Most orgs collect surveys but stop at pretty charts. Decisions need tested claims and interpretable models: “Does stress at work predict care-seeking?” “Which factors most move satisfaction?” This session shows a repeatable path from messy responses to evidence you can present with confidence—fast enough for busy analysts and PMs, rigorous enough for research reviews.

✅ No prior coding required—start at true zero

🤖 AI-assisted coding—prompt → runnable code

🧩 Guided prompts—what to ask AI and why

🧪 Human-behavior datasets—clean synthetic core + optional real-survey snapshot

🛠️ Reusable templates—notebooks you can keep

Work with a real-world-style survey (Sleep Health & Lifestyle) to turn messy data into decisions—without a coding background. You’ll guide AI to draft clean R/Python notebooks, test claims correctly, and build practical models (logistic + decision tree). We focus on evaluation, thresholds, fairness, and clear reporting—so you leave with a 5-slide Decision Brief and a reproducible notebook.

Objective: build a repeatable, AI-assisted, evidence-based workflow for human-behavior data—prompting and verifying code, diagnosing errors, and communicating results clearly.

(Optional Week-1 add-on: a tiny OSMI snapshot for real-survey “reality check.”)

📚 What we’ll cover (from messy to actionable)

📦 Week 1 — Prep & Theory (Async + Office Hours)

Light, self-paced foundations + two live setup clinics so everyone is bootcamp-ready.

🧩 Module A — Environment & Preflight

Use Google Colab (Python) and RStudio on Posit Cloud (R); organize project folders

Starter notebook tour and ✅ Readiness check (imports, data access via Drive / Posit Cloud)

📊 Module B — Quantifying One Variable

Center (mean/median), spread (SD/IQR), shape (skew/kurtosis), percentiles

Pick the right plot: histogram/box/violin/bar

Missingness & data types; tidy columns

🧭 Module C — Relationships & Correlation

Association ≠ causation; Simpson’s paradox (quick)

Pearson vs Spearman; nonlinearity/outliers

Right plots; stratify by BMI/Stress tier

🧪 Module D — Hypothesis Tests & Effects

Choose χ², t-test, or ANOVA (assumptions)

Effect sizes (risk diff, odds ratio) + CIs

“Claim → test → effect → limitation” memo

🧠 Module E — Logistic Regression (concepts)

Intuition: log-odds → probabilities (S-curve)

Target = HasSleepCondition; leakage pitfalls; one-hot BMI/Gender

Interpret coefficients as odds ratios; ROC/AUC, PR, calibration

🌳 Module F — Decision Trees (concepts)

Splits/impurity; depth caps; overfitting

Interpreting path rules; when trees vs logistic

Delivery format: Live coding in Python on Google Colab. R provided as recorded RStudio tutorials with full project files and a Posit Cloud fallback (RStudio in your browser—no install, just click the link and make a copy). We will use ChatGPT; other assistants like Copilot, Codeium, or Tabnine are also welcome. Please verify outputs in Colab.

📌 Week-1 Deliverables (due Fri):

Preflight pass (screenshot)

EDA one-pager: 2 plots + 3 insights

Claim & Check memo (≤150 words)

(Optional): first logistic + decision tree run using provided cells)

🛠️ Live Office Hours (setup clinics):

📅 12:00–1:00 PM (PST): Setup & environment troubleshooting

📅 12:00–1:00 PM (PST): Preflight, plotting help, claim selection

🧑‍💻 Week 2 — Weekend Bootcamp (Live, 2 days)

From one-variable to two-variable to baseline models → thresholded decision.

📅 Day 1 (Sat)

AM (9:00–12:00 PST) — 🔍 One-Variable → Two-Variable

Finalize EDA + tested claim; stratified visuals; peer feedback

PM (1:00–3:00 PST) — 🧮 Logistic Baseline

Train/validation split; odds-ratio table; ROC/AUC; quick calibration

Save mini 📇 model card (data, target, metric, assumptions, risks)

📅 Day 2 (Sun)

AM (9:00–12:00 PST) — 🌲 Decision Tree + Comparison

Depth-capped tree (≤4), path interpretation; compare to logistic

Error analysis & subgroup checks (Gender, Age bands, BMI)

PM (1:00–3:00 PST) — 🎛️ Thresholds → 🧾 Decision Brief

Tune threshold (PR/ROC) → action rule (e.g., sleep-hygiene nudge)

Build & present your 5-slide Decision Brief:

Problem & data

Key chart (from Week 1)

Model results (logistic + tree)

Decision rule & trade-offs

Risks & next steps

🏁 Final Deliverables: two-model baseline + Decision Brief + reproducible notebook

🧰 What you will practice

Data: Sleep Duration, Quality of Sleep, Stress Level, Physical Activity, BMI Category, Systolic/Diastolic BP (parsed), Heart Rate, Daily Steps, Sleep Disorder (+ derived flags/tiers).

Languages: R and Python (parallel materials).

Skills: disciplined EDA; correct test selection; odds-ratio interpretation; model evaluation (ROC/AUC, PR, calibration); error analysis; thresholding; fairness & leakage checks; decision reporting.

🧳 What you will leave with

A methodical workflow: question → data → model → evaluation → decision

A concise portfolio on the Sleep dataset: EDA one-pager, Claim & Check, two-model baseline, and a Decision Brief

The confidence to state assumptions, avoid leakage, and defend recommendations

🎁 What’s included

🎥 Live sessions: interactive instruction with Omid

🛟 Office hours: two Week-1 setup clinics + optional follow-up hour

📼 Lifetime access: recordings & materials

🤝 Community: peer feedback & shared project space

📜 Certificate of completion

✅ Maven Guarantee: full refund up to the halfway point