The School of Numbers & Logic · mathematics, pure and applied
Data Science
Real data from first table to first model — collect it, clean it, chart it honestly, and stop short of claims it cannot support.
The full arc of a small, honest analysis — framing a question, wrangling a real dataset, and reporting what it can support.
Syllabus · 4 units · ~20 hours
Unit I — What Data Is
Observations, variables, and tidy tables · Kinds of variables and why the kind matters · Where datasets come from, and what that implies
Unit II — Questions Worth Asking
Descriptive, predictive, and causal questions · Turning a vague curiosity into an answerable question · The dataset you have versus the question you want
Unit III — First Analysis
Summaries: counts, means, medians, group comparisons · First charts: distributions and relationships · Missing values and suspicious entries · Documenting every cleaning decision
Unit IV — Saying What You Found
Uncertainty in plain language · The difference between shows and suggests · A complete worked case study, start to finish
Encoding data in ink so a reader sees the truth quickly — chart choice, labeling, color, and the failure modes to avoid.
Syllabus · 4 units · ~14 hours
Unit I — Encodings
Position, length, angle, area, color: a ranked toolbox · Why bar charts beat pie charts for comparison · Scales and axes as promises to the reader
Unit II — Choosing the Chart
One variable: distributions · Two variables: relationships and comparison · Change over time · Small multiples instead of overloaded single charts
Unit III — The Craft
Titles that state the finding · Direct labeling over legends where possible · Color for meaning, not decoration · Annotation: guiding the reader's eye honestly
Unit IV — Failure Modes
Truncated and dual axes · Overplotting and how to relieve it · Chart junk and the data-ink ratio · Redesigning a bad chart: a worked critique
Dataframes, cleaning, grouping, and scripted charts — the daily craft of analysis, written down so it can be rerun and checked.
Syllabus · 5 units · ~36 hours
Unit I — Python for People with Data
Values, lists, and dictionaries · Reading a CSV into a dataframe · Selecting rows and columns · The notebook as a lab journal
Unit II — Transforming Tables
Filtering and sorting · New columns from old · Grouping and aggregation · Joining tables and the perils of the merge
Unit III — Cleaning
Missing values: find, count, decide · Inconsistent categories and stray whitespace · Dates and times, tamed · Outliers: investigate before deleting
Unit IV — Charts in Code
Scripted plots that regenerate with the data · Distributions, scatterplots, and time series · Styling for legibility
Unit V — A Complete Analysis
From raw file to findings in one reproducible script · Structuring a small project · Writing up results with the code that made them
Prediction as disciplined generalization — regression, classification, validation, and the constant war against overfitting.
Syllabus · 5 units · ~44 hours
Unit I — Learning from Examples
Features, labels, and the prediction problem · Training versus test data: the cardinal separation · Loss functions: saying what wrong costs · A nearest-neighbor model as a first predictor
Unit II — Regression
Linear regression by least squares · Interpreting coefficients, cautiously · Polynomial features and the flexibility dial · Bias and variance: the central trade
Unit III — Classification
Logistic regression and predicted probabilities · Decision thresholds and their consequences · Accuracy, precision, recall — and when each misleads · Class imbalance
Unit IV — Keeping Models Honest
Overfitting diagnosed · Cross-validation · Regularization: ridge and lasso · Leakage: the subtle way results lie
Unit V — Trees & Ensembles
Decision trees and their appetites · Random forests · Gradient boosting in outline · Explaining a model's prediction to a skeptic
The obligations that come with other people's data — consent, privacy, bias, and analyses that can be checked by strangers.
Syllabus · 4 units · ~12 hours
Unit I — Provenance & Consent
Where the data came from and who agreed to what · Terms of collection versus purposes of use · Scraping, licensing, and the public-but-personal problem
Unit II — Privacy
Anonymization and how re-identification defeats it · Aggregation as protection · Differential privacy, in plain terms
Unit III — Bias
Sampling bias: who is missing from the data · Historical bias: models that learn old injustices · Measuring fairness, and the trade-offs among definitions · Auditing a model's errors by group
Unit IV — Accountability
Reproducibility: analyses a stranger can rerun · Documenting data and decisions · Case studies: recidivism scores, hiring filters, health algorithms · Declining an analysis: when the right answer is no