/README.md

<!-- WEHUB_ZH_README -->
> [!NOTE]
> 本文档由 WeHub 基于上游 README 翻译整理，属于社区翻译，非官方中文文档。  
> [English](./README.en.md) · [原始项目](https://github.com/cleanlab/cleanlab) · [上游 README](https://github.com/cleanlab/cleanlab/blob/HEAD/README.md)  
> 原作者、版权与许可证归属以原始项目及本仓库 LICENSE 文件为准。

<div align="center">
  <img src="https://raw.githubusercontent.com/cleanlab/assets/master/cleanlab/cleanlab_logo_open_source.png" width=60%>
</div>

<div align="center">
<a href="https://pypi.org/pypi/cleanlab/" target="_blank"><img src="https://img.shields.io/pypi/v/cleanlab.svg" alt="pypi_versions"></a>
<a href="https://pypi.org/pypi/cleanlab/" target="_blank"><img src="https://img.shields.io/badge/python-3.10%2B-blue" alt="py_versions"></a>
<a href="https://app.codecov.io/gh/cleanlab/cleanlab" target="_blank"><img src="https://codecov.io/gh/cleanlab/cleanlab/branch/master/graph/badge.svg" alt="coverage"></a>
<a href="https://github.com/cleanlab/cleanlab/stargazers/" target="_blank"><img src="https://img.shields.io/github/stars/cleanlab/cleanlab?style=social&maxAge=2592000" alt="Github Stars"></a>
<a href="https://twitter.com/CleanlabAI" target="_blank"><img src="https://img.shields.io/twitter/follow/CleanlabAI?style=social" alt="Twitter"></a>
</div>

<h4 align="center">
    <p>
        <a href="https://docs.cleanlab.ai/">文档</a> |
        <a href="https://github.com/cleanlab/examples">示例</a> |
        <a href="https://cleanlab.ai/blog/learn/">博客</a> |
        <a href="#citation-and-related-publications">研究</a>
    <p>
</h4>

Cleanlab 的开源库可帮助你通过自动检测 ML 数据集中的问题来**清理**（clean）数据与**标注**（lab**el**）。为促进**在混乱的真实世界数据上进行机器学习**，这一以数据为中心的 AI（data-centric AI）包利用你*现有*的模型来估计数据集中的问题，修复这些问题后即可训练出*更优*的模型。
 

<p align="center">
  <img src="https://raw.githubusercontent.com/cleanlab/assets/master/cleanlab/datalab_issues.png" width=74%>
</p>
<p align="center">
    通过以下代码，cleanlab **自动检测**到的猫/狗（Cat/Dog）数据集中各类问题的示例：    
</p>

```python
        lab = cleanlab.Datalab(data=dataset, label="column_name_for_labels")
        # Fit any ML model, get its feature_embeddings & pred_probs for your data
        lab.find_issues(features=feature_embeddings, pred_probs=pred_probs)
        lab.report()
```

- 使用 cleanlab 自动检查各类数据集：[文本](https://docs.cleanlab.ai/stable/tutorials/datalab/text.html), [音频](https://docs.cleanlab.ai/stable/tutorials/datalab/audio.html), [图像](https://docs.cleanlab.ai/stable/tutorials/datalab/image.html), 或 [表格](https://docs.cleanlab.ai/stable/tutorials/datalab/tabular.html) 数据集。
- 使用 cleanlab 自动：[检测数据问题（异常值、重复项、标注错误等）](https://docs.cleanlab.ai/stable/tutorials/datalab/datalab_quickstart.html), [训练鲁棒模型](https://docs.cleanlab.ai/stable/tutorials/indepth_overview.html), [推断多标注者数据的共识与标注者质量](https://docs.cleanlab.ai/stable/tutorials/multiannotator.html), [建议下一步应（重新）标注的数据（主动学习）](https://github.com/cleanlab/examples/blob/master/active_learning_multiannotator/active_learning.ipynb).


---


## 运行 cleanlab 开源版

此 cleanlab 包运行于 Python 3.10+，支持 Linux、macOS 以及 Windows。

- 从[这里](https://docs.cleanlab.ai/)! 开始入门。可通过 `uv`、`pip` 或 `conda` 安装。
- 从源码安装最新开发版的开发者请参阅[此 master 分支文档](https://docs.cleanlab.ai/master/index.html).

**实践以数据为中心的 AI 可以是这样：**
1. 在原始数据集上训练初始 ML 模型。
2. 利用该模型诊断数据问题（通过 cleanlab 方法）并改进数据集。
3. 在改进后的数据集上训练同一模型。 
4. 尝试各种建模技术以进一步提升性能。

多数人直接从步骤 1 跳到 4，但借助 cleanlab，你无需对建模代码做任何修改也可能获得巨大提升！
通过反复执行步骤 2 → 4 持续提升性能（并尽量使用*清洗后*的数据进行评估）。

![](https://raw.githubusercontent.com/cleanlab/assets/master/cleanlab/flowchart.png)


## 将 cleanlab 用于任意模型及大多数 ML 任务

cleanlab 的全部功能均可配合**任意数据集**与**任意模型**使用。是的，任意模型：PyTorch、Tensorflow、Keras、JAX、HuggingFace、OpenAI、XGBoost、scikit-learn 等。

cleanlab 适用于种类繁多的机器学习任务。此以数据为中心的 AI 包为以下特定任务提供专门功能：
1. [二分类与多分类](https://docs.cleanlab.ai/stable/tutorials/indepth_overview.html)
2. [多标签分类](https://docs.cleanlab.ai/stable/tutorials/multilabel_classification.html)（例如图像/文档标注）
3. [Token 分类](https://docs.cleanlab.ai/stable/tutorials/token_classification.html)（例如文本中的实体识别）
4. [回归](https://docs.cleanlab.ai/stable/tutorials/regression.html)（预测数据集中的数值列）
5. [图像分割](https://docs.cleanlab.ai/stable/tutorials/segmentation.html)（带逐像素标注的图像）
6. [目标检测](https://docs.cleanlab.ai/stable/tutorials/object_detection.html)（带边界框标注的图像）
7. [由多名标注者标注数据的分类](https://docs.cleanlab.ai/stable/tutorials/multiannotator.html)
8. [多名标注者的主动学习](https://github.com/cleanlab/examples/blob/master/active_learning_multiannotator/active_learning.ipynb)（建议标注或重新标注哪些数据最能提升模型）
9. [异常值检测](https://docs.cleanlab.ai/stable/tutorials/outliers.html)（识别看似分布外的非典型数据）

对于其他 ML 任务，若恰当应用，cleanlab 仍可帮助你改进数据集。
请参阅我们的[示例 Notebook](https://github.com/cleanlab/examples) 与[博客](https://cleanlab.ai/blog/learn/).


## 如此清新，如此 cleanlab

除了自动捕获潜藏在你数据中的[各类问题](https://docs.cleanlab.ai/stable/cleanlab/datalab/guide/issue_type_description.html)，这一以数据为中心的 AI 包还可帮助你处理**噪声标注**，并训练更**鲁棒的 ML 模型**。
示例如下：

```python

# cleanlab works with **any classifier**. Yup, you can use PyTorch/TensorFlow/OpenAI/XGBoost/etc.
cl = cleanlab.classification.CleanLearning(sklearn.YourFavoriteClassifier())

# cleanlab finds data and label issues in **any dataset**... in ONE line of code!
label_issues = cl.find_label_issues(data, labels)

# cleanlab trains a robust version of your model that works more reliably with noisy data.
cl.fit(data, labels)

# cleanlab estimates the predictions you would have gotten if you had trained with *no* label issues.
cl.predict(test_data)

# A universal data-centric AI tool, cleanlab quantifies class-level issues and overall data quality, for any dataset.
cleanlab.dataset.health_summary(labels, confident_joint=cl.confident_joint)
```

cleanlab 通过最先进的*置信学习*（confident learning）算法来**清理**你数据的**标注**，相关算法发表于此[论文](https://jair.org/index.php/jair/article/view/12125) 与[博客](https://l7.curtisnorthcutt.com/confident-learning).。可在 [labelerrors.com](https://labelerrors.com). 查看部分经 cleanlab 清洗的数据集。

cleanlab 具备：

1. **有理论支撑**——即使模型不完美，也能对标注噪声进行精确估计，并具有[可证明的保证](https://arxiv.org/abs/1911.00068)。
2. **快速**——代码已并行化且可扩展。
4. **易于使用**——一行代码即可找出错误标注、劣质标注者、异常值，或训练抗噪声模型。
6. **通用**——适用于**[任意数据集](https://labelerrors.com/)****（文本、图像、表格、音频等）+ **任意模型**（PyTorch、OpenAI、XGBoost 等）
<br/>

![](https://raw.githubusercontent.com/cleanlab/assets/master/cleanlab/label-errors-examples.png)
<p align="center">
使用 cleanlab <a href="https://l7.curtisnorthcutt.com/label-errors">发现并纠正</a>的各类图像数据集中错误给定标注的示例。 
尽管这些示例来自图像数据集，该方法同样适用于文本、音频、表格数据。
</p>

## 引用与相关出版物

cleanlab 基于同行评审的研究。若你使用本软件包，可引用以下相关论文：

<details><summary><a href="https://arxiv.org/abs/1911.00068">Confident Learning (JAIR '21)</a> (<b>点击显示 bibtex</b>) </summary>

    @article{northcutt2021confidentlearning,
        title={Confident Learning: Estimating Uncertainty in Dataset Labels},
        author={Curtis G. Northcutt and Lu Jiang and Isaac L. Chuang},
        journal={Journal of Artificial Intelligence Research (JAIR)},
        volume={70},
        pages={1373--1411},
        year={2021}
    }

</details>

<details><summary><a href="https://arxiv.org/abs/1705.01936">Rank Pruning (UAI '17)</a> (<b>点击显示 bibtex</b>) </summary>

    @inproceedings{northcutt2017rankpruning,
        author={Northcutt, Curtis G. and Wu, Tailin and Chuang, Isaac L.},
        title={Learning with Confident Examples: Rank Pruning for Robust Classification with Noisy Labels},
        booktitle = {Proceedings of the Thirty-Third Conference on Uncertainty in Artificial Intelligence},
        series = {UAI'17},
        year = {2017},
        location = {Sydney, Australia},
        numpages = {10},
        url = {http://auai.org/uai2017/proceedings/papers/35.pdf},
        publisher = {AUAI Press},
    }

</details>

<details><summary><a href="https://jonasmueller.org/info/LabelQuality_icml.pdf"> Label Quality Scoring (ICML '22)</a> (<b>点击显示 bibtex</b>) </summary>

    @inproceedings{kuan2022labelquality,
        title={Model-agnostic label quality scoring to detect real-world label errors},
        author={Kuan, Johnson and Mueller, Jonas},
        booktitle={ICML DataPerf Workshop},
        year={2022}
    }

</details>

<details><summary><a href="https://arxiv.org/abs/2210.03920"> Label Errors in Token Classification / Entity Recognition (NeurIPS '22)</a> (<b>点击显示 bibtex</b>) </summary>

    @inproceedings{wang2022tokenerrors,
        title={Detecting label errors in token classification data},
        author={Wang, Wei-Chen and Mueller, Jonas},
        booktitle={NeurIPS Workshop on Interactive Learning for Natural Language Processing (InterNLP)},
        year={2022}
    }

</details>

<details><summary><a href="https://arxiv.org/abs/2211.13895"> Label Errors in Multi-Label Classification (ICLR '23)</a> (<b>点击显示 bibtex</b>) </summary>

    @inproceedings{thyagarajan2023multilabel,
        title={Identifying Incorrect Annotations in Multi-Label Classification Data},
        author={Thyagarajan, Aditya and Snorrason, Elías and Northcutt, Curtis and Mueller, Jonas},
        booktitle={ICLR Workshop on Trustworthy ML},
        year={2023}
    }

</details>

<details><summary><a href="https://arxiv.org/abs/2309.00832"> Label Errors in Object Detection (ICML '23)</a> (<b>点击显示 bibtex</b>) </summary>

    @inproceedings{tkachenko2023objectlab,
        title={ObjectLab: Automated Diagnosis of Mislabeled Images in Object Detection Data},
        author={Tkachenko, Ulyana and Thyagarajan, Aditya and Mueller, Jonas},
        booktitle={ICML Workshop on Data-centric Machine Learning Research},
        year={2023}
    }

</details>

<details><summary><a href="https://arxiv.org/abs/2307.05080"> Label Errors in Image Segmentation (ICML '23)</a> (<b>点击显示 bibtex</b>) </summary>

    @inproceedings{lad2023segmentation,
        title={Estimating label quality and errors in semantic segmentation data via any model},
        author={Lad, Vedang and Mueller, Jonas},
        booktitle={ICML Workshop on Data-centric Machine Learning Research},
        year={2023}
    }

</details>

<details><summary><a href="https://arxiv.org/abs/2305.16583"> Detecting Errors in Numerical Data (DMLR '24)</a> (<b>点击显示 bibtex</b>) </summary>

    @inproceedings{zhou2023errors,
        title={Detecting Errors in a Numerical Response via any Regression Model},
        author={Zhou, Hang and Mueller, Jonas and Kumar, Mayank and Wang, Jane-Ling and Lei, Jing},
        booktitle={Journal of Data-centric Machine Learning Research},
        year={2024}
    }

</details>

<details><summary><a href="https://arxiv.org/abs/2207.03061"> Out-of-Distribution Detection (ICML '22)</a> (<b>点击显示 bibtex</b>) </summary>

    @inproceedings{kuan2022ood,
        title={Back to the Basics: Revisiting Out-of-Distribution Detection Baselines},
        author={Kuan, Johnson and Mueller, Jonas},
        booktitle={ICML Workshop on Principles of Distribution Shift},
        year={2022}
    }

</details>

<details><summary><a href="https://arxiv.org/abs/2210.06812"> CROWDLAB for Data with Multiple Annotators (NeurIPS '22)</a> (<b>点击显示 bibtex</b>) </summary>

    @inproceedings{goh2022crowdlab,
        title={CROWDLAB: Supervised learning to infer consensus labels and quality scores for data with multiple annotators},
        author={Goh, Hui Wen and Tkachenko, Ulyana and Mueller, Jonas},
        booktitle={NeurIPS Human in the Loop Learning Workshop},
        year={2022}
    }

</details>

<details><summary><a href="https://arxiv.org/abs/2301.11856"> ActiveLab: Active learning with data re-labeling (ICLR '23)</a> (<b>点击显示 bibtex</b>) </summary>

    @inproceedings{goh2023activelab,
        title={ActiveLab: Active Learning with Re-Labeling by Multiple Annotators},
        author={Goh, Hui Wen and Mueller, Jonas},
        booktitle={ICLR Workshop on Trustworthy ML},
        year={2023}
    }

</details>

<details><summary><a href="https://arxiv.org/abs/2305.15696"> Detecting Dataset Drift and Non-IID Sampling (ICML '23)</a> (<b>点击显示 bibtex</b>) </summary>

    @inproceedings{cummings2023drift,
        title={Detecting Dataset Drift and Non-IID Sampling via k-Nearest Neighbors},
        author={Cummings, Jesse and Snorrason, Elías and Mueller, Jonas},
        booktitle={ICML Workshop on Data-centric Machine Learning Research},
        year={2023}
    }

</details>

要了解/引用上文未介绍的 cleanlab 其他功能，请查阅我们的 [Blog](https://cleanlab.ai/blog/learn).


## 其他资源

- [展示本软件包实际应用的 Example Notebooks](https://github.com/cleanlab/examples)

- [Cleanlab 博客](https://cleanlab.ai/blog/learn)

- [博客文章：Confident Learning 简介](https://l7.curtisnorthcutt.com/confident-learning)

- [NeurIPS 2021 论文：Pervasive Label Errors in Test Sets Destabilize Machine Learning Benchmarks](https://arxiv.org/abs/2103.14749)

- [数据-centric AI（Data-centric AI）简介（MIT IAP 课程）](https://dcai.csail.mit.edu/)

- [过往版本的发布说明](https://github.com/cleanlab/cleanlab/releases)

**有兴趣参与贡献？** 请参阅 [贡献指南](CONTRIBUTING.md)、[开发指南](DEVELOPMENT.md) 以及[实用贡献建议](https://github.com/cleanlab/cleanlab/wiki#ideas-for-contributing-to-cleanlab).

**有问题？** 请查看[我们的常见问题](https://docs.cleanlab.ai/stable/tutorials/faq.html) 和 [Github Issues](https://github.com/cleanlab/cleanlab/issues?q=is%3Aissue).