fixml 0.0.1

Package for automated test evaluation and creation

FixML

A tool for providing context-aware evaluations using a checklist-based approach on the Machine Learning project code bases.

Motivation

Testing codes in Machine Learning project mostly revolves around ensuring the findings are reproducible. To achieve this, currently it requires a lot of manual efforts. It is because such projects usually have assumptions that are hard to quantify in traditional software engineering approach i.e. code coverage. One such example would be testing the model's performance, which will not result in any errors, but we do expect this result to be reproducible by others. Testing such codes, therefore, require us to not only quantitatively, but also to qualitatively gauge how effective the tests are.

A common way to handle this currently is to utilize expertise from domain experts in this area. Researches and guidelines have been done on how to incorporate such knowledge through the use of checklists. However, this requires manually validating the checklist items which usually results in poor scalability and slow feedback loop for developers, which are incompatible with today's fast-paced, competitive landscape in ML developments.

This tool aims to bridge the gap between these two different approaches, by adding Large Language Models (LLMs) into the loop, given LLMs' recent advancement in multiple areas including NLU tasks and code-related tasks. They have been shown to some degrees the ability to analyze codes and to produce context-aware suggestions. This tool simplifies such workflow by providing a command line tool as well as a high-level API for developers and researchers alike to quickly validate if their tests satisfy common areas that are required for reproducibility purposes.

Given LLMs' tendency to provide plausible but factually incorrect information, extensive analyses have been done on ensuring the responses are aligned with ground truths and human expectations both accurately and consistently. Based on these analyses, we are also able to continuously refine our prompts and workflows.

Installation

This tool is on PyPI. To install, please run:

$ pip install fixml

Usage

CLI tool

Once installed, the tool offers a Command Line Interface (CLI) command fixml. By using this command you will be able to evaluate your project code bases, generate test function specifications, and perform various relevant tasks.

Run fixml --help for more details.

[!IMPORTANT] By default, this tool uses OpenAI's gpt3.5-turbo for evaluation. To run any command that requires calls to LLM (i.e. fixml evaluate, fixml generate), an environment variable OPENAI_API_KEY needs to be set. To do so, either use export to set the variable in your current session, or create a .env file with a line OPENAI_API_KEY={your-api-key} saved in your working directory.

[!TIP] Currently, only calls to OpenAI endpoints are supported. This tool is still in ongoing development and integrations with other service providers and locally hosted LLMs are planned.

Test Evaluator

The test evaluator command is used to evaluate the tests of your repository. It generates an evaluation report and provides various options for customization, such as specifying a checklist file, output format, and verbosity.

Example calls:

# Evaluate repo, and output the evalutions as a JSON file in working directory
$ fixml evaluate /path/to/your/repo

# Perform the above verbosely, and use the JSON file to export a HTML report
$ fixml evaluate /path/to/your/repo -e ./eval_report.html -v

# Perform the above, but use a custom checklist, and to overwrite existing report
$ fixml evaluate /path/to/your/repo -e ./eval_report.html -v -o -c checklist/checklist.csv

# Perform the above, and to use gpt-4o as the evaluation model
$ fixml evaluate /path/to/your/repo -e ./eval_report.html -v -o -c checklist/checklist.csv -m gpt-4o

Test Spec Generator

The test spec generator command is used to generate a test specification from a checklist. It allows for the inclusion of an optional checklist file to guide the test specification generation process.

Example calls:

# Generate test function specifications and to write them into a .py file
$ fixml generate test.py

# Perform the above, but to use a custom checklist
$ fixml generate test.py -c checklist/checklist.csv

Package

Alternatively, you can use the package to import all components necessary for running the evaluation/generation workflows listed above.

The workflows used in the package have been designed to be fully modular. You can easily switch between different prompts, models and checklists to use. You can also write your own custom classes to extend the capability of this library.

Consult the API documentation on Readthedocs for more information and example calls.

Development Build

If you are interested in helping the development of this tool, or you would like to get the cutting-edge version of this tool, you can install this tool via conda.

To do this, ensure you have Miniconda/Anaconda installed on your system. You can download miniconda on their official website.

1. Clone this repository from GitHub:

$ git clone [e-mail]:UBC-MDS/fixml.git

2. Create a conda environment:

$ conda env create -f environment.yaml

3. Activate the newly created conda environment (default name fixml):

$ conda activate fixml

4. Use poetry which is preinstalled in the conda environment to create a local package install:

$ poetry install

5. You now should be able to run fixml, try:

fixml --help

Rendering API Documentation

Make sure you have installed dev dependencies listed in pyproject.toml.

cd docs/

python -m sphinx -T -b html  -D language=en . _build

Running the Tests

Navigate to the project root directory and use the following command in terminal to run the test suite:

# skip integration tests
$ pytest -m "not integeration"

# run ALL tests, which requires OPENAI_API_KEY to be set
$ pytest

Contributing

Interested in contributing? Check out the contributing guidelines. Please note that this project is released with a Code of Conduct. By contributing to this project, you agree to abide by its terms.

License

fixml was created by John Shiu, Orix Au Yeung, Tony Shum, and Yingzi Jin as a deliverable product during our capstone project of the UBC-MDS program in collaboration with Dr. Tiffany Timbers and Dr. Simon Goring. It is licensed under the terms of the MIT license for software code. Reports and instructional materials are licensed under the terms of the CC-BY 4.0 license.

Citation

If you use fixml in your work, please cite:

@misc{mds2024fixml,
  author =       {John Shiu, Orix Au Yeung, Tony Shum, and Yingzi Jin},
  title =        {fixml: A Comprehensive Tool for Test Evaluation and Specification Generation},
  howpublished = {\url{https://https://github.com/UBC-MDS/fixml}},
  year =         {2024}
}

Acknowledgements

We'd like to thank everyone who has contributed to the development of the fixml package. This is a new project aimed at enhancing the robustness and reproducibility of applied machine learning software. It is meant to be a research tool and is currently hosting on GitHub as an open source project. We welcome it to be read, revised, and supported by data scientists, machine learning engineers, educators, practitioners, and hobbyists alike. Your contributions and feedback are invaluable in making this package a reliable resource for the community.

Special thanks to the University of British Columbia (UBC) and the University of Wisconsin-Madison for their support and resources. We extend our gratitude to Dr. Tiffany Timbers and Dr. Simon Goringfor their guidance and expertise, which have been instrumental in the development of this project.