Lime

This project is an agnostic model interpretability framework and explainability tool designed to provide local interpretable explanations for individual predictions. It functions as a local surrogate model that approximates the behavior of any machine learning classifier or regression model to identify the most influential features for a specific instance.

The framework is designed to be model-agnostic, meaning it can explain predictions across tabular, text, and image data regardless of the underlying architecture. It employs local linear approximations and feature importance visualization to render contributions as bar plots, highlighted text, or segmented image regions.

Capabilities cover several data domains, including text classification analysis, tabular feature influence, and regression model diagnostics. For image data, the tool utilizes pixel contribution analysis and segment visualization to highlight areas driving a classification decision.

The library includes utilities for data preprocessing, such as continuous feature binning, categorical feature management, and mixed feature processing.

Features

Model Interpretability Frameworks - Provides a comprehensive framework for generating local interpretable explanations for predictions across diverse data types.

Model Interpretability Tools - Uses local surrogate models to approximate complex model behavior and explain individual predictions.

Interpretability Visualizations - Visualizes specific image segments that drive a machine learning model's classification decision.

Feature Contribution Analysis - Identifies the most influential pixels in an image by perturbing inputs and observing the change in model predictions.

Local Interpretability Tools - Generates local interpretable approximations to explain why a specific instance was assigned a particular class.

Local Surrogate Models - Approximates complex model behavior locally using linear models to identify the most influential features.

Regression Diagnostics - Generates local explanations for continuous value predictions to understand the drivers of regression outputs.

Explanation - Identifies specific words or phrases within a document that positively or negatively influence a model's label.

Text Attribution - Identifies specific words in a document that contribute positively or negatively to a model prediction.

Local Surrogate Approximations - Uses local linear approximations to identify features influencing a single model prediction.

Model Agnostic Interpretability - Implements an architecture-independent approach to interpret any classifier via a generic prediction interface.

Model Explainability - Generates local interpretable explanations for individual predictions made by any machine learning classifier.

Perturbation-Based Sampling - Creates local datasets by randomly altering features of a single instance to observe changes in model predictions.

Text Classifiers - Identifies the influence of individual words on a classifier's prediction for single text instances.

Applied Machine Learning - Provides a framework for identifying the most influential features for individual machine learning predictions.

Classifier Interpretability - Visualizes the specific segments or pixels of an image that most strongly drive classification decisions.

Tabular Explanations - Generates local explanations for tabular data by perturbing features and training a surrogate linear model.

Tabular Model Explanation - Analyzes how numerical and categorical features in a dataset contribute to classifier or regression model predictions.

Feature Binning - Converts numerical features into quartiles or bins to improve the human interpretability of model explanations.

Feature Importance Attribution - Renders feature contributions as bar plots and highlighted image segments to visualize prediction drivers.

Distribution-Based Sampling - Generates perturbed categorical data by sampling values according to their original training set distribution.

Linear Regression - Fits a weighted linear regression model to perturbed samples to identify local feature influences.

Linear Regression Models - Highlights the most influential features contributing to a continuous value prediction in regression models.

Linear Model Interpretability - Provides local interpretable explanations for the output of regression models to clarify prediction drivers.

Regression Approximations - Generates interpretable approximations for regression models to explain predicted continuous values.

Cross-Modal Explanation - Generates local interpretable explanations for individual predictions across text, tabular, and image data.

Multiclass Interpretations - Generates interpretations for multiple target labels to compare how different features influence various class predictions.

Super-Pixel Segmentation - Groups pixels into contiguous regions to treat blocks of pixels as single features during model analysis.

Interactive Visualization Rendering - Displays interactive visual representations of feature contributions and prediction probabilities within notebook environments.

Numerical Binning - Provides utilities to convert continuous numerical variables into discrete bins to simplify feature influence explanations.

Feature Interaction Visualizations - Renders model explanations as bar plots, interactive pages, or highlighted text to visualize feature weights.

Explainable AI Libraries - Local interpretable model-agnostic explanations for any classifier.

Model Interpretability - Explains predictions of any black-box machine learning classifier.

Model Interpretation - Explaining predictions of any machine learning classifier.

Guardrails and AI Safety - Listed in the “Guardrails and AI Safety” section of the The Incredible Pytorch awesome list.

marcotcrlime

Features

Star history