joshuarauh's picture
Update app.py
109daa0 verified
raw
history blame
6.4 kB
import os
import gradio as gr
from anthropic import Anthropic
import wolframalpha
from datetime import datetime, timedelta
from collections import deque
import re
# Initialize clients
anthropic = Anthropic(api_key=os.environ.get('ANTHROPIC_API_KEY'))
wolfram_client = wolframalpha.Client(os.environ.get('WOLFRAM_APPID'))
def parse_questions(content):
"""Parse questions and their solutions from Claude's output"""
# Split content into questions
questions = []
current_text = ""
question_pattern = re.compile(r'\d+\)')
# Split the content by question numbers
parts = re.split(question_pattern, content)
if len(parts) > 1: # Skip the first empty part if it exists
parts = parts[1:]
for part in parts:
# Try to extract the problem and solution
try:
# Split into problem and solution (assuming "Solution:" marks the divide)
problem_solution = part.split("Solution:", 1)
if len(problem_solution) == 2:
problem = problem_solution[0].strip()
solution = problem_solution[1].strip()
# Extract the final numerical answer if possible
# This is a simple example - you'll need to adjust based on your output format
final_answer = re.search(r'=\s*([-+]?\d*\.?\d+)', solution)
if final_answer:
final_answer = final_answer.group(1)
else:
final_answer = "Not found"
questions.append((problem, final_answer))
except Exception as e:
print(f"Error parsing question: {e}")
continue
return questions
def verify_solution(problem, claimed_solution):
"""Verify a mathematical solution using Wolfram Alpha"""
try:
# Clean up the problem and solution for Wolfram Alpha
query = f"Solve {problem}"
result = wolfram_client.query(query)
# Extract the solution from Wolfram Alpha
wolfram_solution = next(result.results).text
# Compare solutions (needs sophisticated parsing based on your problem types)
solutions_match = compare_solutions(wolfram_solution, claimed_solution)
return {
'verified': solutions_match,
'wolfram_solution': wolfram_solution,
'match': solutions_match
}
except Exception as e:
return {
'verified': False,
'error': str(e),
'wolfram_solution': None
}
def compare_solutions(wolfram_sol, claude_sol):
"""Compare two solutions for mathematical equivalence"""
try:
# Convert both solutions to floats for comparison
w_val = float(wolfram_sol)
c_val = float(claude_sol)
return abs(w_val - c_val) < 0.001
except (ValueError, TypeError):
return False
def generate_test(subject):
"""Generate and verify a math test"""
try:
# Generate the test using Claude
system_prompt = """Generate 3 university-level math questions with numerical solutions that can be verified.
For each question:
1. State the problem clearly
2. Provide your step-by-step solution
3. End each solution with a clear final numerical answer in the format: "Final answer = [number]"
Use simple $$ for all math expressions."""
message = anthropic.messages.create(
model="claude-3-opus-20240229",
max_tokens=1500,
temperature=0.7,
messages=[{
"role": "user",
"content": f"{system_prompt}\n\nWrite an exam for {subject}."
}]
)
# Extract questions and solutions
content = message.content[0].text
# Add verification results
verification_results = []
# Parse and verify each question
verification_note = "\n\n## Solution Verification:\n"
for i, (problem, solution) in enumerate(parse_questions(content)):
result = verify_solution(problem, solution)
verification_note += f"\nQuestion {i+1}:\n"
if result['verified']:
verification_note += "✅ Solution verified by Wolfram Alpha\n"
else:
verification_note += "⚠️ Solution needs verification\n"
if result['wolfram_solution']:
verification_note += f"Wolfram Alpha got: {result['wolfram_solution']}\n"
verification_results.append(result)
# Add usage statistics
usage_stats = f"""
\n---\nUsage Statistics:
• Input Tokens: {message.usage.input_tokens:,}
• Output Tokens: {message.usage.output_tokens:,}
• Wolfram Alpha calls: {len(verification_results)}
Cost Breakdown:
• Claude Cost: ${((message.usage.input_tokens / 1000) * 0.015) + ((message.usage.output_tokens / 1000) * 0.075):.4f}
• Wolfram API calls: {len(verification_results)}
"""
return content + verification_note + usage_stats
except Exception as e:
return f"Error: {str(e)}"
# Subject choices and interface configuration remain the same...
subjects = [
"Single Variable Calculus",
"Multivariable Calculus",
"Linear Algebra",
"Differential Equations",
"Real Analysis",
"Complex Analysis",
"Abstract Algebra",
"Probability Theory",
"Numerical Analysis",
"Topology"
]
# Create Gradio interface
interface = gr.Interface(
fn=generate_test,
inputs=gr.Dropdown(
choices=subjects,
label="Select Mathematics Subject",
info="Choose a subject for the exam questions"
),
outputs=gr.Markdown(
label="Generated Test",
latex_delimiters=[
{"left": "$$", "right": "$$", "display": True},
{"left": "$", "right": "$", "display": False}
]
),
title="Advanced Mathematics Test Generator",
description="""Generates university-level mathematics exam questions with solutions using Claude 3 Opus.
Limited to 25 requests per day. Please use responsibly.""",
theme="default",
allow_flagging="never"
)
# Launch the interface
if __name__ == "__main__":
interface.launch()