CS 310 — Python for Data Science

Bellevue College · Fall 2026
Pedro Albuquerque · pedro.albuquerque@bellevuecollege.edu

---

About the Course

This course explains the engine behind the newest and most powerful data science models — so that we remain accountable for their usage. That means being able to:

• Provide criticism — question model outputs instead of accepting them blindly.
• Make adjustments — tune and correct models when they fall short.
• Drive innovation — build new solutions on top of a solid understanding.

Understanding is only half of the story — we also build: practical Python implementations of every model we study, leveraging LLM tools and agents such as Gemini, working hands-on in Google Colab — no local setup required.

In one sentence: understand the models deeply enough to use, judge, and improve them — with the human in charge and AI as technological support.

Lectures

Every lecture is a self-contained Jupyter notebook. Click the badge to open it directly in Google Colab.

#	Lecture	Topics	Open in Colab
1	Introduction	The 4th Industrial Revolution · Data Generating Process · Supervised Learning · Loss Functions & ERM · Strong Law of Large Numbers · Uniform Convergence · VC Dimension & PAC Learning
2	Complexity Control, Validation, and Cross-Validation	Why Control Complexity · Structural Risk Minimization · Nested Polynomial Models · Bias–Variance Decomposition · Train/Test Splits · Implicit vs. Explicit SRM · Ridge, Lasso, AIC/BIC · K-Fold & Nested Cross-Validation
3	Inferential Models: Cobb–Douglas & Differential Equations	Two Paradigms (Model First vs. Data First) · The Cobb–Douglas Production Function · Solving PDEs by Separation of Variables · Estimation & Hypothesis Testing on the 1899–1922 Data · Jensen's Inequality · A Modeling Recipe with Differential Equations · Elasticities · Reading a Regression Table
4	Structural Equation Modeling	One Covariance-Structure Framework (semopy) · Correlation, Regression, Path Analysis & Mediation, Confirmatory Factor Analysis, and Simultaneous Equations as Special Cases · The Full ACSI Model · Why a Fixed Model Needs No Train/Validation Split · Hyperparameters in Disguise (Features, Polynomial Degree, Transformations, Modification Indices) · The Three-Way Split & K-Fold Cross-Validation · Post-Selection Inference & the Garden of Forking Paths

Each notebook includes worked examples with runnable code and a set of exercises, each accompanied by a Suggested Gemini Prompt you can paste into Colab's built-in Gemini assistant.

Getting Started

Option 1 — Google Colab (recommended). Click a Colab badge above. Everything runs in the browser; no installation needed.

Option 2 — Run locally.

git clone https://github.com/PedroBSB/CS310.git
cd CS310
pip install numpy pandas matplotlib scipy statsmodels scikit-learn semopy jupyter
jupyter notebook

To load a dataset from this repository inside any notebook (works in Colab and locally):

import pandas as pd
url = "https://raw.githubusercontent.com/PedroBSB/CS310/main/Data/College.csv"
college = pd.read_csv(url)

Datasets

The Data/ folder contains all datasets used in lectures, exercises, and assignments. Most come from An Introduction to Statistical Learning (James, Witten, Hastie, Tibshirani).

Dataset	Description
Advertising.csv	Sales vs. TV, radio, and newspaper advertising budgets
AirPassengers.csv	Classic monthly airline passengers time series (1949–1960)
Auto.csv / Auto.data	Gas mileage, horsepower, and other specs for 392 vehicles
Bikeshare.csv	Hourly usage of a bike sharing program in Washington, DC
Boston.csv	Housing values and neighborhood attributes in Boston suburbs
BrainCancer.csv	Survival times for patients with primary brain tumors
Caravan.csv	Caravan insurance purchases with 85 customer attributes
Carseats.csv	Simulated child car seat sales at 400 stores
Ch12Ex13.csv	Gene expression data for unsupervised learning exercises
cobb_douglas_1928.csv	The original Cobb & Douglas (1928) data: US manufacturing output, capital, and labor indexes, 1899–1922
College.csv	Statistics for 777 US colleges (applications, tuition, graduation rate)
Credit.csv	Credit card balance and customer demographics
CreditCard.csv	Credit card applications and spending behavior
Default.csv	Simulated credit card default data
Fund.csv	Returns of 2,000 fund managers over 50 months
Heart.csv	Cleveland Clinic heart disease diagnostic data (303 patients)
Hitters.csv	Major League Baseball player statistics and salaries
Income1.csv / Income2.csv	Simulated income vs. education (and seniority)
OJ.csv	Orange juice purchase choices (Citrus Hill vs. Minute Maid)
Portfolio.csv	Simulated asset returns for portfolio risk estimation
quadratic_model.csv	Simulated quadratic DGP for model selection (SRM, AIC/BIC, CV)
Publication.csv	Time to publication of clinical trial results
Smarket.csv	Daily S&P 500 returns, 2001–2005
Wage.csv	Wages and demographics for male workers in the Mid-Atlantic
Weekly.csv	Weekly S&P 500 returns, 1990–2010
Customers.csv / Orders.csv / Products.csv / Location.csv	Relational retail tables for pandas joins and aggregation
bank_transactions.csv	Bank transaction records for large-scale data wrangling
marketing_campaign.csv	Customer personality and campaign response data
acsi_data.csv	American Customer Satisfaction Index survey data
bfi.csv	Big Five personality inventory (25 items) for factor analysis

Using AI in This Course

You are encouraged to use Gemini (and other LLM tools) throughout this course — that is part of the point. But two failure modes must be avoided:

1. Fear: rejecting AI and losing its power as a tool.
2. Blind delegation: handing all reasoning and responsibility to the machine.

"A computer can never be held accountable, therefore a computer must never make a management decision." — IBM internal training manual, 1979

Run, read, and criticize every line of AI-generated code before you submit it. Accountability — legal, ethical, and professional — always rests with the human who acts on the outputs.

---

_{CS 310 · Python for Data Science · Bellevue College}