41  Module 8 Cheat Sheet

Key concepts, definitions, and code from Chapters 19–20

A quick-reference summary of the essential ideas from Module 8. Click any section heading to jump to the full coverage in the book.


The ML Vocabulary

Artificial Intelligence (AI) is the broad goal of making machines perform tasks that require human-like intelligence. Machine Learning (ML) is the subset of AI where systems learn patterns from data rather than following hand-written rules. Data mining is the practice of discovering patterns in data — it overlaps heavily with ML but emphasizes discovery over prediction.

Term What it means Example
Supervised learning Learn from labeled examples to predict a known target Predicting house price from square footage
Unsupervised learning Find structure with no target variable Segmenting customers into groups
Reinforcement learning Learn by trial and error against a reward signal Game playing, robotics
Generative AI Produce new content resembling training data LLMs, image generation

Supervised learning splits into two problem types:

Type Target variable Example question
Regression Continuous number How much will this customer spend?
Classification Category Will this customer churn — yes or no?

Core Terminology

Term Definition
Feature (predictor, X) An input variable used to make a prediction
Label (target, y) The thing you are trying to predict
Observation One row — one example the model learns from
Model The learned mapping from features to label
Training The process of fitting a model to data
Generalization How well the model performs on data it has never seen

Framing the Problem

Before writing any code, establish what you are predicting, why it matters, and how you will know if the model is good enough.

A pre-modeling checklist:

  1. What is the business question? State it in one sentence, without jargon.
  2. What is the target variable? If you cannot name a column, it is not yet an ML problem.
  3. Is it regression or classification? This determines everything downstream.
  4. What does success look like? Define the metric and the threshold before you model.
  5. Do you have the data? Features must be available at prediction time, not just historically.
  6. What is the cost of being wrong? False positives and false negatives rarely cost the same.

Distinguish model performance metrics (R², accuracy) from business performance metrics (revenue retained, hours saved). A model can win on the first and lose on the second.


The Train/Test Split

Evaluating a model on the data it learned from tells you how well it memorized, not how well it predicts. Hold out a portion of the data before training and never touch it until the end.

from sklearn.model_selection import train_test_split

X = df[['feature_1', 'feature_2']]   # features (2-D)
y = df['target']                      # label (1-D)

X_train, X_test, y_train, y_test = train_test_split(
    X, y,
    test_size=0.2,       # 20% held out
    random_state=123     # reproducible split
)
Parameter Purpose
test_size Fraction held out (0.2–0.3 is typical)
random_state Fixes the shuffle so results reproduce
stratify=y Preserves class balance — use for classification

Data Leakage

Data leakage is when information that would not be available at prediction time sneaks into training. It produces models that look excellent in development and fail in production.

Leak Why it breaks Fix
Scaling before splitting Test set statistics influence training Split first, then fit the scaler on training data only
A feature caused by the target days_until_cancellation predicts churn perfectly Drop features that only exist after the outcome
Repeatedly tuning on the test set The test set stops being unseen Use cross-validation for tuning (Module 12)
Random splits on time series Future data leaks into the past Split chronologically

Ethical Considerations

Concern Question to ask
Fairness Does the model perform equally well across demographic groups?
Privacy Should these features be used, even if legally available?
Interpretability Can you explain a decision to the person it affects?
Accountability Who is responsible when the model is wrong?

Common Pitfalls

Mistake Fix
Passing a 1-D Series as X Use double brackets: df[['col']] keeps it 2-D
Forgetting random_state Results change every run and cannot be compared
Evaluating on training data Always report metrics on the held-out test set
Choosing a metric after seeing results Define success before modeling
Treating a modeling task as ML when rules would do If a simple rule works, use the rule