PyXAI (Python eXplainable AI) is a Python library (version 3.6 or later) allowing to bring formal explanations suited to (regression or classification) tree-based ML models (Decision Trees, Random Forests, Boosted Trees, ...). PyXAI generates explanations that are post-hoc and local. In contrast to many popular approaches to XAI (SHAP, LIME, ...), PyXAI generates explanations that are also correct. Being correct (aka sound or faithful) indicates that the explanations that are provided actually reflect the exact behaviour of the model by guaranteeing certain properties about the explanations generated. They can be of several types:
PyXAI also includes algorithms for correcting tree-based models when their predictions conflict with pieces of user knowledge. This more tricky facet of XAI is seldom offered by existing XAI systems. When some domain knowledge is available and a prediction (or an explanation) contradicts it, the model must be corrected. Rectification is a principled approach for such a correction operation.
New features in version 1.1:
- Rectification for DT (Decision Tree) and RF (Random Forest) models dedicated to binary classification.
- Visualization displayed in a notebook or on screen, and now also for time series problems.
- Enhanced compatibility with Mac OS and Windows
New features in version 1.0:
- Regression for Boosted Trees with XGBoost or LightGBM
- Adding Theories (knowledge about the dataset)
- Easier model import (automatic detection of model types)
- PyXAI's Graphical User Interface (GUI): displaying, loading and saving explanations.
- Supports multiple image formats for imaging datasets
- Supports data pre-processing (tool for preparing and cleansing a dataset)
- Unit Tests with the unittest module
User interaction with PyXAI.
PyXAI's Graphical User Interface (GUI) for visualizing explanations.
Visualization in a notebook of an explanation for an instance from a time series problem.
The most popular approaches (SHAP, LIME, ...) to XAI are model-agnostic, but they do not offer any guarantees of rigor. A number of works by Marques-Silva and Huang, Ignatiev have highlighted several misconceptions about such approaches to XAI. Correctness is paramount when dealing with high-risk or sensitive applications, which is the type of applications that are targeted by PyXAI. When the correctness property is not satisfied, one can find ”counterexamples” for the explanations that are generated, i.e., pairs of instances sharing an explanation but leading to distinct predictions. Contrastingly, PyXAI algorithms rely on logic-based, model-precise approaches for computing explanations. Although formal explainability has a number of drawbacks, particularly in terms of the computational complexity of logical reasoning needed to derive explanations, steady progress has been made since its inception.
Models are the resulting objects of an experimental ML protocol through a chosen cross-validation method (for example, the result of a training phase on a classifier). Importantly, in PyXAI, there is a complete separation between the learning phase and the explaining phase: you produce/load/save models, and you find explanations for some instances given such models. Currently, with PyXAI, you can use methods to find explanations suited to different ML models for classification or regression tasks:
In addition to finding explanations, PyXAI also provides methods that perform operations (production, saving, loading) on models and instances. Currently, these methods are available for three ML libraries:
It is possible to also leverage PyXAI to find explanations suited to models learned using other libraries.
In this website, you can find all what you need to know about PyXAI, with more than 10 Jupyter Notebooks, including:
Here is an example (it comes from the Quick Start page):
from pyxai import Learning, Explainer
learner = Learning.Scikitlearn("tests/iris.csv", learner_type=Learning.CLASSIFICATION)
model = learner.evaluate(method=Learning.HOLD_OUT, output=Learning.DT)
instance, prediction = learner.get_instances(model, n=1, correct=True, predictions=[0])
explainer = Explainer.initialize(model, instance)
print("instance:", instance)
print("binary representation:", explainer.binary_representation)
sufficient_reason = explainer.sufficient_reason(n=1)
print("sufficient_reason:", sufficient_reason)
print("to_features:", explainer.to_features(sufficient_reason))
instance, prediction = learner.get_instances(model, n=1, correct=False)
explainer.set_instance(instance)
contrastive_reason = explainer.contrastive_reason()
print("contrastive reason", contrastive_reason)
print("to_features:", explainer.to_features(contrastive_reason, contrastive=True))
explainer.visualisation.screen(instance, contrastive_reason, contrastive=True)
As illustrated by this example, with a few lines of code, PyXAI allows you to train a model, extract instances, and get explanations about the classifications made.
FAQs
Explaining Machine Learning Classifiers in Python
We found that pyxai demonstrated a healthy version release cadence and project activity because the last version was released less than a year ago. It has 1 open source maintainer collaborating on the project.
Did you know?
Socket for GitHub automatically highlights issues in each pull request and monitors the health of all your open source dependencies. Discover the contents of your packages and block harmful activity before you install or update your dependencies.
Security News
PyPI confirms no security flaws were exploited in the Ultralytics supply chain attack and highlights improvements for safer package publishing.
Security News
Research
The Socket Research Team breaks down a malicious wrapper package that uses obfuscation to harvest credentials and exfiltrate sensitive data.
Research
Security News
Attackers used a malicious npm package typosquatting a popular ESLint plugin to steal sensitive data, execute commands, and exploit developer systems.