L2X - Code for replicating the experiments in the paper Learning to Explain: An Information-Theoretic Perspective on Model Interpretation.

Jianbo Chen

Last update: Sep 6, 2022

Overview

L2X

Code for replicating the experiments in the paper Learning to Explain: An Information-Theoretic Perspective on Model Interpretation at ICML 2018, by Jianbo Chen, Mitchell Stern, Martin J. Wainwright, Michael I. Jordan.

Dependencies

The code for L2X runs with Python and requires Tensorflow of version 1.2.1 or higher and Keras of version 2.0 or higher. Please pip install the following packages:

numpy
tensorflow
keras
pandas
nltk

Or you may run the following and in shell to install the required packages:

git clone https://github.com/Jianbo-Lab/L2X
cd L2X
sudo pip install -r requirements.txt

See README.md in respective folders for details.

Citation

If you use this code for your research, please cite our paper:

@arxiv{chen2018learning,
title = {Learning to Explain: An Information-Theoretic Perspective on Model Interpretation},
author = {Chen, Jianbo and Song, Le and Wainwright, Martin J and Jordan, Michael I}, 
journal={arXiv preprint arXiv:1802.07814}, 
year = {2018}  
}

You might also like...

Net2Vis automatically generates abstract visualizations for convolutional neural networks from Keras code.

Automatic neural network visualizations generated in your browser!

99 Nov 30, 2022

Code for visualizing the loss landscape of neural nets

Visualizing the Loss Landscape of Neural Nets This repository contains the PyTorch code for the paper Hao Li, Zheng Xu, Gavin Taylor, Christoph Studer

2.2k Dec 30, 2022

A data-driven approach to quantify the value of classifiers in a machine learning ensemble.

Documentation | External Resources | Research Paper Shapley is a Python library for evaluating binary classifiers in a machine learning ensemble. The

187 Dec 27, 2022

Algorithms for monitoring and explaining machine learning models

Alibi is an open source Python library aimed at machine learning model inspection and interpretation. The focus of the library is to provide high-qual

1.9k Dec 30, 2022

A library for debugging/inspecting machine learning classifiers and explaining their predictions

ELI5 ELI5 is a Python package which helps to debug machine learning classifiers and explain their predictions. It provides support for the following m

2.6k Dec 30, 2022

Lime: Explaining the predictions of any machine learning classifier

lime This project is about explaining what machine learning classifiers (or models) are doing. At the moment, we support explaining individual predict

10.3k Jan 1, 2023

FairML - is a python toolbox auditing the machine learning models for bias.

======== FairML: Auditing Black-Box Predictive Models FairML is a python toolbox auditing the machine learning models for bias. Description Predictive

338 Nov 9, 2022

A library that implements fairness-aware machine learning algorithms

Themis ML themis-ml is a Python library built on top of pandas and sklearnthat implements fairness-aware machine learning algorithms. Fairness-aware M

105 Dec 30, 2022

Interpretability and explainability of data and machine learning models

AI Explainability 360 (v0.2.1) The AI Explainability 360 toolkit is an open-source library that supports interpretability and explainability of datase

1.2k Dec 29, 2022

Comments

variational lower bound

It seems that this question is not relevant to the codes, but may I ask why the lower bound of the conditional expectation of log Pm(Y | x) is the conditional expectation of log Qs(Y | x)? I think applying the Jensen’s inequality yields the lower bound to a function of Pm. I can't find any relationship between Pm and Qs.

opened by yalunar 2
L2X for hierarchical LSTM returns always first sentence as selection

Hello and congratulations for your great work. I tried to implement the L2X explanations with hierarchical LSTM on a fake news dataset, as shown in the imdb_sent example. However, the explainer always selects the first sentence as an explanation. Do you have any idea why is this happening? I would also like to ask you (sorry if it is a stupid question), as the method is model agnostic, why do you use different neural network architectures for each imdb example?

opened by ikonstas-ds 1
Add note about l2x_synthetic package

Hi! I extracted the synthetic dataset generation code and bundled it as an easy-to-install package, called l2x_synthetic.

If you like, we could add a small link to the package in the README file; it can perhaps come useful to some people. Cheers! ☀️

opened by dunnkers 0
Post-hoc accuracy for IMDB-word

When running the experiments IMDB-word I tend to get significantly lower post-hoc accuracy than the one listed in Table 4 in the paper (0.908). Could you tell me if I might have to change some hyperparameters etc.?

Thanks!

opened by mniepert 0