📊 Vectorization / Embeddings Index

🗺️ Learning Plan

Vectors / Embeddings

1. Why Token IDs Are Not Enough
2. One-Hot vs Dense Vectors
3. Embeddings
4. Embedding Dimensions
5. How Embeddings Are Learned
6. Similarity in Vector Space
7. Static vs Contextual Embeddings
8. Vectorization to Attention Bridge

📌 Core overview

• 00-Vectorization

✅ What is already covered in this folder

This folder already covers the important foundations from the older note:

• token IDs vs meaning
• one-hot vs dense vectors
• embeddings
• embedding dimensions
• how embeddings are learned
• similarity in vector space
• contextual embeddings
• bridge to attention

♻️ What should still be reused from the older note

The older single-file note also had valuable material like:

• strong real-world scenarios
• practical intuition and analogies
• setup steps for running examples locally
• external tools and visualizations
• code-heavy demos
• exercises
• debugging tips
• resources

That content should be reused across the folder instead of being lost.

The current reuse plan is:

• 00-Vectorization.md now keeps the stronger big-picture intuition and learning path from the older note
• 06-Similarity-in-Vector-Space.md keeps semantic search, recommendation, and analogy intuition
• 07-Static-vs-Contextual-Embeddings.md keeps the bank-style context examples and context-sensitive meaning
• 08-Vectorization-to-Attention-Bridge.md keeps the transition from embeddings to attention
• 3.d-Positional-Encoding.md keeps positional information in the later attention/transformer stage
• practical exercises like Build a Semantic Search Engine and Word Analogy Solver currently still exist only in 00-Vectorization-old.md and have not yet been promoted into a new lab or companion note

Missing practical content still living in the old note

The following useful hands-on material is not gone, but it is still only in 00-Vectorization-old.md:

• Exercise 1: Build a Semantic Search Engine
• Exercise 2: Word Analogy Solver
• additional code-heavy demos
• debugging-oriented practice material

So right now:

• the new folder has the cleaner conceptual structure
• the old note still holds some practical exercise content that has not yet been redistributed

Useful external tools kept as references

• TensorFlow Embedding Projector
• Word2Vec Visualizer
• Embedding Comparator
• GloVe Visualization
• UMAP Explorer

These do not need to live in one giant note, but they should remain available from the folder index.