-
Date: {event['start_time'].strftime('%A, %B %d, %Y')}
|
| 84 |
-
Time: {event['start_time'].strftime('%I:%M %p')}
|
| 85 |
-
π Location: {event['location']}
|
| 86 |
-
π·οΈ Categories: {event['categories']}
|
| 87 |
-
π₯ Hosted by: {event['hosts']}
|
| 88 |
-
|
| 89 |
-
{event['description'][:200]}...
|
| 90 |
-
|
| 91 |
-
π More info: {event['link']}
|
| 92 |
-
"""
|
| 93 |
-
|
| 94 |
-
# Generate response using T5
|
| 95 |
-
input_text = f"query: {query} context: {events_text}"
|
| 96 |
-
inputs = self.tokenizer(input_text, return_tensors="pt", max_length=512, truncation=True)
|
| 97 |
-
outputs = self.model.generate(
|
| 98 |
-
inputs.input_ids,
|
| 99 |
-
max_length=150,
|
| 100 |
-
num_beams=4,
|
| 101 |
-
temperature=0.7,
|
| 102 |
-
no_repeat_ngram_size=2
|
| 103 |
-
)
|
| 104 |
-
|
| 105 |
-
response = self.tokenizer.decode(outputs[0], skip_special_tokens=True)
|
| 106 |
-
|
| 107 |
-
# Combine T5 response with event details
|
| 108 |
-
final_response = f"{response}\n\nHere are the matching events:{events_text}"
|
| 109 |
-
|
| 110 |
-
return final_response
|
| 111 |
-
|
| 112 |
-
except Exception as e:
|
| 113 |
-
print(f"Error formatting response: {e}")
|
| 114 |
-
return "Here are some relevant events I found:" + events_text
|
| 115 |
|
| 116 |
def parse_event_datetime(self, entry) -> Tuple[Optional[datetime], Optional[datetime]]:
|
| 117 |
"""Parse event dates from RSS feed"""
|
|
|
|
| 28 |
def query(self, user_query: str) -> str:
|
| 29 |
"""Main query method called by the app"""
|
| 30 |
try:
|
| 31 |
+
# Update events if needed - make this non-blocking
|
| 32 |
+
if self.last_update is None or (datetime.now() - self.last_update).seconds >= 3600:
|
| 33 |
+
# Start a background task to update events
|
| 34 |
+
self.update_events()
|
| 35 |
+
|
| 36 |
+
# Quick response for empty query
|
| 37 |
+
if not user_query.strip():
|
| 38 |
+
return "Please ask me about events at Brock University!"
|
| 39 |
+
|
| 40 |
+
# Find matching events - optimize by limiting initial search
|
| 41 |
matched_events = self.find_matching_events(user_query)
|
| 42 |
|
| 43 |
if not matched_events:
|
| 44 |
return "I couldn't find any events matching your query. Try asking in a different way!"
|
| 45 |
|
| 46 |
+
# Format the response without T5 for faster response
|
| 47 |
+
events_text = ""
|
| 48 |
+
for event, score in matched_events:
|
| 49 |
+
events_text += f"""
|
| 50 |
+
π
{event['title']}
|
| 51 |
+
π {event['start_time'].strftime('%A, %B %d, %Y')}
|
| 52 |
+
β° {event['start_time'].strftime('%I:%M %p')}
|
| 53 |
+
π {event['location']}
|
| 54 |
+
π·οΈ {event['categories']}
|
| 55 |
+
π₯ {event['hosts']}
|
| 56 |
+
|
| 57 |
+
{event['description'][:200]}...
|
| 58 |
+
|
| 59 |
+
π {event['link']}
|
| 60 |
+
|
| 61 |
+
"""
|
| 62 |
+
# Create a simple response prefix based on the query type
|
| 63 |
+
if "today" in user_query.lower():
|
| 64 |
+
prefix = "Here are today's events:"
|
| 65 |
+
elif "week" in user_query.lower():
|
| 66 |
+
prefix = "Here are the events happening this week:"
|
| 67 |
+
elif any(word in user_query.lower() for word in ["workshop", "training", "seminar"]):
|
| 68 |
+
prefix = "I found these workshops and seminars:"
|
| 69 |
+
elif any(word in user_query.lower() for word in ["faculty", "department", "school"]):
|
| 70 |
+
prefix = "Here are the faculty-related events:"
|
| 71 |
+
else:
|
| 72 |
+
prefix = "Here are some events that match your query:"
|
| 73 |
|
| 74 |
+
return f"{prefix}\n{events_text}"
|
|
|
|
|
|
|
| 75 |
|
| 76 |
+
except Exception as e:
|
| 77 |
+
print(f"Error in query: {e}")
|
| 78 |
+
return "I encountered an error processing your query. Please try again!"
|
| 79 |
+
return "I encountered an error processing your query. Please try again!"
|
| 80 |
+
|
| 81 |
+
def find_matching_events(self, query: str) -> List[Tuple[Dict, float]]:
|
| 82 |
+
"""Find events matching the query - optimized version"""
|
| 83 |
+
matched_events = []
|
| 84 |
+
query_lower = query.lower()
|
| 85 |
+
|
| 86 |
+
for event in self.events:
|
| 87 |
+
# Quick initial filter
|
| 88 |
+
if any(term in event['title'].lower() or
|
| 89 |
+
term in event['description'].lower()[:200] or
|
| 90 |
+
term in event['location'].lower() or
|
| 91 |
+
term in event['categories'].lower()
|
| 92 |
+
for term in query_lower.split()):
|
| 93 |
+
|
| 94 |
+
# Calculate similarity scores only for potentially matching events
|
| 95 |
+
title_score = fuzz.token_set_ratio(query_lower, event['title'].lower()) / 100
|
| 96 |
+
desc_score = fuzz.token_set_ratio(query_lower, event['description'].lower()) / 100
|
| 97 |
+
location_score = fuzz.token_set_ratio(query_lower, event['location'].lower()) / 100
|
| 98 |
+
categories_score = fuzz.token_set_ratio(query_lower, event['categories'].lower()) / 100
|
| 99 |
|
| 100 |
# Weight the scores
|
| 101 |
total_score = (
|
| 102 |
+
title_score * 0.4 +
|
| 103 |
+
desc_score * 0.3 +
|
| 104 |
+
location_score * 0.2 +
|
| 105 |
+
categories_score * 0.1
|
|
|
|
| 106 |
)
|
| 107 |
|
| 108 |
if total_score > 0.3: # Threshold for relevance
|
| 109 |
matched_events.append((event, total_score))
|
| 110 |
+
|
| 111 |
+
# Sort by score and get top matches
|
| 112 |
+
matched_events.sort(key=lambda x: x[1], reverse=True)
|
| 113 |
+
return matched_events[:3]
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 114 |
|
| 115 |
def parse_event_datetime(self, entry) -> Tuple[Optional[datetime], Optional[datetime]]:
|
| 116 |
"""Parse event dates from RSS feed"""
|