How fastai/course22 works
Project type and purpose
• This is not a traditional “application” or microservice-based system: it’s a content repository for the 2022 edition of the “Practical Deep Learning for Coders” course (course.fast.ai).
• It bundles Jupyter notebooks, slide decks, Excel spreadsheets, and a small set of tooling scripts to assemble/clean the content.
• It includes a .devcontainer configuration to provide a reproducible environment (Docker + Codespaces) for running the notebooks.
File structure
Top-level directories and their roles:
• / (root): the _working copies_ of the notebooks (00-… .ipynb through 10-… .ipynb).
• /clean: a cleaned version of each notebook (no outputs or prose) – used for distribution or automated testing.
• /slides: pre-built slide decks in PDF/PPTX and some notebook-based slides.
• /xl: Excel files used as teaching datasets or worked examples.
• /tools: Python scripts (clean.py, download.py) and metadata (kaggles.txt) that automate cleaning/notebook preparation or data fetch.
• /.devcontainer: Dockerfile, environment spec (requirements.txt), and Codespaces configuration.
Key files:
• .gitignore
• getting-started-with-codespaces.md – instructions for launching this repo in GitHub Codespaces.
• README.md – high-level overview.
README for additional insights
• Confirms that the repo root holds notebooks, with a parallel clean folder.
• tools are for building/maintaining the repo (not directly lesson content).
• There is a companion website (course.fast.ai) that actually hosts the interactive lessons.
a. Main components:
• Developer Environment (DevContainer + Codespaces)
• Raw Notebooks (root folder)
• Clean Notebooks (clean folder)
• Slide Decks (/slides)
• Data Spreadsheets (/xl)
• Tooling Scripts (/tools)
• External Hosting/Delivery (GitHub, course.fast.ai, Binder/Colab)
b. Relationships & interactions:
• DevContainer spins up a Jupyter server with the dependencies from .devcontainer/requirements.txt.
• /tools/download.py fetches or updates raw data (e.g., Kaggle files listed in kaggles.txt) into /xl if needed.
• /tools/clean.py reads the raw notebooks in / → strips outputs & text → writes cleaned copies into /clean.
• Authors iterate on raw notebooks in the DevContainer; commit changes to GitHub.
• GitHub Actions or CI (implied) could run tools/clean.py automatically, validate that /clean is up to date, and upload notebooks to the public site.
• Slide decks are maintained separately in /slides; they may be generated from notebooks or hand-edited, then stored in Git.
• End users can pull the repo and launch it via Codespaces or Binder, or view the course materials on course.fast.ai.
c. Architectural patterns/design principles:
• Separation of concerns: content (notebooks/slides/data) vs. build tooling vs. environment config.
• Reproducible dev environment: containerized dev via Docker / Codespaces.
• Content pipeline: raw → cleaned → published.
d. Key technologies/frameworks:
• Python & Jupyter notebooks (FastAI framework within notebooks)
• Docker / DevContainer / GitHub Codespaces
• Git/GitHub for version control & hosting
• Optional CI (e.g., GitHub Actions) for automated validation/cleaning
• Excel (.xlsx) as static data source
• Slide formats: PDF, PPTX, Jupyter-based slides