Self_correcting_coding_agent
1.0.0
В этом репозитории реализован агент LLM, который получает приглашение пользователя и соответствующие модульные тесты. Агент генерирует код, проверяет его правильность выполнения и обеспечивает прохождение модульных тестов. Если код не выполняется или не проходит какой-либо модульный тест, агент регенерирует код до тех пор, пока он не завершится успешно.
Этот проект основан на коде из LangChain's Notebook (https://github.com/mistralai/cookbook/blob/main/ Third_party/langchain/langgraph_code_assistant_mistral.ipynb), где агент генерирует код на основе запроса пользователя и проверяет, выполняется ли работает.
Основной вклад этого проекта — интеграция модульных тестов в процесс выполнения кода.
Кроме того, исходный код блокнота Jupyter был преобразован в структурированный проект Python, пригодный для развертывания. Придерживаясь лучших практик кодирования, код был модульным, были созданы многократно используемые функции, а также обеспечена надлежащая документация и тестирование. В результате получается хорошо организованный проект .py, который легко обслуживать и готов к работе.
git clone https : // github . com / paulomuraroferreira / Self_correcting_coding_agent . git
cd Self_correcting_coding_agent$ pip install - e .Создайте файл .env и заполните следующие переменные среды:
OPENAI_API_KEY = your_openai_api_key
OPENAI_CHAT_MODEL = "gpt-4o-2024-08-06"Введите приглашение кода в файл main.py. Подсказка должна быть описательной и содержать название классов.
Например:
QUESTION3 = '''
Python Class Description
Class Name: BankAccount
Description:
The BankAccount class represents a user's bank account, allowing for deposits, withdrawals, and viewing the transaction history. The class ensures that withdrawals cannot exceed the current balance and that deposits and withdrawals are properly recorded in the transaction history.
Attributes:
balance (float): The current balance of the account, initialized to 0.
transactions (list): A list to store the history of transactions. Each transaction is stored as a dictionary with keys type (either 'deposit' or 'withdrawal'), amount, and date.
Methods:
deposit(amount: float) -> None: Adds the specified amount to the balance and records the transaction.
withdraw(amount: float) -> bool: Attempts to subtract the specified amount from the balance. Returns True if successful, otherwise returns False. Records the transaction if successful.
get_balance() -> float: Returns the current balance.
get_transaction_history() -> list: Returns a list of all transactions.
'''
main = Main(QUESTION3)
Добавьте код модульного теста в файл .py в папке src/unit_test_folder. Также импортируйте соответствующий класс из файла code_solution.py. Например, для приведенного выше запроса LLM сгенерирует класс BankAccount. Следовательно, src/unit_test_folder/test_unit_test.py должен включать:
из src.code_solution импортировать банковский счет
Запустите файл main.py.
python main . py В приведенном выше примере код отобразит следующий вывод:
================================ Human Message =================================
Python Class Description
Class Name: BankAccount
Description:
The BankAccount class represents a user's bank account, allowing for deposits, withdrawals, and viewing the transaction history. The class ensures that withdrawals cannot exceed the current balance and that deposits and withdrawals are properly recorded in the transaction history.
Attributes:
balance (float): The current balance of the account, initialized to 0.
transactions (list): A list to store the history of transactions. Each transaction is stored as a dictionary with keys type (either 'deposit' or 'withdrawal'), amount, and date.
Methods:
deposit(amount: float) -> None: Adds the specified amount to the balance and records the transaction.
withdraw(amount: float) -> bool: Attempts to subtract the specified amount from the balance. Returns True if successful, otherwise returns False. Records the transaction if successful.
get_balance() -> float: Returns the current balance.
get_transaction_history() -> list: Returns a list of all transactions.
2024-08-18 14:51:24,955 - INFO - agent.py - ---GENERATING CODE SOLUTION---
================================== Ai Message ==================================
Here is my attempt to solve the problem: The BankAccount class is designed to manage a user's bank account by allowing deposits, withdrawals, and tracking transaction history. It ensures that withdrawals do not exceed the available balance and records each transaction with details such as type, amount, and date.
Imports: from datetime import datetime
Code: class BankAccount:
def __init__(self):
self.balance = 0.0
self.transactions = []
def deposit(self, amount: float) -> None:
self.balance += amount
transaction = {
'type': 'deposit',
'amount': amount,
'date': datetime.now()
}
self.transactions.append(transaction)
def withdraw(self, amount: float) -> bool:
if amount > self.balance:
return False
self.balance -= amount
transaction = {
'type': 'withdrawal',
'amount': amount,
'date': datetime.now()
}
self.transactions.append(transaction)
return True
def get_balance(self) -> float:
return self.balance
def get_transaction_history(self) -> list:
return self.transactions
2024-08-18 14:51:28,979 - INFO - agent.py - ---CHECKING CODE---
2024-08-18 14:51:29,170 - INFO - agent.py -
============================= test session starts ==============================
platform linux -- Python 3.12.3, pytest-8.3.2, pluggy-1.5.0 -- /home/paulo/Python_projects/Self_correcting_coding_agent/venv/bin/python
cachedir: .pytest_cache
rootdir: /home/paulo/Python_projects/Self_correcting_coding_agent
configfile: pyproject.toml
plugins: anyio-4.4.0
collecting ... collected 7 items
src/unit_test_folder/test_unit_test.py::test_initial_balance PASSED [ 14%]
src/unit_test_folder/test_unit_test.py::test_deposit PASSED [ 28%]
src/unit_test_folder/test_unit_test.py::test_withdrawal_successful PASSED [ 42%]
src/unit_test_folder/test_unit_test.py::test_withdrawal_insufficient_funds PASSED [ 57%]
src/unit_test_folder/test_unit_test.py::test_transaction_history PASSED [ 71%]
src/unit_test_folder/test_unit_test.py::test_invalid_deposit FAILED [ 85%]
src/unit_test_folder/test_unit_test.py::test_invalid_withdrawal FAILED [100%]
=================================== FAILURES ===================================
_____________________________ test_invalid_deposit _____________________________
account = <src.code_solution.BankAccount object at 0x7fac0d4f0b90>
def test_invalid_deposit(account):
"""Test that depositing a negative amount raises ValueError."""
> with pytest.raises(ValueError):
E Failed: DID NOT RAISE <class 'ValueError'>
src/unit_test_folder/test_unit_test.py:48: Failed
___________________________ test_invalid_withdrawal ____________________________
account = <src.code_solution.BankAccount object at 0x7fac0d4f1f40>
def test_invalid_withdrawal(account):
"""Test that withdrawing a negative amount raises ValueError."""
> with pytest.raises(ValueError):
E Failed: DID NOT RAISE <class 'ValueError'>
src/unit_test_folder/test_unit_test.py:53: Failed
=========================== short test summary info ============================
FAILED src/unit_test_folder/test_unit_test.py::test_invalid_deposit - Failed:...
FAILED src/unit_test_folder/test_unit_test.py::test_invalid_withdrawal - Fail...
========================= 2 failed, 5 passed in 0.02s ==========================
2024-08-18 14:51:29,170 - INFO - agent.py - Some tests failed.
2024-08-18 14:51:29,172 - INFO - agent.py - ---DECISION: RE-TRY SOLUTION---
================================ Human Message =================================
Your solution failed the unit test: ============================= test session starts ==============================
platform linux -- Python 3.12.3, pytest-8.3.2, pluggy-1.5.0 -- /home/paulo/Python_projects/Self_correcting_coding_agent/venv/bin/python
cachedir: .pytest_cache
rootdir: /home/paulo/Python_projects/Self_correcting_coding_agent
configfile: pyproject.toml
plugins: anyio-4.4.0
collecting ... collected 7 items
src/unit_test_folder/test_unit_test.py::test_initial_balance PASSED [ 14%]
src/unit_test_folder/test_unit_test.py::test_deposit PASSED [ 28%]
src/unit_test_folder/test_unit_test.py::test_withdrawal_successful PASSED [ 42%]
src/unit_test_folder/test_unit_test.py::test_withdrawal_insufficient_funds PASSED [ 57%]
src/unit_test_folder/test_unit_test.py::test_transaction_history PASSED [ 71%]
src/unit_test_folder/test_unit_test.py::test_invalid_deposit FAILED [ 85%]
src/unit_test_folder/test_unit_test.py::test_invalid_withdrawal FAILED [100%]
=================================== FAILURES ===================================
_____________________________ test_invalid_deposit _____________________________
account = <src.code_solution.BankAccount object at 0x7fac0d4f0b90>
def test_invalid_deposit(account):
"""Test that depositing a negative amount raises ValueError."""
> with pytest.raises(ValueError):
E Failed: DID NOT RAISE <c ... (truncated)
2024-08-18 14:51:29,175 - INFO - agent.py - ---GENERATING CODE SOLUTION---
================================== Ai Message ==================================
Here is my attempt to solve the problem: The previous solution did not handle cases where negative amounts were deposited or withdrawn. The unit tests expect a ValueError to be raised in such cases, but the original implementation did not include this validation. To fix this, we need to add checks in the deposit and withdraw methods to raise a ValueError if the amount is negative.
Imports: from datetime import datetime
Code: class BankAccount:
def __init__(self):
self.balance = 0.0
self.transactions = []
def deposit(self, amount: float) -> None:
if amount < 0:
raise ValueError("Deposit amount must be positive.")
self.balance += amount
transaction = {
'type': 'deposit',
'amount': amount,
'date': datetime.now()
}
self.transactions.append(transaction)
def withdraw(self, amount: float) -> bool:
if amount < 0:
raise ValueError("Withdrawal amount must be positive.")
if amount > self.balance:
return False
self.balance -= amount
transaction = {
'type': 'withdrawal',
'amount': amount,
'date': datetime.now()
}
self.transactions.append(transaction)
return True
def get_balance(self) -> float:
return self.balance
def get_transaction_history(self) -> ... (truncated)
2024-08-18 14:51:34,913 - INFO - agent.py - ---CHECKING CODE---
2024-08-18 14:51:35,131 - INFO - agent.py -
============================= test session starts ==============================
platform linux -- Python 3.12.3, pytest-8.3.2, pluggy-1.5.0 -- /home/paulo/Python_projects/Self_correcting_coding_agent/venv/bin/python
cachedir: .pytest_cache
rootdir: /home/paulo/Python_projects/Self_correcting_coding_agent
configfile: pyproject.toml
plugins: anyio-4.4.0
collecting ... collected 7 items
src/unit_test_folder/test_unit_test.py::test_initial_balance PASSED [ 14%]
src/unit_test_folder/test_unit_test.py::test_deposit PASSED [ 28%]
src/unit_test_folder/test_unit_test.py::test_withdrawal_successful PASSED [ 42%]
src/unit_test_folder/test_unit_test.py::test_withdrawal_insufficient_funds PASSED [ 57%]
src/unit_test_folder/test_unit_test.py::test_transaction_history PASSED [ 71%]
src/unit_test_folder/test_unit_test.py::test_invalid_deposit PASSED [ 85%]
src/unit_test_folder/test_unit_test.py::test_invalid_withdrawal PASSED [100%]
============================== 7 passed in 0.01s ===============================
2024-08-18 14:51:35,131 - INFO - agent.py - All tests passed!
2024-08-18 14:51:35,131 - INFO - agent.py - ---NO CODE TEST FAILURES---
2024-08-18 14:51:35,133 - INFO - agent.py - ---DECISION: FINISH---
