A visual companion to machine learning

Machine learning, illustrated.

See how a model works, from the whole pipeline to a single weight. Follow one penguin from raw measurements to a prediction, turn the dial on a model, and compare the different ways of answering the same question. Built for Machine Learning (CSI 330) at Muhlenberg College, Fall 2026. Every number is what scikit-learn computes on the course’s own data.

No. 00The whole pipeline, one bird at a time
→ next bird · 1–6 a stage · click a panel
Three-class logistic regression on bill length and depth: 239 penguins to train, 103 to test. The page recomputes every number and agrees with scikit-learn on all 103.

9 entries · 10 interactive studies · weeks 1–8, a growing reference

01 Explore by topic

Foundations 1 entry

Bias and variance Too simple, too flexible, and the error that tells them apart

Classification 3 entries

k-nearest neighbors Classify a new bird by asking the most similar ones Logistic regression A line, then a squash, then a probability Gaussian naive Bayes Prior × likelihood, weighed on a balance

Regression 2 entries

Linear regression and least squares The line the springs settle on k-NN regression Predict a number by averaging the neighbours

Evaluation 2 entries

Classification metrics One threshold, four kinds of answer Regression metrics MAE, RMSE and R² from the same misses

Validation 1 entry

Cross-validation Every observation tested once, and the spread kept

02 Building a model, in order

From a question to a number you can trust. Each step points to the machines that explain it; the number on the right is the panel of the pipeline above where it happens.

  1. Frame the question What is predicted, from what: a class or a number. k-nearest neighbors · Linear regression and least squares
  2. Split the data Hold back a test set before anything is learned. Cross-validation · Bias and variance
  3. Scale the features Put measurements on one footing, using the training data only. k-nearest neighbors
  4. Choose and fit a model A rule with weights, chosen to make the training loss small. Logistic regression · Gaussian naive Bayes · Linear regression and least squares · k-NN regression
  5. Turn scores into answers Probabilities, then a decision, with a threshold you choose. Logistic regression · Classification metrics
  6. Measure on the test set Count the kinds of right and wrong, or the sizes of the misses. Classification metrics · Regression metrics
  7. Validate the choice Rotate the held-out part to see how much the score moves. Cross-validation

Then compare the choices you made along the way in Why this, not that.

03 The interactive collection

  1. 01Bias and varianceweek 1
  2. 02k-nearest neighborsweek 2
  3. 03Logistic regressionweek 2
  4. 04Gaussian naive Bayesweek 3
  5. 05Linear regression and least squaresweek 4
  6. 06k-NN regressionweek 5
  7. 07Classification metricsweek 6
  8. 08Regression metricsweek 7
  9. 09Cross-validationweek 8