Ecosyste.ms: Awesome
An open API service indexing awesome lists of open source software.
https://github.com/PaddlePaddle/InterpretDL
InterpretDL: Interpretation of Deep Learning Models，基于『飞桨』的模型可解释性算法库。
https://github.com/PaddlePaddle/InterpretDL
convolutional-neural-networks explanations grad-cam interpretation-algorithms lime model-interpretation nlp-models paddlepaddle smoothgrad vision-transformer visualizations
Last synced: about 1 month ago
JSON representation
InterpretDL: Interpretation of Deep Learning Models，基于『飞桨』的模型可解释性算法库。
- Host: GitHub
- URL: https://github.com/PaddlePaddle/InterpretDL
- Owner: PaddlePaddle
- License: apache-2.0
- Created: 2020-06-22T01:26:45.000Z (about 4 years ago)
- Default Branch: master
- Last Pushed: 2023-10-14T09:33:07.000Z (11 months ago)
- Last Synced: 2024-05-14T12:02:10.470Z (4 months ago)
- Topics: convolutional-neural-networks, explanations, grad-cam, interpretation-algorithms, lime, model-interpretation, nlp-models, paddlepaddle, smoothgrad, vision-transformer, visualizations
- Language: Python
- Homepage: https://interpretdl.readthedocs.io
- Size: 31.9 MB
- Stars: 231
- Watchers: 12
- Forks: 36
- Open Issues: 4
-
Metadata Files:
- Readme: README.md
- License: LICENSE
Awesome Lists containing this project
- Awesome-explainable-AI - https://github.com/PaddlePaddle/InterpretDL
README
[**中文**](./README_CN.md) | English
[](https://github.com/PaddlePaddle/InterpretDL/releases)
[](https://pypi.org/project/interpretdl)
[](https://circleci.com/gh/PaddlePaddle/InterpretDL)
[](https://interpretdl.readthedocs.io/en/latest/index.html)
[](https://pepy.tech/project/interpretdl)
# InterpretDL: Interpretation of Deep Learning Models based on PaddlePaddle
InterpretDL, short for *interpretations of deep learning models*, is a model interpretation toolkit for [PaddlePaddle](https://github.com/PaddlePaddle/Paddle) models. This toolkit contains implementations of many interpretation algorithms, including LIME, Grad-CAM, Integrated Gradients and more. Some SOTA and new interpretation algorithms are also implemented.
*InterpretDL is under active construction and all contributions are welcome!*
# Why InterpretDL
The increasingly complicated deep learning models make it impossible for people to understand their internal workings. Interpretability of black-box models has become the research focus of many talented researchers. InterpretDL provides a collection of both classical and new algorithms for interpreting models.
By utilizing these helpful methods, people can better understand why models work and why they don't, thus contributing to the model development process.
For researchers working on designing new interpretation algorithms, InterpretDL gives an easy access to existing methods that they can compare their work with.
# :fire: :fire: :fire: News :fire: :fire: :fire:
- (2023.2) One paper on explaining Transformers got accepted by TMLR. See implementations at [BT](https://github.com/PaddlePaddle/InterpretDL/blob/master/interpretdl/interpreter/bidirectional_transformer.py).
> Jiamin Chen, Xuhong Li, Lei Yu, Dejing Dou, Haoyi Xiong. “Beyond Intuition: Rethinking Token Attributions inside Transformers.” TMLR. [paper link](https://openreview.net/forum?id=rm0zIzlhcX).
- (2022.11) Two papers got accepted by AAAI'23 and Artificial Intelligence respectively. See implementations at [G-LIME](https://github.com/PaddlePaddle/InterpretDL/blob/master/interpretdl/interpreter/glime.py) and [TrainingDynamics](https://github.com/PaddlePaddle/InterpretDL/blob/master/interpretdl/interpreter/training_dynamics.py).
> Xuhong Li, Haoyi Xiong, Xingjian Li, Xiao Zhang, Ji Liu, Haiyan Jiang, Zeyu Chen, Dejing Dou. “G-LIME: Statistical Learning for Local Interpretations of Deep Neural Networks using Global Priors.” Artificial Intelligence, 2023. [pdf link](https://github.com/PaddlePaddle/InterpretDL/files/10110787/glime-aij-paper.pdf).
> Qingrui Jia, Xuhong Li, Lei Yu, Penghao Zhao, Jiang Bian, Shupeng Li, Haoyi Xiong, Dejing Dou. “Learning from Training Dynamics: Identifying Mislabeled Data Beyond Manually Designed Features”. AAAI 2023. [pdf link](https://arxiv.org/abs/2212.09321).
# Demo
Interpretation algorithms give a hint of why a black-box model makes its decision.
The following table gives visualizations of several interpretation algorithms applied to the original image to tell us why the model predicts "bull_mastiff."
| Original Image | IntGrad ([demo](https://github.com/PaddlePaddle/InterpretDL/blob/master/tutorials/example_int_grad_cv.ipynb)) | SG ([demo](https://github.com/PaddlePaddle/InterpretDL/blob/master/tutorials/example_smooth_grad_cv.ipynb)) | LIME ([demo](https://github.com/PaddlePaddle/InterpretDL/blob/master/tutorials/example_lime_cv.ipynb)) | Grad-CAM ([demo](https://github.com/PaddlePaddle/InterpretDL/blob/master/tutorials/example_grad_cam_cv.ipynb)) |
| :------------------: | :----------------------------------------------------------: | :----------------------------------------------------------: | :----------------------------------------------------------: | :----------------------------------------------------------: |
|  |  |  |  |  |
For sentiment analysis task, the reason why a model gives positive/negative predictions can be visualized as follows. A quick demo can be found [here](https://github.com/PaddlePaddle/InterpretDL/blob/master/tutorials/ernie-2.0-en-sst-2.ipynb). Samples in Chinese are also available [here](https://github.com/PaddlePaddle/InterpretDL/blob/master/tutorials/ernie-1.0-zh-chnsenticorp.ipynb).
# Contents
- [InterpretDL: Interpretation of Deep Learning Models based on PaddlePaddle](#interpretdl-interpretation-of-deep-learning-models-based-on-paddlepaddle)
- [Why InterpretDL](#why-interpretdl)
- [:fire: :fire: :fire: News :fire: :fire: :fire:](#fire-fire-fire-news-fire-fire-fire)
- [Demo](#demo)
- [Contents](#contents)
- [Installation](#installation)
- [Pip installation](#pip-installation)
- [Developer installation](#developer-installation)
- [Unit Tests](#unit-tests)
- [Documentation](#documentation)
- [Getting Started](#getting-started)
- [Examples and Tutorials](#examples-and-tutorials)
- [Roadmap](#roadmap)
- [Implemented Algorithms with Taxonomy](#implemented-algorithms-with-taxonomy)
- [Implemented Trustworthiness Evaluation Algorithms](#implemented-trustworthiness-evaluation-algorithms)
- [Presentations](#presentations)
- [References of Algorithms](#references-of-algorithms)
- [Copyright and License](#copyright-and-license)
- [Recent News](#recent-news)
# Installation
It requires the deep learning framework [paddlepaddle](https://www.paddlepaddle.org.cn/install/quick), versions with CUDA support are recommended.
## Pip installation
```bash
pip install interpretdl
# or with tsinghua mirror
pip install interpretdl -i https://pypi.tuna.tsinghua.edu.cn/simple
```
## Developer installation
```bash
git clone https://github.com/PaddlePaddle/InterpretDL.git
# ... fix bugs or add new features
cd InterpretDL && pip install -e .
# welcome to propose pull request and contribute
yapf -i # code style: column_limit=120
```
## Unit Tests
```bash
# run gradcam unit tests
python -m unittest -v tests.interpreter.test_gradcam
# run all unit tests
python -m unittest -v
```
# Documentation
Online link: [interpretdl.readthedocs.io](https://interpretdl.readthedocs.io/en/latest/index.html).
Or generate the docs locally:
```bash
git clone https://github.com/PaddlePaddle/InterpretDL.git
cd docs
make html
open _build/html/index.html
```
# Getting Started
All interpreters inherit the abstract class [`Interpreter`](https://github.com/PaddlePaddle/InterpretDL/blob/4f7444160981e99478c26e2a52f8e40bd06bf644/interpretdl/interpreter/abc_interpreter.py), of which `interpret(**kwargs)` is the function to call.
```python
# an example of SmoothGradient Interpreter.
import interpretdl as it
from paddle.vision.models import resnet50
paddle_model = resnet50(pretrained=True)
sg = it.SmoothGradInterpreter(paddle_model, use_cuda=True)
gradients = sg.interpret("test.jpg", visual=True, save_path=None)
```
A quick [Getting-Started tutorial](https://github.com/PaddlePaddle/InterpretDL/blob/master/tutorials/Getting_Started.ipynb) (or [on NBviewer](https://nbviewer.org/github/PaddlePaddle/InterpretDL/blob/master/tutorials/Getting_Started.ipynb)) is provided. It takes only a few minutes to be familiar with InterpretDL.
# Examples and Tutorials
We have provided at least one example for each interpretation algorithm and each trustworthiness evaluation algorithm, hopefully covering applications for both CV and NLP.
We are currently preparing tutorials for easy usages of InterpretDL.
Both examples and tutorials can be accessed under [tutorials](https://github.com/PaddlePaddle/InterpretDL/tree/master/tutorials) folder.
# Roadmap
We are planning to create a useful toolkit for offering the model interpretations as well as evaluations.
We have now implemented the interpretation algorithms as follows, and we are planning to add more algorithms that are desired.
Welcome to contribute or just tell us which algorithms are desired.
## Implemented Algorithms with Taxonomy
Two dimensions (representations of explanation results and types of the target model) are used to categorize the interpretation algorithms. This taxonomy can be an indicator to find the best suitable algorithm for the target task and model.
| Methods | Representation | Model Type |
| ------------------------------------------------------------ | --------------------- | ---------------------- |
| [LIME](https://github.com/PaddlePaddle/InterpretDL/blob/master/interpretdl/interpreter/lime.py) | Input Features | Model-Agnostic |
| [LIME with Prior](https://github.com/PaddlePaddle/InterpretDL/blob/master/interpretdl/interpreter/lime_prior.py) | Input Features | Model-Agnostic |
| [GLIME](https://github.com/PaddlePaddle/InterpretDL/blob/master/interpretdl/interpreter/glime.py) | Input Features | Model-Agnostic |
| [NormLIME/FastNormLIME](https://github.com/PaddlePaddle/InterpretDL/blob/master/interpretdl/interpreter/_normlime_base.py) | Input Features | Model-Agnostic |
| [LRP](https://github.com/PaddlePaddle/InterpretDL/blob/master/interpretdl/interpreter/lrp.py) | Input Features | Differentiable* |
| [SmoothGrad](https://github.com/PaddlePaddle/InterpretDL/blob/master/interpretdl/interpreter/smooth_grad.py) | Input Features | Differentiable |
| [IntGrad](https://github.com/PaddlePaddle/InterpretDL/blob/master/interpretdl/interpreter/integrated_gradients.py) | Input Features | Differentiable |
| [GradSHAP](https://github.com/PaddlePaddle/InterpretDL/blob/master/interpretdl/interpreter/gradient_shap.py) | Input Features | Differentiable |
| [Occlusion](https://github.com/PaddlePaddle/InterpretDL/blob/master/interpretdl/interpreter/occlusion.py) | Input Features | Model-Agnostic |
| [GradCAM/CAM](https://github.com/PaddlePaddle/InterpretDL/blob/master/interpretdl/interpreter/gradient_cam.py) | Intermediate Features | Specific: CNNs |
| [ScoreCAM](https://github.com/PaddlePaddle/InterpretDL/blob/master/interpretdl/interpreter/score_cam.py) | Intermediate Features | Specific: CNNs |
| [Rollout](https://github.com/PaddlePaddle/InterpretDL/blob/master/interpretdl/interpreter/rollout.py) | Intermediate Features | Specific: Transformers |
| [TAM](https://github.com/PaddlePaddle/InterpretDL/blob/master/interpretdl/interpreter/transition_attention_maps.py) | Input Features | Specific: Transformers |
| [Generic Attention](https://github.com/PaddlePaddle/InterpretDL/blob/master/interpretdl/interpreter/generic_attention.py) | Input Features | Specific: Transformers |
| [Bidirectional](https://github.com/PaddlePaddle/InterpretDL/blob/master/interpretdl/interpreter/bidirectional_transformer.py) | Input Features | Specific: Transformers |
| [ForgettingEvents](https://github.com/PaddlePaddle/InterpretDL/blob/master/interpretdl/interpreter/forgetting_events.py) | Dataset-Level | Differentiable |
| [TIDY (Training Data Analyzer)](https://github.com/PaddlePaddle/InterpretDL/blob/master/tutorials/TIDY.ipynb) | Dataset-Level | Differentiable |
| [BHDF](https://github.com/PaddlePaddle/InterpretDL/blob/master/interpretdl/interpreter/training_dynamics.py) | Dataset-Level** | Differentiable |
| [Consensus](https://github.com/PaddlePaddle/InterpretDL/blob/master/interpretdl/interpreter/consensus.py) | Features | Cross-Model |
\* LRP requires that the model is of specific implementations for relevance back-propagation.
\*\* Dataset-Level Interpreters require a training process.
## Implemented Trustworthiness Evaluation Algorithms
- [x] [Perturbation Tests](https://github.com/PaddlePaddle/InterpretDL/blob/master/interpretdl/evaluate_interpreter/perturbation.py)
- [x] [Deletion & Insertion](https://github.com/PaddlePaddle/InterpretDL/blob/master/interpretdl/evaluate_interpreter/deletion_insertion.py)
- [x] [Infidelity](https://github.com/PaddlePaddle/InterpretDL/blob/master/interpretdl/evaluate_interpreter/infidelity.py)
## Interpretability Evaluation Algorithms
- [x] [Localization Ability](https://github.com/PaddlePaddle/InterpretDL/blob/master/interpretdl/evaluate_interpreter/localization.py)
## Planning Alorithms
* Intermediate Features Interpretation Algorithm
- [x] More Transformers Specific Interpreters
* Dataset-Level Interpretation Algorithms
- [ ] Influence Function
* Evaluations
- [ ] Local Fidelity
- [ ] Monotonicity
- [x] Infidelity
- [ ] Sensitivity
# Presentations
**Linux Foundation Project AI & Data** -- Interpretable Deep Learning: Interpretation, Interpretability, Trustworthiness, and Beyond. [Video Link](https://wiki.lfaidata.foundation/download/attachments/7733341/GMT20220324-130226_Recording_3840x2160.mp4?version=1&modificationDate=1649079184753&api=v2) (00:20:30 -- 00:45:00).
**Baidu Create 2021 (in Chinese)**: [Video Link](https://live.baidu.com/m/media/pclive/pchome/live.html?room_id=5073321791&source=h5pre) (01:18:40 -- 01:36:30).
**ICML 2021 Expo** -- Interpretable Deep Learning: Interpretation, Interpretability, Trustworthiness, and Beyond. [Video Link](https://icml.cc/ExpoConferences/2021/workshop/11429#wse-detail-11435).
# References of Algorithms
* `SGDNoise`: [On the Noisy Gradient Descent that Generalizes as SGD, Wu et al 2019](https://arxiv.org/abs/1906.07405)
* `IntegratedGraients`: [Axiomatic Attribution for Deep Networks, Mukund Sundararajan et al. 2017](https://arxiv.org/abs/1703.01365)
* `CAM`, `GradCAM`: [Grad-CAM: Visual Explanations from Deep Networks via Gradient-based Localization, Ramprasaath R. Selvaraju et al. 2017](https://arxiv.org/abs/1610.02391.pdf)
* `SmoothGrad`: [SmoothGrad: removing noise by adding noise, Daniel Smilkov et al. 2017](https://arxiv.org/abs/1706.03825)
* `GradientShap`: [A Unified Approach to Interpreting Model Predictions, Scott M. Lundberg et al. 2017](http://papers.nips.cc/paper/7062-a-unified-approach-to-interpreting-model-predictions)
* `Occlusion`: [Visualizing and Understanding Convolutional Networks, Matthew D Zeiler and Rob Fergus 2013](https://arxiv.org/abs/1311.2901)
* `Lime`: ["Why Should I Trust You?": Explaining the Predictions of Any Classifier, Marco Tulio Ribeiro et al. 2016](https://arxiv.org/abs/1602.04938)
* `NormLime`: [NormLime: A New Feature Importance Metric for Explaining Deep Neural Networks, Isaac Ahern et al. 2019](https://arxiv.org/abs/1909.04200)
* `ScoreCAM`: [Score-CAM: Score-Weighted Visual Explanations for Convolutional Neural Networks, Haofan Wang et al. 2020](https://arxiv.org/abs/1910.01279)
* `ForgettingEvents`: [An Empirical Study of Example Forgetting during Deep Neural Network Learning, Mariya Toneva et al. 2019](http://arxiv.org/abs/1812.05159)
* `LRP`: [On Pixel-Wise Explanations for Non-Linear Classifier Decisions by Layer-Wise Relevance Propagation, Bach et al. 2015](https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0130140)
* `Rollout`: [Quantifying Attention Flow in Transformers, Abnar et al. 2020](https://arxiv.org/abs/2005.00928)
* `TAM`: [Explaining Information Flow Inside Vision Transformers Using Markov Chain. Yuan et al. 2021](https://openreview.net/forum?id=TT-cf6QSDaQ)
* `Consensus`: [Cross-Model Consensus of Explanations and Beyond for Image Classification Models: An Empirical Study. Li et al 2021](https://arxiv.org/abs/2109.00707)
* `Perturbation`: [Evaluating the visualization of what a deep neural network has learned.](https://arxiv.org/abs/1509.06321)
* `Deletion&Insertion`: [RISE: Randomized Input Sampling for Explanation of Black-box Models.](https://arxiv.org/abs/1806.07421)
* `PointGame`: [Top-down Neural Attention by Excitation Backprop.](https://arxiv.org/abs/1608.00507)
* `Generic Attention`: [Generic Attention-model Explainability for Interpreting Bi-Modal and Encoder-Decoder Transformers
](https://arxiv.org/abs/2103.15679)
# Copyright and License
InterpretDL is provided under the [Apache-2.0 license](https://github.com/PaddlePaddle/InterpretDL/blob/master/LICENSE).
# Recent News
- (2022.08) The paper with this repository is accepted by Journal of Machine Learning Research (JMLR). If this repo is helpful for your work, please consider citing this paper:
> Xuhong Li, Haoyi Xiong, Xingjian Li, Xuanyu Wu, Zeyu Chen, and Dejing Dou. “InterpretDL: Explaining Deep Models in PaddlePaddle.” Journal of Machine Learning Research, 2022. https://jmlr.org/papers/v23/21-0738.html.
```
@article{JMLR:v23:21-0738,
author = {Xuhong Li and Haoyi Xiong and Xingjian Li and Xuanyu Wu and Zeyu Chen and Dejing Dou},
title = {InterpretDL: Explaining Deep Models in PaddlePaddle},
journal = {Journal of Machine Learning Research},
year = {2022},
volume = {23},
number = {197},
pages = {1--6},
url = {http://jmlr.org/papers/v23/21-0738.html}
}
```
- (2022.08) Two research works got accepted by ECML and Machine Learning Journal. Cross-model consensus explanations are exploited to create a pseudo semantic segmentation dataset named PSSL, which contains 1.2M pseudo masks corresponding to the whole ImageNet training set. The dataset is publicly available.
Refer to [PaddleSeg:PSSL](https://github.com/PaddlePaddle/PaddleSeg/tree/develop/configs/pssl) for downloading the dataset and the pretrained models.
Refer to the papers for the details about the Consensus explanation and how it helps for the semantic segmentation task:
> Xuhong Li, Haoyi Xiong, Siyu Huang, Shilei Ji, Dejing Dou. Cross-Model Consensus of Explanations and Beyond for Image Classification Models: An Empirical Study. ECML'22, Machine Learning Journal Track. https://arxiv.org/abs/2109.00707.
> Xuhong Li, Haoyi Xiong, Yi Liu, Dingfu Zhou, Zeyu Chen, Yaqing Wang, and Dejing Dou. "Distilling ensemble of explanations for weakly-supervised pre-training of image segmentation models." Machine Learning (2022): 1-17. https://arxiv.org/abs/2207.03335.
- (2022/04/27) A getting-started tutorial is provided. Check it from [GitHub](https://github.com/PaddlePaddle/InterpretDL/blob/master/tutorials/Getting_Started.ipynb) or [NBViewer](https://nbviewer.org/github/PaddlePaddle/InterpretDL/blob/master/tutorials/Getting_Started.ipynb). Usage examples have been provided for each algorithm (both Interpreter and Evaluator). We are currently preparing tutorials for easy usages of InterpretDL. Both tutorials and examples can be assessed under the [tutorial](https://github.com/PaddlePaddle/InterpretDL/tree/master/tutorials) folder.
- (2022/01/06) Implemented the Cross-Model Consensus Explanation method. In brief, this method averages the explanation results from several models. Instead of interpreting individual models, this method is able to identify the discriminative features in the input data with accurate localization. See the [paper](https://arxiv.org/abs/2109.00707) for details.
* `Consensus`: Xuhong Li, Haoyi Xiong, Siyu Huang, Shilei Ji, Dejing Dou. Cross-Model Consensus of Explanations and Beyond for Image Classification Models: An Empirical Study. arXiv:2109.00707.
We show a demo with six models (the last column shows the consensus explanation), while more models (around 15) could give a much better result. See the [example](https://github.com/PaddlePaddle/InterpretDL/blob/master/tutorials/example_consensus_cv.ipynb) for more details.
- (2021/10/20) Implemented the Transition Attention Maps (TAM) explanation method for PaddlePaddle [Vision Transformers](https://github.com/PaddlePaddle/PaddleClas/blob/release/2.3/ppcls/arch/backbone/model_zoo/vision_transformer.py). As always, several lines call this interpreter. See details from the [example](https://github.com/PaddlePaddle/InterpretDL/blob/master/tutorials/example_tam_cv_ViT.ipynb), and the [paper](https://openreview.net/forum?id=TT-cf6QSDaQ):
* `TAM`: Tingyi Yuan, Xuhong Li, Haoyi Xiong, Hui Cao, Dejing Dou. Explaining Information Flow Inside Vision Transformers Using Markov Chain. In *Neurips 2021 XAI4Debugging Workshop*.
```python
import paddle
import interpretdl as it
# load vit model and weights
# !wget -c https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ViT_base_patch16_224_pretrained.pdparams -P assets/
from assets.vision_transformer import ViT_base_patch16_224
paddle_model = ViT_base_patch16_224()
MODEL_PATH = 'assets/ViT_base_patch16_224_pretrained.pdparams'
paddle_model.set_dict(paddle.load(MODEL_PATH))
# Call the interpreter.
tam = it.TAMInterpreter(paddle_model, use_cuda=True)
img_path = 'samples/el1.png'
heatmap = tam.interpret(
img_path,
start_layer=4,
label=None, # elephant
visual=True,
save_path=None)
heatmap = tam.interpret(
img_path,
start_layer=4,
label=340, # zebra
visual=True,
save_path=None)
```
| image | elephant | zebra |
| :----------------------------------------------------------: | :----------------------------------------------------------: | :----------------------------------------------------------: |
|  |  |  |
- (2021/07/22) Implemented Rollout Explanations for PaddlePaddle [Vision Transformers](https://github.com/PaddlePaddle/PaddleClas/blob/release/2.3/ppcls/arch/backbone/model_zoo/vision_transformer.py). See the [notebook](https://github.com/PaddlePaddle/InterpretDL/blob/master/tutorials/example_rollout_cv_ViT.ipynb) for the visualization.
```python
import paddle
import interpretdl as it
# wget -c https://paddle-imagenet-models-name.bj.bcebos.com/dygraph/ViT_small_patch16_224_pretrained.pdparams -P assets/
from assets.vision_transformer import ViT_small_patch16_224
paddle_model = ViT_small_patch16_224()
MODEL_PATH = 'assets/ViT_small_patch16_224_pretrained.pdparams'
paddle_model.set_dict(paddle.load(MODEL_PATH))
img_path = 'assets/catdog.png'
rollout = it.RolloutInterpreter(paddle_model, use_cuda=True)
heatmap = rollout.interpret(img_path, start_layer=0, visual=True)
```