A La Carte ML: Notebooks from Scratch to BERT

What it does A curated collection of Jupyter notebooks spanning core ML concepts, NLP, computer vision, and Transformers. Each notebook isolates a specific technique—ranging from linear regression implemented from scratch to fine-tuned BERT emotion classification—intended as minimal, reusable reference material for learning and research.

The interesting bit The repo treats notebooks as modular units rather than a monolithic course. You can jump from bare-metal stochastic gradient descent to a Graph Neural Network on the Cora dataset, or compare bag-of-words classifiers against a Transformer-based NER pipeline without wading through chapter one.

Key highlights

• Foundational math (computational graphs, SGD) alongside modern architectures (Transformers, GNNs, Siamese networks)
• “From scratch” implementations (linear regression, logistic regression, neural networks) mixed with PyTorch abstractions
• NLP progression from classic BoW and CBOW through text augmentation to fine-tuned BERT
• Transformer section implements attention, positional embeddings, and practical tasks like date-format translation and extractive question answering
• Most notebooks open directly in Colab; several also link to Kaggle
• Includes a dev container configuration for GitHub Codespaces

Caveats

• The README is essentially a large table of contents; you will need to open individual notebooks to judge depth and code quality
• The Computer Vision section is truncated in the source, so the full scope there is unclear

Verdict Useful if you are teaching yourself ML or need isolated, runnable examples to adapt for coursework. Skip it if you are looking for a unified library or production pipelines—this is a reading list that happens to execute.