🤖 ChatGPT Prompt Collection

AI-Guided SIR Modeling Hackathon

This document contains curated ChatGPT prompts for the
AI-Guided SIR Modeling Hackathon.

It is designed to teach participants how to work with ChatGPT as a scientific coding co-pilot, not as a black box.

🧠 Why Prompt Structure Matters

In this hackathon, ChatGPT is used to:

Generate scientific Python code
Explain epidemiological concepts
Assist with debugging and inference
Help write clear summaries

Poorly structured prompts often lead to:

Wrong assumptions (units, data meaning)
Silent modeling errors
Overconfident but incorrect explanations

Good prompts dramatically reduce these risks.

⭐ Core Prompt Template (RECOMMENDED)

Use this template for all scientific prompts during the hackathon.

🔹 Core Prompt Template

Persona:
You are an expert epidemiological modeler and Python educator.

Context:
- Dataset: biweekly measles case counts from Niamey, Niger
- Columns: biweek, community, measles
- Data represent: observed case counts (incidence)
- Time unit: biweeks
- Infectious period ≈ 2 weeks (gamma ≈ 1 per biweek)

Task:
[Describe ONE specific task clearly and precisely]

Constraints:
- Use Python compatible with Google Colab
- Allowed libraries: numpy, scipy, pandas, matplotlib
- Keep code minimal, readable, and well commented
- Do not rewrite unrelated code

Verification:
- Include at least one plot
- Include at least one sanity check
- Explain assumptions clearly

Output style:
- Step-by-step explanation
- Code blocks clearly separated
- Avoid unnecessary verbosity

📌 Prompting Rules (Hackathon Ground Rules)

Always specify units (biweeks, not days)
Always state what the data represent (case counts vs model states)
One task per prompt
Always ask for plots
Always ask for sanity checks
Iterate in small steps
Treat ChatGPT as an assistant, not an authority

🧩 Task-Specific Prompts

All prompts below are instances of the Core Prompt Template.

🧪 Prompt 1 — Load and Explore the Data

Persona:
You are an expert epidemiological modeler and Python educator.

Context:
- Dataset contains biweekly measles case counts from Niamey, Niger
- Columns: biweek, community, measles

Task:
Write Python code to:
1. Load the CSV file
2. Plot measles cases over time for each community (A, B, C)
3. Briefly explain what biweekly case counts mean for modeling

Constraints:
- Use pandas and matplotlib
- Google Colab compatible

Verification:
- One time-series plot with labeled axes and legend

🧬 Prompt 2 — Implement the SIR Model (Biweekly Units)

Persona:
You are an expert epidemiological modeler.

Context:
- Closed SIR model
- Time unit: biweeks
- Infectious period ≈ 2 weeks (gamma ≈ 1 per biweek)

Task:
Implement an SIR model using SciPy ODE solving and plot S, I, R.

Constraints:
- Use scipy.integrate
- Comment each equation

Verification:
- Plot S, I, R
- Verify S + I + R ≈ N at all times

📈 Prompt 3 — Feature-Based R₀ Estimation (Early Growth)

Persona:
You are an expert in infectious disease modeling.

Context:
- Community A measles outbreak
- Biweekly case counts
- Early outbreak phase shows exponential growth

Task:
Estimate early exponential growth by:
1. Selecting an early biweek window
2. Fitting log(measles) vs time
3. Plotting the semi-log data and fitted line
4. Converting slope to R0 assuming gamma = 1 per biweek

Constraints:
- Use numpy or scipy
- Explain assumptions clearly

Verification:
- Semi-log plot
- Printed slope and R0 estimate

🔁 Prompt 4 — Incidence Mapping via Accumulator H(t)

Persona:
You are an expert epidemic modeler.

Context:
- Observed data are biweekly case counts (incidence)
- SIR model outputs state variables

Task:
Extend the SIR model by:
- Adding an accumulator H(t) with dH/dt = beta * S * I / N
- Computing predicted biweekly cases as ΔH
- Aligning ΔH with observed biweeks

Constraints:
- Keep implementation minimal and readable

Verification:
- Plot ΔH over time
- Explain why ΔH matches the data structure

📉 Prompt 5 — Least Squares Parameter Fitting

Persona:
You are an expert in numerical optimization.

Context:
- Observed data: biweekly case counts
- Model output: predicted ΔH

Task:
Fit beta (and optionally initial conditions) using least squares.

Constraints:
- Use scipy.optimize
- Use sensible initial values from early-growth estimates

Verification:
- Overlay plot of observed vs fitted cases
- Print best-fit parameter values

🎯 Prompt 6 — Poisson Likelihood Inference

Persona:
You are an expert in statistical inference for epidemic models.

Context:
- Observed data are count data
- Poisson noise is appropriate
- μ_t = rho * ΔH_t (rho = reporting probability)

Task:
Implement maximum likelihood estimation assuming:
y_t ~ Poisson(μ_t)

Constraints:
- Use parameter transforms (log for beta, logit for rho)
- Explain why transforms are needed

Verification:
- Plot observed vs expected counts
- Print estimated parameters

📊 Prompt 7 — Profile Likelihood (Uncertainty)

Persona:
You are an expert in likelihood-based inference.

Context:
- Poisson likelihood already implemented

Task:
Create a profile likelihood for beta by:
1. Fixing beta over a grid
2. Re-optimizing remaining parameters
3. Plotting profile log-likelihood
4. Indicating an approximate 95% confidence interval

Constraints:
- Keep computation lightweight

Verification:
- Profile likelihood plot
- Explanation of uncertainty interpretation

🧾 Prompt 8 — Generate a Markdown Summary

Persona:
You are preparing a short scientific report for hackathon participants.

Task:
Write a Markdown-formatted summary covering:
- Model description
- Parameter estimates and R0
- Comparison of least squares vs Poisson inference
- Interpretation of uncertainty
- Suggested next steps (SEIR, time-varying beta)

Constraints:
- Scientific but accessible language
- Use Markdown headers and bullet points

Verification:
- Clean Markdown output suitable for a notebook cell

🚫 Bad Prompt vs ✅ Good Prompt

Learning How to Talk to ChatGPT

Example 1 — Missing Context & Units

🚫 Bad Prompt

Help me fit an SIR model to this data.

Problems

No units
No data meaning
Incorrect assumptions likely

✅ Good Prompt

Persona:
You are an expert epidemiological modeler and Python educator.

Context:
- Dataset: biweekly measles case counts from Niamey, Niger
- Data represent incidence
- Time unit: biweeks

Task:
Fit an SIR model using Poisson likelihood.

Verification:
- Plot observed vs fitted cases

Example 2 — Too Many Tasks at Once

🚫 Bad Prompt

Build an SIR model, fit it, explain R0, and say if the outbreak is controlled.

✅ Good Prompt

Task:
Implement and simulate the SIR model only.

➡️ Follow with a second prompt for inference.

Example 3 — No Verification

🚫 Bad Prompt

Estimate beta and gamma.

✅ Good Prompt

Task:
Estimate beta using Poisson likelihood.

Verification:
- Plot observed vs expected counts
- Print log-likelihood

Example 4 — Treating ChatGPT as an Oracle

🚫 Bad Prompt

Is this model correct?

✅ Good Prompt

Task:
Critically review the model assumptions and list weaknesses.

🧠 Final Takeaways

ChatGPT is a co-pilot, not a decision-maker
You control:
- Assumptions
- Units
- Validation
Good prompts → good science
Bad prompts → confident nonsense

Happy hacking! 🚀