Ten notebooks that trace NLP's evolution from counting to attention

What it does

This repository is a collection of ten Jupyter notebooks that form a guided curriculum in natural language processing. It starts with tokenization and classical techniques like bag-of-words and TF-IDF, progresses through Naive Bayes and topic modeling, and ends with word embeddings, RNNs, attention, and Transformers. The material is built to run in Google Colab or Kaggle, so you can ignore your local GPU situation.

The interesting bit

The notebooks are sequenced to mirror the field’s actual history—counting words, then adding statistics, then neural networks, then self-attention—so you understand why each technique was necessary. The author emphasizes “core intuition” over black-box library calls, which is rarer than it should be in tutorial repositories.

Key highlights

• Ten sequential notebooks covering NLP from preprocessing to Transformers
• Mixes classical ML (LDA, Naive Bayes) with deep learning (RNNs, Attention, Transformers)
• Explicitly designed for cloud kernels; no local setup required
• Focuses on conceptual evolution and intuition rather than API syntax
• MIT licensed

Caveats

• Several notebooks rely on gigabyte-scale Kaggle datasets that do not auto-attach; you must manually link them before running
• Assumes you already know basic Python; this is not a programming primer
• The code is pedagogical by design, so don’t expect production-hardened pipelines or novel architectures

Verdict

Worth bookmarking if you want a structured, historical tour of NLP fundamentals without wrangling CUDA drivers. Give it a pass if you need a reusable framework or pre-trained model weights.