CS114 Machine Learning Notes

This repository contains a set of Quarto notebooks for studying core machine learning topics. The notes are written as exam-friendly learning documents: each notebook combines conceptual explanation, mathematical derivation, visual intuition, executable demonstrations, and hand-written worked problems.

The current language of the notebooks is Vietnamese, with English technical terms kept where they are standard in machine learning, such as Logistic Regression, GLM, K-means, and Gradient Descent.

What This Project Is Doing

The goal is to turn dense machine learning theory into notes that are easier to study from:

• Explain the idea before the formula.
• Break long derivations into smaller algebraic steps.
• Add visual demonstrations for abstract concepts.
• Keep Python chunks executable so plots and demonstrations are reproducible.
• Write problem-set solutions by hand, not as code output, so the PDFs work as printable exam cheat sheets.
• Export every notebook to PDF in the output/ folder.

Notebook Set

The full combined document is available at output/CS114_Full_Notes.pdf. It is generated from all 9 source notebooks by scripts/build_full_notes.py.

Source notebook	Main topic	Generated PDF
quarto/LR_OLS_Derivation.qmd	Ordinary Least Squares derivation, residual checks, simple linear regression	output/LR_OLS_Derivation.pdf
quarto/LR_MLE_Matrix_Derivation.qmd	Linear regression from Gaussian MLE and matrix normal equations	output/LR_MLE_Matrix_Derivation.pdf
quarto/Classification_Logistic_Regression.qmd	Binary classification, sigmoid, cross-entropy, logistic regression optimization	output/Classification_Logistic_Regression.pdf
quarto/Exponential_Family_GLMs.qmd	Exponential family, GLMs, inverse links, Bernoulli/Poisson/Gaussian cases	output/Exponential_Family_GLMs.pdf
quarto/GLA.qmd	Generative learning, Bayes rule, Naive Bayes, Gaussian Discriminant Analysis	output/GLA.pdf
quarto/Decision_Tree.qmd	Decision trees, entropy, Gini, pruning, ensemble intuition	output/Decision_Tree.pdf
quarto/Neural_Network.qmd	Neural networks, layer shapes, activation functions, backpropagation	output/Neural_Network.pdf
quarto/Optimization_Bias_Variance_Regularization.qmd	Gradient descent, bias-variance decomposition, regularization, cross-validation	output/Optimization_Bias_Variance_Regularization.pdf
quarto/Unsupervised_Learning_K_Means.qmd	Unsupervised learning, K-means, centroid updates, elbow method	output/Unsupervised_Learning_K_Means.pdf

Repository Layout

.
+-- quarto/               # Quarto source notebooks
+-- html/                 # Rendered HTML previews and their *_files assets
+-- output/               # Rendered PDF files
+-- exam/                 # Sample and practice exams with step-by-step answer keys
+-- scripts/              # Rebuild scripts for generated source files
+-- temp/                 # Quarto/Jupyter intermediate files and scratch artifacts
+-- README.md             # Project overview

The .qmd files in quarto/ are the source of truth. The HTML, PDF, and temp files are generated artifacts.

Practice Exams

In addition to the notebooks, this repository contains a set of practice exams located in the exam/ directory. These are designed to test both mechanical calculation and deep conceptual understanding:

• sample_exam: The original recovered exam and its detailed answer key.
• exam_01 to exam_05: Five additional practice exams, each containing 6 standard questions covering the full scope of the notebooks, plus a final advanced question designed to challenge deeper theoretical understanding (e.g., Generalization in SGD, Curse of Dimensionality, MAP estimation for Regularization, Inductive Bias).
• Every exam is accompanied by a step-by-step mathematical answer key.

Exam Coverage Audit

The exam directory was checked against the notebook knowledgebase. The current notebooks cover the following exam skills:

Exam skill	Where to study
Simple OLS by hand, residuals, prediction	quarto/LR_OLS_Derivation.qmd
Matrix OLS, normal equations, Gaussian MLE to OLS	quarto/LR_MLE_Matrix_Derivation.qmd
Logistic probability, cross-entropy, gradient descent, Hessian convexity, Newton's method	quarto/Classification_Logistic_Regression.qmd
Bernoulli/Gaussian/Poisson exponential-family forms, GLM inverse links, MLE convexity	quarto/Exponential_Family_GLMs.qmd
GDA priors, means, shared variance/covariance, Gaussian density, linear boundary proof	quarto/GLA.qmd
Naive Bayes, conditional counts, Laplace smoothing, posterior normalization	quarto/GLA.qmd
Decision-tree entropy, Gini, information gain, regression-tree SSE/MSE	quarto/Decision_Tree.qmd
Neural-network shapes, forward pass, backpropagation, Hadamard product, vanishing gradient	quarto/Neural_Network.qmd
Dropout, inverted dropout, Batch Normalization forward and backward	quarto/Neural_Network.qmd
Gradient descent, Ridge/Lasso, cross-validation, bias-variance decomposition	quarto/Optimization_Bias_Variance_Regularization.qmd
K-means trace, distortion, elbow method, centroid proof, convergence and local minima	quarto/Unsupervised_Learning_K_Means.qmd

Current known caveat: sample_exam still contains a recovered question marked as missing in the original recovered exam file. The rest of the sample exam and exams 01-05 map to the notebooks above.

Requirements

To render the notebooks, the local setup should have:

• Quarto
• A working Python/Jupyter environment
• The science_env Jupyter kernel
• Common scientific Python packages:
  • numpy
  • pandas
  • matplotlib
• A LaTeX engine for PDF rendering, such as TinyTeX

The notebooks are configured with:

jupyter: science_env

So Quarto will try to execute code chunks using the science_env kernel.

Rendering

Preview a single notebook:

quarto preview .\quarto\Classification_Logistic_Regression.qmd

Render one notebook to PDF:

Push-Location .\quarto
quarto render .\Classification_Logistic_Regression.qmd --to pdf --output Classification_Logistic_Regression.pdf
Pop-Location
Move-Item -LiteralPath .\quarto\Classification_Logistic_Regression.pdf -Destination .\output\Classification_Logistic_Regression.pdf -Force

Important: avoid using --output-dir output for single-file renders in this folder. In this project, the safer pattern is to render the PDF beside the source file and then move that one PDF into output/.

Render one notebook to HTML and move the preview artifacts:

Push-Location .\quarto
quarto render .\Classification_Logistic_Regression.qmd --to html --output Classification_Logistic_Regression.html
Pop-Location
Move-Item -LiteralPath .\quarto\Classification_Logistic_Regression.html -Destination .\html\Classification_Logistic_Regression.html -Force
Move-Item -LiteralPath .\quarto\Classification_Logistic_Regression_files -Destination .\html\Classification_Logistic_Regression_files -Force

Build the full combined source and render the big PDF:

python .\scripts\build_full_notes.py
quarto render .\quarto\CS114_Full_Notes.qmd --to pdf --output CS114_Full_Notes.pdf
Move-Item -LiteralPath .\CS114_Full_Notes.pdf -Destination .\output\CS114_Full_Notes.pdf -Force

The combined PDF has a detailed table of contents and PDF bookmarks. Each source notebook starts at a new page for faster navigation.

Study Design

Each notebook is structured to support three study modes:

• Concept mode: read the explanations and visual intuition.
• Formula mode: use the boxed equations and cheat-sheet tables.
• Exam mode: practice the worked problems, which show written computation steps instead of relying on code.

Python chunks are kept for demonstrations and plots, but the problem-set solutions are written out manually so the notes remain useful on paper.

Current Status

All 9 notebooks have been rendered successfully to PDF in output/. The Python chunks compile under science_env, and the worked problem sections are written as step-by-step mathematical solutions.

The combined document output/CS114_Full_Notes.pdf has also been rendered successfully from all 9 notebooks.