References: From a line to a probability: logistic regression

Source material

Source material (conceptual spine):
• StatQuest with Josh Starmer: "Logistic Regression"
  Creator: Josh Starmer
  YouTube: https://www.youtube.com/watch?v=yIYKR4sgzI8
  Channel / site: https://statquest.org/
  License: as published on StatQuest's public YouTube channel (link-out only)

Source material (hands-on companion):
• Microsoft: "ML For Beginners" (Classification module)
  Repository: https://github.com/microsoft/ML-For-Beginners
  License: MIT

Clawdemy provides original notes, summaries, and quizzes derived from this material
for educational purposes. All rights to the original videos and curriculum remain
with their creators.

What this lesson draws from each source

• StatQuest’s “Logistic Regression” anchors the core idea: keep the linear combination, then map it to a probability with the S-curve, and use it to classify. StatQuest has a follow-on three-part series (coefficients, maximum likelihood, and R-squared with p-values) for readers who want the fitting details.
• Microsoft’s ML-For-Beginners Classification module is the hands-on companion: it builds logistic-regression classifiers in Python with scikit-learn on real datasets.

The “line plus a squash” framing, the exam worked example, and the explicit callback to gradient descent as the fitting method are Clawdemy’s own connective tissue across the track.

Going deeper

• StatQuest with Josh Starmer. The logistic regression intro plus its details series. StatQuest also has a clear explainer on maximum likelihood, the principle behind the loss logistic regression minimizes.
• Microsoft ML-For-Beginners: Classification. Project-based classification lessons in scikit-learn, the place to fit a logistic regression to real data yourself.

Adjacent topics

• Decision trees (the next lesson). A completely different approach to classification: ask a sequence of yes/no questions rather than draw a single straight boundary, which lets the model carve out non-linear regions.
• Softmax. The multi-class generalization of the sigmoid, used when there are more than two classes. The output layer of many neural-network classifiers.
• Precision, recall, and ROC (Phase 4). The tools that make the movable-threshold tradeoff in this lesson precise. The decision threshold is the dial; those metrics measure what it costs to turn it.

Community discussion

None selected for this lesson. Logistic regression is well covered by the StatQuest and Microsoft resources above. If a canonical discussion surfaces, it will be added at the next review.