# papers-in-100-lines-of-code: one directory per paper, no line counts, and a badge pointing at master

> A collection that reimplements published machine learning papers in short Python files, indexed by title, authors, link and date. The index never states how long any implementation is, one entry is sorted out of order, and the top badge links to a branch the repository does not use.

**MaximeVandegar/Papers-in-100-Lines-of-Code** — Implementation of papers in 100 lines of code.

- Repository: https://github.com/MaximeVandegar/Papers-in-100-Lines-of-Code
- Stars: 2,898 · Forks: 256
- Language: Python
- License: MIT
- Published: 2026-09-10 · Updated: 2026-09-10 · Language: en
- Canonical page: https://hysenlabs.com/projects/maximevandegar-papers-in-100-lines-of-code

## The README's own badge links to a branch the repository does not use

The first line of the file is a badge whose target is a README path on a `master` branch:

```
https://github.com/MaximeVandegar/Papers-in-100-Lines-of-Code/blob/master/README.md
```

The default branch for this repository is `main`, so the one self-referential link in the file, the one that points at the README from the README, is written against a branch name that is not the one checked out. The second badge points at the MIT license page on opensource.org, which agrees with the recorded license. Nothing else in the file is affected, since every other link is an outbound arXiv or nature URL, but it does mean the file opens with a broken reference to itself and that anyone following it lands on a branch path rather than on the current contents.

## No entry states a line count, so the 100 in the name is unchecked

Every entry in the index has the same four parts: a level-five heading with the paper title, a line repeating the title with an arXiv or nature link, a line of authors in italics, and a date in backticks. Maxout Networks is dated `2013-02-18`, Adam is dated `2014-12-22`, Wasserstein GAN is dated `2017-01-26`. Nowhere in that structure is a file name, a line count, a language, a dependency, or a result. The constraint that gives the project its name appears in exactly one place, the single description line reading Implementation of papers in 100 lines of code, and nothing in the index enforces it or reports it. So a reader cannot tell from the listing whether an implementation is ninety lines or nine hundred, which is the one number the project is named after.

## One entry is dated two months before the entry above it

The index reads as a chronological list, and almost entirely is. The break comes between Adam, dated `2014-12-22`, and NICE, which follows it dated `2014-10-30`. NICE carries arXiv 1410.8516, an October 2014 identifier, so the date and the identifier agree with each other and disagree with the position in the list. Two months of entries are out of order at that point. A second inconsistency sits in the same style of line: the DCGAN heading reads Unsupervised Representation Learning with Deep Convolutional Generative Adversarial Networks, while the link line below it is shortened to Convolutional Generative Adversarial Networks. Neither problem changes what the code does, but both are the kind of drift that appears when an index is maintained by hand over years.

## One directory per paper, named from the title with spaces replaced

The repository is a flat set of top-level directories, one per paper, with each space in the title replaced by a low line and the capitalization left as the paper writes it. Examples visible in the tree include `Adam_a_Method_For_Stochastic_Optimization/`, `Deep_Image_Prior/`, `Fourier_Features_Let_Networks_Learn_High_Frequency_Functions_in_Low_Dimensional_Domains/`, and `Fast_and_Accurate_Deep_Network_Learning_by_Exponential_Linear_Units_ELUs/`. That naming carries the paper's own capitalization, so the Adam directory capitalizes For while the title in the index writes it lowercase, and a script that derives a directory name from an index entry would not reproduce it. Punctuation is dropped rather than transliterated, so the entry for Gaussian Error Linear Units becomes a directory name carrying only the letters and the parenthesized acronym. The non-paper entries sit among them rather than in a separate area: `.github/`, `.gitignore`, and `CODE_OF_CONDUCT.md` are all at the top level, next to the paper directories.

## The subject matter reaches past 2017 into 3D rendering and diffusion

The entries in the index run from Maxout Networks in early 2013 to the reinforcement learning and generative modelling work of 2016 and 2017. The directory names go considerably further. Among them are `3D_Gaussian_Splatting_for_Real_Time_Radiance_Field_Rendering/`, `Instant_Neural_Graphics_Primitives_with_a_Multiresolution_Hash_Encoding/`, `FastNeRF_High_Fidelity_Neural_Rendering_at_200FPS/`, `FreeNeRF_Improving_Few_shot_Neural_Rendering_with_Free_Frequency_Regularization/`, `InfoNeRF_Ray_Entropy_Minimization_for_Few_Shot_Neural_Volume_Rendering/`, `Implicit_Neural_Representations_with_Periodic_Activation_Functions/`, `Denoising_Diffusion_Probabilistic_Models/`, `Denoising_Diffusion_Implicit_Models/`, `High_Resolution_Image_Synthesis_with_Latent_Diffusion_Models/`, and `DreamBooth_Fine_Tuning_Text_to_Image_Diffusion_Models_for_Subject_Driven_Generation/`. So the collection is not a single era of deep learning, and a paper from the 2020s sits beside a maxout network implementation from 2013.

## One Nature link, one shortened title, and a list that stops mid-author

Citation style is uniform except where it is not. Every entry links to arXiv except Human-level control through deep reinforcement learning, which links to a nature.com article and carries seventeen authors against the three to seven of most entries. The same entry is also one of the few whose date, `2015-02-25`, is a magazine publication rather than a preprint. The list then runs out mid-line. The final entry shown is Proximal Policy Optimization Algorithms, linking to arXiv 1707.06347, and its author line begins *John Schulman, and stops there, with no second author, no date, and no following entry. What is present for that paper is the heading, the link, and the first of its authors.

## Nothing in the file tells you how to run any of it

There is no fenced code block anywhere in the visible file, no installation command, no dependency list, no Python version, and no usage section. The project is presented as an index and nothing more: title, link, authors, date, repeated. That leaves two practical gaps. A reader who wants to run an implementation has no starting point, since the file does not name the file inside the directory or the library it expects. And a reader who wants to compare two papers on brevity has nothing to compare, because the index reports no size. The Python language recorded for the repository is the only other hint about what the directories contain, and it applies to the whole collection rather than to any single paper. There is a further asymmetry between the early and late entries. A maxout network or an adam optimizer is a few dozen lines of arithmetic that a reader can check against the paper by eye, while a NeRF, a latent diffusion model or a DreamBooth fine-tune depends on pretrained weights and prepared data that a short file cannot carry with it. The index gives those two kinds of entry the same four lines of description, so nothing in the listing tells a reader which kind they are about to open.

## Conclusion

Use this collection the way its directory names invite, as a set of small reference implementations to read rather than a library to depend on, and judge each paper on its own terms because the index carries no line counts and no run instructions to check them against. Two things to know before you start. The scope has drifted well past the 2013 to 2017 core into 3D reconstruction and diffusion work, where a short file can still assume pretrained weights and datasets that are not in the repository. And the index is maintained by hand, with at least one entry out of date order and one link text that does not match its heading, so treat the listing as a starting point rather than an index you can trust without opening the file.

## FAQ

### What is papers-in-100-lines-of-code?

A collection of implementations of published papers, described in one line as Implementation of papers in 100 lines of code. Each paper gets its own top-level directory, and the index gives the title, a link, the authors, and a date for each one.

### Which papers are implemented in papers-in-100-Lines-of-Code?

The indexed entries run from Maxout Networks dated 2013-02-18 through Generative Adversarial Networks, Adam, Wasserstein GAN and Deep Residual Learning. The directory names reach further, to 3D Gaussian Splatting, Instant Neural Graphics Primitives, denoising diffusion models and latent diffusion.

### How does papers-in-100-Lines-of-Code organise the papers?

As a flat set of top-level directories, one per paper, with each space in the paper title replaced by a low line and its capitalization preserved, as in Adam_a_Method_For_Stochastic_Optimization. The .github/, .gitignore and CODE_OF_CONDUCT.md entries sit at the same level.

### Does papers-in-100-lines-of-code say how long each implementation is?

No. Each index entry carries a heading, a linked title, an author list and a date, and no entry reports a line count, a file name or a result. The 100 line figure appears only in the project name and the one-line description.

### How do I run one of the papers-in-100-Lines-of-Code implementations?

The file gives no installation command, no dependency list, no Python version and no usage section, and it does not name the file inside each paper directory. The recorded language for the repository is Python.

### What license is papers-in-100-lines-of-code under?

MIT, recorded for the repository and matching the badge at the top of the file, which points at the MIT license page on opensource.org.

## Sources

- [Issues](https://github.com/MaximeVandegar/Papers-in-100-Lines-of-Code/issues)
- [License: MIT](https://github.com/MaximeVandegar/Papers-in-100-Lines-of-Code/blob/main/LICENSE)
- [MaximeVandegar/Papers-in-100-Lines-of-Code on GitHub](https://github.com/MaximeVandegar/Papers-in-100-Lines-of-Code)
- [README](https://github.com/MaximeVandegar/Papers-in-100-Lines-of-Code/blob/main/README.md)

---

Hysen Labs editorial analysis, written from the project's own repository and release notes. Cite the canonical page: https://hysenlabs.com/projects/maximevandegar-papers-in-100-lines-of-code