Unit 06: Classification Classics
Three algorithms in three weeks, each introduced with a different dataset and a different motivating question. The consistent pattern throughout: devise or derive the algorithm in class first, then implement it, then reach for the library. You earn the library.
Concepts You’ll Learn About
- Logistic regression — the sigmoid function; log-odds; maximum likelihood vs. least squares; interpreting coefficients
- k-Nearest Neighbors — distance metrics; the curse of dimensionality; why k matters; approximate NN at scale
- Decision trees — recursive partitioning; information gain; entropy as an impurity measure
- Shannon entropy — bits as a measure of uncertainty; encoding strings; connection to Huffman coding
Topics
-
Logistic Regression on cancer data — binary classification on the Wisconsin breast cancer dataset; reading the classification report carefully because false negatives cost lives. View Download Run
-
Logistic Regression on loan data — open-ended: do the best logistic regression you can on loan_data.csv and defend your choices. Assignment: submit your notebook.
-
k-Nearest Neighbors — devise the algorithm on the board first, then implement it. Applied to handwritten digit recognition. A survey of approximate NN algorithms connects the classroom version to Spotify- and Netflix-scale systems. View Download Run View Download Run
-
Decision Trees — hand-classify the WillWait? restaurant dataset to motivate entropy and information gain, then run the lab on income data. View Download Run
-
Shannon entropy — information as surprise; encoding strings efficiently; connection to the compression ideas coming in Unit 07.
What’s next
Unit 07 closes out the classical-ML arc with SVMs, proper cross-validation, word embeddings, and ensemble methods — the professional-practice layer on top of the algorithms you now know.