Spaces:
Running
Running
Update app.py
Browse files
app.py
CHANGED
|
@@ -12,62 +12,82 @@ wolfram_client = wolframalpha.Client(os.environ.get('WOLFRAM_APPID'))
|
|
| 12 |
|
| 13 |
def parse_questions(content):
|
| 14 |
"""Parse questions and their solutions from Claude's output"""
|
| 15 |
-
# Split content into questions
|
| 16 |
questions = []
|
| 17 |
-
|
| 18 |
|
| 19 |
-
#
|
| 20 |
-
|
| 21 |
-
parts = parts[1:]
|
| 22 |
|
| 23 |
-
|
| 24 |
-
|
| 25 |
-
|
| 26 |
-
|
| 27 |
-
|
| 28 |
-
|
| 29 |
-
|
| 30 |
-
|
| 31 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 32 |
else:
|
| 33 |
-
|
| 34 |
|
| 35 |
-
|
| 36 |
-
|
| 37 |
-
|
| 38 |
-
|
| 39 |
-
|
| 40 |
-
|
| 41 |
-
|
| 42 |
-
|
| 43 |
-
|
| 44 |
-
|
| 45 |
-
|
| 46 |
-
|
| 47 |
-
|
| 48 |
-
|
| 49 |
-
except Exception as e:
|
| 50 |
-
print(f"Error parsing question {question_num}: {e}")
|
| 51 |
-
continue
|
| 52 |
|
| 53 |
return questions
|
| 54 |
|
| 55 |
def verify_solution(problem, answer):
|
| 56 |
"""Verify a mathematical solution using Wolfram Alpha"""
|
| 57 |
try:
|
| 58 |
-
#
|
| 59 |
-
query = problem.replace('$$', '').replace('$', '')
|
|
|
|
|
|
|
|
|
|
|
|
|
| 60 |
result = wolfram_client.query(query)
|
| 61 |
|
| 62 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 63 |
for pod in result.pods:
|
| 64 |
-
if pod.title in ['Result', 'Solution', 'Numerical result']:
|
| 65 |
wolfram_answer = pod.text
|
| 66 |
# Extract numerical value
|
| 67 |
wolfram_nums = re.findall(r'[-+]?(?:\d*\.)?\d+', wolfram_answer)
|
| 68 |
if wolfram_nums:
|
|
|
|
|
|
|
|
|
|
|
|
|
| 69 |
return {
|
| 70 |
-
'verified':
|
| 71 |
'wolfram_solution': wolfram_answer,
|
| 72 |
'error': None
|
| 73 |
}
|
|
@@ -75,13 +95,13 @@ def verify_solution(problem, answer):
|
|
| 75 |
return {
|
| 76 |
'verified': False,
|
| 77 |
'wolfram_solution': None,
|
| 78 |
-
'error': "
|
| 79 |
}
|
| 80 |
except Exception as e:
|
| 81 |
return {
|
| 82 |
'verified': False,
|
| 83 |
'wolfram_solution': None,
|
| 84 |
-
'error': str(e)
|
| 85 |
}
|
| 86 |
|
| 87 |
def generate_test(subject):
|
|
@@ -94,7 +114,9 @@ def generate_test(subject):
|
|
| 94 |
3. Include "Solution:" before the solution
|
| 95 |
4. Show step-by-step work
|
| 96 |
5. End each solution with "Final answer = [number]"
|
| 97 |
-
6. Keep problems relatively simple (basic calculus, algebra, etc.)
|
|
|
|
|
|
|
| 98 |
|
| 99 |
message = anthropic.messages.create(
|
| 100 |
model="claude-3-opus-20240229",
|
|
@@ -115,17 +137,20 @@ def generate_test(subject):
|
|
| 115 |
verification_results = []
|
| 116 |
|
| 117 |
for q in questions:
|
| 118 |
-
|
| 119 |
-
|
| 120 |
-
|
| 121 |
-
|
| 122 |
-
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 123 |
else:
|
| 124 |
-
verification_note += "⚠️
|
| 125 |
-
if result['wolfram_solution']:
|
| 126 |
-
verification_note += f"Wolfram Alpha result: {result['wolfram_solution']}\n"
|
| 127 |
-
if result['error']:
|
| 128 |
-
verification_note += f"Note: {result['error']}\n"
|
| 129 |
|
| 130 |
# Add usage statistics
|
| 131 |
usage_stats = f"""
|
|
|
|
| 12 |
|
| 13 |
def parse_questions(content):
|
| 14 |
"""Parse questions and their solutions from Claude's output"""
|
|
|
|
| 15 |
questions = []
|
| 16 |
+
current_question = {}
|
| 17 |
|
| 18 |
+
# Split content into lines for more reliable parsing
|
| 19 |
+
lines = content.split('\n')
|
|
|
|
| 20 |
|
| 21 |
+
for line in lines:
|
| 22 |
+
# Start of new question
|
| 23 |
+
if re.match(r'^\s*\d+\)', line):
|
| 24 |
+
if current_question:
|
| 25 |
+
questions.append(current_question)
|
| 26 |
+
current_question = {
|
| 27 |
+
'number': re.match(r'^\s*(\d+)\)', line).group(1),
|
| 28 |
+
'problem': line.split(')', 1)[1].strip(),
|
| 29 |
+
'solution': '',
|
| 30 |
+
'final_answer': None
|
| 31 |
+
}
|
| 32 |
+
# Solution marker
|
| 33 |
+
elif 'Solution:' in line and current_question:
|
| 34 |
+
current_question['problem'] = current_question['problem'].strip()
|
| 35 |
+
current_question['solution'] = line.split('Solution:', 1)[1].strip()
|
| 36 |
+
# Add to current problem or solution
|
| 37 |
+
elif current_question:
|
| 38 |
+
if current_question['solution']:
|
| 39 |
+
current_question['solution'] += '\n' + line
|
| 40 |
else:
|
| 41 |
+
current_question['problem'] += '\n' + line
|
| 42 |
|
| 43 |
+
# Extract final answer
|
| 44 |
+
if current_question and 'final answer' in line.lower():
|
| 45 |
+
matches = re.findall(r'[-+]?(?:\d*\.)?\d+', line)
|
| 46 |
+
if matches:
|
| 47 |
+
current_question['final_answer'] = matches[-1]
|
| 48 |
+
|
| 49 |
+
# Add last question
|
| 50 |
+
if current_question:
|
| 51 |
+
questions.append(current_question)
|
| 52 |
+
|
| 53 |
+
# Clean up questions
|
| 54 |
+
for q in questions:
|
| 55 |
+
q['problem'] = q['problem'].strip()
|
| 56 |
+
q['solution'] = q['solution'].strip()
|
|
|
|
|
|
|
|
|
|
| 57 |
|
| 58 |
return questions
|
| 59 |
|
| 60 |
def verify_solution(problem, answer):
|
| 61 |
"""Verify a mathematical solution using Wolfram Alpha"""
|
| 62 |
try:
|
| 63 |
+
# Clean up the problem for Wolfram Alpha
|
| 64 |
+
query = problem.replace('$$', '').replace('$', '')
|
| 65 |
+
# Remove any text instructions, keep only the mathematical expression
|
| 66 |
+
query = re.sub(r'(?i)find|calculate|solve|evaluate|determine', '', query)
|
| 67 |
+
query = query.strip()
|
| 68 |
+
|
| 69 |
result = wolfram_client.query(query)
|
| 70 |
|
| 71 |
+
if not result.success:
|
| 72 |
+
return {
|
| 73 |
+
'verified': False,
|
| 74 |
+
'wolfram_solution': None,
|
| 75 |
+
'error': "Wolfram Alpha could not process the query"
|
| 76 |
+
}
|
| 77 |
+
|
| 78 |
+
# Look for numerical results in multiple pods
|
| 79 |
for pod in result.pods:
|
| 80 |
+
if pod.title in ['Result', 'Solution', 'Numerical result', 'Decimal approximation']:
|
| 81 |
wolfram_answer = pod.text
|
| 82 |
# Extract numerical value
|
| 83 |
wolfram_nums = re.findall(r'[-+]?(?:\d*\.)?\d+', wolfram_answer)
|
| 84 |
if wolfram_nums:
|
| 85 |
+
wolfram_value = float(wolfram_nums[0])
|
| 86 |
+
user_value = float(answer)
|
| 87 |
+
# Allow for small numerical differences
|
| 88 |
+
is_verified = abs(wolfram_value - user_value) < 0.01
|
| 89 |
return {
|
| 90 |
+
'verified': is_verified,
|
| 91 |
'wolfram_solution': wolfram_answer,
|
| 92 |
'error': None
|
| 93 |
}
|
|
|
|
| 95 |
return {
|
| 96 |
'verified': False,
|
| 97 |
'wolfram_solution': None,
|
| 98 |
+
'error': "No numerical solution found in Wolfram Alpha response"
|
| 99 |
}
|
| 100 |
except Exception as e:
|
| 101 |
return {
|
| 102 |
'verified': False,
|
| 103 |
'wolfram_solution': None,
|
| 104 |
+
'error': f"Error during verification: {str(e)}"
|
| 105 |
}
|
| 106 |
|
| 107 |
def generate_test(subject):
|
|
|
|
| 114 |
3. Include "Solution:" before the solution
|
| 115 |
4. Show step-by-step work
|
| 116 |
5. End each solution with "Final answer = [number]"
|
| 117 |
+
6. Keep problems relatively simple (basic calculus, algebra, etc.)
|
| 118 |
+
7. Make sure problems have clear numerical answers
|
| 119 |
+
8. Avoid word problems - focus on pure mathematical expressions"""
|
| 120 |
|
| 121 |
message = anthropic.messages.create(
|
| 122 |
model="claude-3-opus-20240229",
|
|
|
|
| 137 |
verification_results = []
|
| 138 |
|
| 139 |
for q in questions:
|
| 140 |
+
if q['final_answer'] is not None:
|
| 141 |
+
result = verify_solution(q['problem'], q['final_answer'])
|
| 142 |
+
verification_results.append(result)
|
| 143 |
+
verification_note += f"\nQuestion {q['number']}:\n"
|
| 144 |
+
if result['verified']:
|
| 145 |
+
verification_note += "✅ Solution verified by Wolfram Alpha\n"
|
| 146 |
+
else:
|
| 147 |
+
verification_note += "⚠️ Solution needs verification\n"
|
| 148 |
+
if result['wolfram_solution']:
|
| 149 |
+
verification_note += f"Wolfram Alpha result: {result['wolfram_solution']}\n"
|
| 150 |
+
if result['error']:
|
| 151 |
+
verification_note += f"Note: {result['error']}\n"
|
| 152 |
else:
|
| 153 |
+
verification_note += f"\nQuestion {q['number']}:\n⚠️ Could not extract final answer\n"
|
|
|
|
|
|
|
|
|
|
|
|
|
| 154 |
|
| 155 |
# Add usage statistics
|
| 156 |
usage_stats = f"""
|