Skip to content

Deep Learning Reference Hub

From-scratch implementations of core deep learning methods, written to be read.

Every module in this hub exposes the mathematics rather than hiding it. The implementations are teaching artifacts first and working code second, though they are held to both standards: each is covered by tests that assert against hand-computed values, closed-form results, or finite-difference gradient checks.

Install

git clone https://github.com/eima40x4c/Deep-Learning-Reference-Hub.git
cd Deep-Learning-Reference-Hub
pip install -e .

That covers NumPy, SciPy, and Matplotlib, which is everything the from-scratch implementations use. Once installed, a module is reachable the way any package is:

from dlhub.optimizers.adam import AdamOptimizer

Some documents also show the same idea in TensorFlow or PyTorch. To run those, add the frameworks:

pip install -e ".[frameworks]"

How this documentation is organised

The four sections answer four different questions, and each page belongs to exactly one of them. This is the Diátaxis split, adopted because a single page that tries to teach a concept, list its defaults, and walk through a task serves none of the three well.

Section Answers Read it when
Tutorials "Can you teach me to build one?" you are learning by doing, start to finish
How-to guides "How do I accomplish this task?" you have a goal and need the steps
Reference "What are the equations, shapes, and defaults?" you know what you want and need the detail
Explanation "Why does this work?" you want the derivation and the reasoning

Each section's own page states what belongs there and what does not, so a contributor can file a new page without reading this one.