Machine Learning · Handwritten Notes · Collection guide

Handwritten Machine Learning Notes

October 2026 · 3 min read

These eight guides expand my six handwritten PDFs into worked explanations. Each begins with visible data, follows the calculation, and then connects it to the general formula. The original PDFs and selected page scans sit alongside the explanations.

Follow the ideas from estimation to prediction

Start with a coin to understand likelihood and Bayesian updating. Carry the same ideas into linear regression, then compare two ways to build a classifier. Then group observations with K-means, study PCA and LDA projections separately, and use LDA to classify a new point.

MLE, MAP, and Bayesian Prediction

Five coin tosses take us from likelihood to a posterior distribution and the next-toss prediction.

Linear Regression: From Gaussian Noise to Ridge and Lasso

Fit three students' scores, derive squared error, and see how Gaussian and Laplace priors give ridge and lasso penalties.

Naïve Bayes: From Probability Models to Classification

Follow the notes from probability estimation to feature models, smoothing, and the linear and sigmoid decision rules.

Logistic Regression: From Scores to Learning

Turn scores into probabilities and work through cross-entropy, gradients, and one training update.

K-means: Build and Check the Clusters

Group six points, move their centers, and calculate elbow, silhouette, and Dunn scores.

PCA: Preserve Variation in Fewer Dimensions

Calculate covariance, principal directions, explained variance, and reconstruction from four visible points.

Fisher LDA: Find a Projection That Separates Classes

Use the original ten-point dataset to calculate class means, within-class scatter, and a Fisher LDA projection.

LDA Classification: From Class Distributions to a Prediction

Fit class means, a shared covariance, and priors from eight rows, then calculate class scores and a posterior prediction.

Keep the data beside the calculation

The first four guides build on likelihood, but each repeats the data needed for its own example. The clustering and projection guides can be read independently. Basic algebra is enough to start; derivatives and matrix notation are introduced where they explain a calculation.

Each guide has a section called “From the handwritten notes.” Open it to compare a selected original page with the expanded explanation. Where the handwritten derivation skips an intermediate step or uses a different convention, the surrounding text spells out the calculation. All essential formulas are also written as searchable text.

The six original PDFs

These are the complete source documents. Page counts include blank trailing pages.
Original notes	Pages	Expanded guide
MLE.pdf	6	MLE, MAP, and prediction
Linear Regression.pdf	3	Linear regression
Naïve bayes.pdf	8	Naïve Bayes
Logistic Regression.pdf	4	Logistic regression
Clustering.pdf	5	K-means clustering
PCA and LDA.pdf	6	PCA, LDA projection, and LDA classification

Reproduce the worked numbers

The companion Python script recreates the numerical examples and explanatory figures. It checks the hand calculations against numerical solvers and scikit-learn, including the original LDA projection and the new classification example. The small tables printed in each guide contain all the observations used.

python -m pip install -r requirements.txt
python calculations.py

Run these commands from this notes folder. The script writes the figures and numerical results into its assets directory. The original PDFs and their page scans are preserved separately.

Continue into the topic guides

For broader coverage, continue with linear regression, logistic regression, or dimensionality reduction. To explain an individual fitted prediction, follow the linear model explainability series.

The earlier decision-tree handwritten notes remain available in the decision-tree section.