Skip to content
Projects

emptymalei/deep-learning has 28 GitHub stars and 1 forks. GitHot measured 0 stars in 24 hours. Last checked 2026-11-15 17:44 UTC. Last push 2026-11-13.

How these counts are measured

Measured star growth

in 24h
0
in 7 days
Update delayed
in 15 days
Update delayed
in 1 month
Update delayed

Net star changes over overlapping periods, not daily totals. Missing measurements are left blank. Stars measure attention, not software quality.

About

time series forecasting with deep learning

Machine Learning

README.md

Original

Deep Learning

When I switched to data science, I built my digital garden, datumorphism. I deliberately designed this digital garden as my second brain. As a result, most of the articles are fragments of knowledge and require context to understand them.

Making bricks is easy but assembling them into a house is not. So I have decided to use this repository to practice my house-building techniques.

I do not have a finished blueprint yet. But I have a framework in my mind: I want to consolidate some of my thoughts and learnings in a good way. However, I do not want to compile a reference book, as datumorphism already serves this purpose. I should create stories.

How to Contribute

This repository contains mostly markdown files. To make sure we have the same conventions, we have added markdownlint tools to pre-commit. So please install pre-commit then run the following command the first time you cloned the repository.

pre-commit install

Build and Preview the Site

dl.leima.is is built with Zensical, the successor to Material for MkDocs, by scripts/build_site.sh. The deploy workflow runs the same script. It only needs uv:

scripts/build_site.sh                           # builds build/site/
uvx zensical==0.1.1 serve -f build/mkdocs.yml   # live preview

Zensical reads mkdocs.yml but cannot run third-party MkDocs plugins yet. So the script first exports the marimo playgrounds, then runs scripts/stage_site.py, which converts notebooks and citations into plain Markdown under build/ (git-ignored). zensical serve watches build/docs/, so after editing a page under dl/, rerun uv run --python 3.12 scripts/stage_site.py.

Notebook outputs are taken from your local, git-ignored .ipynb files when they exist. CI has none, so the published notebook pages show code only. Run python3 scripts/fix_links.py to check that every Markdown link resolves relative to its source file.

Python Environment

The project uses Python 3.12 (see .python-version) and Poetry 2 for the notebooks. Install the requirements using

poetry env use 3.12
poetry install

The site build does not need this environment; it only uses uv.

Developing Notebooks

We use jupytext to sync the .py files to .ipynb files. .ipynb files are ignore in git. Please pair the .py file with the .ipynb using jupytext in jupyterlab first.

Interactive Widgets

Interactive explainers run entirely in the reader's browser and need no build plugin. Put a dl-widget block in a page; whatever is inside it is the fallback for readers without JavaScript, for print, and for text exports.

<div class="dl-widget" data-widget="backtest-windows" markdown>

![Sliding Blocks](../assets/timeseries-evaluation.forecasting/block-slide.png)

</div>

Available widgets live in dl/assets/widgets/:

Widget What it shows Source of truth
backtest-windows Sliding vs expanding backtest folds, with a gap backtest-windows.js
vega A Vega-Lite chart exported from Python with Altair (data-spec="path/from/docs/root.vl.json") scripts/build_charts.py
embed A click-to-load iframe, e.g. a marimo WebAssembly playground (data-src, data-height) playgrounds/*.py

To add a widget:

  1. Create dl/assets/widgets/<name>.js that calls dlWidgets.register("<name>", (el, api) => { ... }). The api gives theme colours (api.colors(), api.onThemeChange(cb)), api.reducedMotion(), api.resolve(path) for files relative to the docs root, and api.loadScript(url) for pinned CDN libraries.
  2. Add the file to extra_javascript in mkdocs.yml after assets/widgets/loader.js.
  3. Respect prefers-reduced-motion (no autoplay) and check the widget in light mode, dark mode and at phone width.

To regenerate Altair charts, run uv run --python 3.12 scripts/build_charts.py.

marimo playgrounds (playgrounds/*.py) are exported to WebAssembly by scripts/build_playgrounds.sh into the git-ignored dl/assets/playgrounds/. scripts/build_site.sh runs it, with marimo pinned there, so every build and deploy includes them. When the export is missing, the embed widget shows its Markdown fallback.