Find ML Foundation Gaps with Mathematics and Python | CMU 10-301 Assignment 1

🌏 中文版

The official handout inside the ZIP is titled Homework 1: Background and contains written plus programming work. Written sections cover course policies, probability and statistics, linear algebra, calculus, geometry, and CS foundations. The programming component is a majority-vote classifier that reads TSV files and writes train/test labels and error metrics. The ZIP supplies heart and education splits, a LaTeX template, and reference outputs, but no majority_vote.py starter; hidden grading remains on Gradescope.

The capability being tested

Translate notation into array shapes, loops, and exact command-line output. Record inputs, outputs, and dimensions after each derivation. For code, hand-compute a tiny case before comparing with the supplied example output. Classify disagreements as mathematical, indexing, or formatting failures.

First executable action

After extracting the ZIP and creating majority_vote.py, run the handout's interface:

python majority_vote.py heart_train.tsv heart_test.tsv heart_train_labels.txt heart_test_labels.txt heart_metrics.txt

Then add tiny cases for reordered columns and constant labels. Completion means explaining the shapes in the main mathematics sections and matching both heart and education reference label/metric files line by line. It does not establish success on private tests.

Hand-compute a majority-vote classifier first

This is invented data, not an official test case. Training labels are 1, 0, 1, 1: three ones and one zero. Ignore features and predict 1 for every row. One of four training predictions is wrong, so training error is 1/4=0.25.

Test labels are 0, 0, 1. Keep the majority learned from training: predictions are 1, 1, 1, with error 2/3. Do not select a new majority from test labels; that would use test answers to change the classifier.

Check: For test labels 1, 1, 0, 1, what are predictions and error? Four ones, with error 1/4=0.25. This example deliberately avoids tied training counts. Follow the official handout for tie-breaking, columns, and output formatting rather than inferring them here.

Changelog

• 2026-10-11: Added an invented teaching example, worked reasoning, and a comprehension check with an answer.

References

• Spring 2026 coursework
• HW1 public bundle