# competition-baseline: A Curated Collection of Data Science Competition Baselines > competition-baseline is a Datawhale China repository that collects baseline solutions for data science and machine learning competitions. Each baseline is a Jupyter Notebook or Python script presenting a working starting point for a specific contest, covering data mining, computer vision, NLP, and recommendation systems. **datawhalechina/competition-baseline** — 数据挖掘、计算机视觉、自然语言处理、推荐系统竞赛知识、代码、思路 - Repository: https://github.com/datawhalechina/competition-baseline - Website: http://coggle.club/ - Stars: 4,763 · Forks: 1,082 - Language: Jupyter Notebook - License: GPL-3.0 - Published: 2026-09-10 · Updated: 2026-09-10 · Language: en - Canonical page: https://hysenlabs.com/projects/datawhalechina-competition-baseline ## What a Baseline Is and Why This Collection Exists In data science competitions, a baseline is the simplest viable solution that produces a valid submission. It establishes the score floor, demonstrates the data pipeline from raw files to prediction output, and gives competitors a reference point for how much their more complex models actually improve on a working starting point. The README explains the motivation: compared to winning solutions, baselines are simpler, easier to follow, and more practical for learning. Winning code often depends on hardware, lucky hyperparameter choices, or ensemble tricks that are hard to reproduce. A baseline is reproducible by design. Datawhale, the organization behind this repository, runs it as a shared resource for competition newcomers and enthusiasts. ## Repository Structure and Navigation The top-level directory contains a LICENSE file, a README.md in Simplified Chinese, a competition/ folder, a docs/ folder, and a tutorial/ folder. Each competition gets its own subfolder inside competition/, named after the contest. Those subfolders contain Jupyter Notebooks, Python scripts, and sometimes data preprocessing utilities. The README lists competitions in reverse chronological order, so the most recently added entries appear first. The competition calendar that tracks active contests is at coggle.club. A Gitee mirror at gitee.com/coggle/competition-baseline serves users in mainland China who encounter slow access speeds from the GitHub version. ## How to Access and Run a Baseline To use any baseline in the collection, clone the repository and navigate to the relevant competition folder. There is no package to install for the repository itself: ```bash git clone https://github.com/datawhalechina/competition-baseline ``` Open the Jupyter Notebook for the competition you are working on. Each notebook documents its own dependencies at the top. Most baselines use standard Python data science packages. The competition datasets are not included in the repository; they must be downloaded from the platform hosting the contest. The README links each competition to its registration page or data download URL, so the path from the baseline notebook to the required data is one click. For Chinese-language users who want to follow along with guidance as they study the material, the Coggle数据科学 WeChat public account and the Zhihu column link listed in the README are companion resources outside the repository. ## Competition Coverage: Scope and Depth The collection covers a broad range of tasks. On the NLP side, past entries include Chinese question similarity (iFLYTEK 2021) and dialogue generation tasks. Computer vision entries include flower classification with PaddlePaddle, image retrieval, deepfake detection (both image and audio/video tracks), and figure skating pose estimation. Tabular and time-series tasks include supply chain demand forecasting, offline store sales prediction, and wind power output prediction. AI safety tasks such as adversarial attack and defense are also represented. The depth varies: some baselines are minimal end-to-end pipelines, while others include study notes and links to further reading. The CCF BDCI 2021 section alone lists fourteen separate competition tracks with individual baselines. ## Limitations: What the Repository Does Not Provide The baselines in this collection are starting points, not winning solutions. A baseline typically scores in the lower half of a competition leaderboard. The repository does not collect winning code or advanced ensemble techniques. It also does not maintain the notebooks over time: once a competition ends and its datasets are no longer available, the baseline becomes a reference document rather than runnable code. Many competition platforms eventually close dataset access, making older baselines difficult to run as written. The README is entirely in Chinese; the navigation, the resource links, and the competition descriptions all assume a reader comfortable with Simplified Chinese. ## License and Organizational Context The repository is licensed under GPL-3.0. That means any code you derive from these baselines and distribute must also be released under GPL-3.0. For competition submissions, this is rarely a concern since you submit predictions, not source code. For anyone who wants to build a training pipeline or tool based on the baseline code and distribute it commercially, the GPL terms require disclosure of the modified source. Datawhale China is an open-source learning community; the repository is one of several they maintain for educational purposes alongside other resources in machine learning, NLP, and computer vision. The homepage at coggle.club provides the competition calendar and further context about the organization's activities. The docs/ and tutorial/ directories at the top level contain additional learning materials beyond the competition-specific baselines. ## Comparison with Other Competition Learning Resources Kaggle and Tianchi both publish competition notebooks and discussion threads that fulfill a similar purpose: helping participants understand how to approach a problem. The key difference is access: those resources are tied to specific platform accounts and scattered across competition pages. competition-baseline aggregates materials from multiple platforms into a single repository that can be cloned and browsed offline. The trade-off is that the repository does not have a search interface or tagging system; finding relevant baselines for a specific task type requires browsing the competition/ folder manually or using git log to find recent additions. For participants in Chinese-language competition platforms specifically, the repository covers events that receive little attention on Kaggle forums, including iFLYTEK, DataFountain, and Tianchi competitions, making it a complementary resource for that audience. ## Conclusion competition-baseline suits Chinese-speaking data scientists who want a working starting point for competitions hosted on platforms such as Tianchi, iFLYTEK, and DataFountain. The material assumes familiarity with Python, Jupyter Notebooks, and the competition platforms where the datasets live. It is not a general-purpose machine learning library: each notebook solves a specific competition, and transferring techniques across competitions requires understanding the underlying methods, not just copying code. The README is in Chinese and the resource links point to Chinese-language platforms; international users will find the navigation difficult without reading Chinese. The last push was on 2026-07-22. Clone the repository and open the competition folder for your target contest, then download the required dataset from the platform link in the README entry before running the notebook. ## FAQ ### What is a competition baseline in data science? A baseline is the simplest working solution to a competition task: it takes the raw data, runs a basic model, and produces a valid submission. It establishes a score floor and lets competitors measure how much their more complex approaches actually improve on a reproducible starting point. ### Are the competition datasets included in this repository? No. The competition/ subfolders contain notebooks and scripts, not datasets. The README links each competition to its registration or data download page on the original platform. Datasets must be downloaded separately from those platforms. ### Is there a mirror for users in mainland China? Yes. The README points to a Gitee mirror at gitee.com/coggle/competition-baseline for users who encounter slow access speeds from the GitHub version. ## Sources - [datawhalechina/competition-baseline on GitHub](https://github.com/datawhalechina/competition-baseline) - [Issues](https://github.com/datawhalechina/competition-baseline/issues) - [License: GPL-3.0](https://github.com/datawhalechina/competition-baseline/blob/master/LICENSE) - [Project website](http://coggle.club/) - [README](https://github.com/datawhalechina/competition-baseline/blob/master/README.md) --- Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/datawhalechina-competition-baseline