python-perf-env 0.1.0

A Python package providing an environment for AI agents to test their code and receive detailed profiling feedback on execution time and memory usage. Facilitates performance analysis and optimization during AI development.

python-perf-env

A Python package providing an environment for AI agents to test their code and receive detailed profiling feedback on execution time and memory usage.

Installation

Install the package using pip:

pip install python-perf-env

Quick Start / Get Started

This guide provides a quick introduction to using the python_perf_env package. It demonstrates how to create and interact with the SimpleEvaluator environment.

1. Import the necessary modules:

from python_perf_env import SimpleEvaluator

2. Create the environment:

You can create the environment using the default configuration:

env = SimpleEvaluator()

Or, you can customize the environment by providing a configuration dictionary:

# Say we want focused on optimizing algorithm on time
custom_config = {
    "max_input_len": 2048,
    "max_time_cost": 1,  # The maximum allowed execution time (in seconds)
    "max_memory_cost": 1 * GB,  # The maximum allowed memory usage (in bytes)
    "time_weight": 2, # Weight for time cost in reward calculation (default: 1).
    "memory_weight": 0.5, # Weight for memory cost in reward calculation (default: 1).
    "entry_point": "my_function",
}
env_custom = SimpleEvaluator(config=custom_config)

3. Reset the environment:

Reset the environment to get the initial observation:

env.reset()

4. Define the code to be executed:

The code you submit to the environment must include a function with the name specified in the entry_point configuration where we setted "my_function" in the custom_config (defaults to "env_main").

code = """
import time

def my_function():
    start_time = time.time()
    for _ in range(1000000):
        pass  # Simulate some work
    end_time = time.time()
    return end_time - start_time  # Return the execution time
"""

5. Take a step in the environment:

Submit the Python code as an action to the step() method:

observation, reward, terminated, truncated, info = env.step(code)

print("Observation:")
print(observation)
print("Reward:", reward)

Using Test-Driven Development (TDD) with TestDrivenEvaluator

The TestDrivenEvaluator allows you to integrate unit tests directly into the evaluation process. This is particularly useful for guiding the AI agent towards solutions that satisfy specific test cases.

:warning: Known issue: The class name for the unittest must be TestEnvMain(unittest.TestCase) otherwise test case will not run.

# Define unit tests
unittest_code = """
import unittest

class TestEnvMain(unittest.TestCase):
    def test_execution_time(self):
        self.assertLess(my_function(), 0.1)  # Adjust the threshold as needed
"""

# Configure the environment for TDD
tdd_config = {
    "max_input_len": 2048,
    "max_time_cost": 1,
    "max_memory_cost": 1 * GB,
    "exception_reward": -9,
    "unittest": unittest_code,
    "entry_point": "my_function"
}
env_tdd = TestDrivenEvaluator(config=tdd_config)

# Reset the environment
env_tdd.reset()

# Define the code (same as before, or modified to pass the tests)
code = """
import time

def my_function():
    start_time = time.time()
    for _ in range(100000): # Reduced iterations for faster testing
        pass
    end_time = time.time()
    return end_time - start_time
"""

# Take a step in the environment
observation, reward, terminated, truncated, info = env_tdd.step(code)

print("Observation (TDD):")
print(observation)
print("Reward (TDD):", reward)

Further Information

• Documentation: Refer to the docstrings within the code for detailed information about the SimpleEvaluator class, its attributes, and its methods.
• Security: Always prioritize security when using this environment as it does not provide encapsulated environment for code execution. Consider providing secure environment using container in future versions.
• Profiling Output: The format of the profiling output might need to be parsed by your AI agent.

License

This project is licensed under the MIT License. See the LICENSE file for details.