hierophant

This Python library use the power of large language models to provide intuitive, human-readable explanations for the predictions made by machine learning models.

Features

• Support for scikit-learn and PyTorch models
• Integration with OpenAI's language models for explanation generation
• Works with both classification and regression models

from sklearn.ensemble import RandomForestClassifier
from sklearn.datasets import load_iris

from hierophant.explainer import Explainer


iris = load_iris()
X, y = iris.data, iris.target

random_forest = RandomForestClassifier()
random_forest.fit(X, y)

# Create an explainer
explainer = Explainer(random_forest, language_model="openai/gpt-4")

explanations = explainer.explain(X)

For more detailed usage, see our Jupyter notebooks in the examples/ directory.

Contributing

We welcome contributions! See our contribution guide for more details.

License

This project is licensed under the terms of the MIT license.

TODO

• Support for more model types: Currently, hierophant supports scikit-learn models. In the future, we aim to add support for other types of models, such as PyTorch and TensorFlow models.
• Support for Hugging Face models: In addition to scikit-learn models, we aim to add support for Hugging Face models. This would allow hierophant to generate explanations for a wide range of state-of-the-art natural language processing models.
• Improved explanation generation: The current explanation generation process is quite basic. We need to further refine this process to generate more detailed and useful explanations.
• Model inspection capabilities: For more complex models, we might need to add functionality to inspect the internal workings of the model. This could involve using model interpretation techniques like LIME or SHAP.
• Data preprocessing functionality: We may need to add functionality to preprocess the data before feeding it to the model or the explanation generation system.
• Postprocessing of explanations: After generating the explanations, we may want to add postprocessing steps to make the explanations more readable or understandable. This could include summarization, highlighting, or conversion to other formats.
• Testing: We need to add comprehensive testing to ensure the reliability and robustness of hierophant.
• Documentation: While we have made a start on documentation, we need to continue to expand and improve it.
• Examples and tutorials: We should create more example notebooks and tutorials demonstrating how to use hierophant with different types of data and models.
• Community engagement: As an open-source project, we want to encourage community involvement. We need to continue improving our contribution guidelines and fostering an inclusive and welcoming community.