# Design a simulation or machine learning experiment for my research

An experiment plan with research questions, data and simulation setup, baselines, ablations, metrics, and reproducibility checks.

- Topic: Research & Summaries
- Framework: Few-Shot
- Level: Standard
- Tags: Research, Data & Analytics, Fill-in Template
- Source: https://www.thinkwithprompt.com/prompts/design-a-simulation-or-machine-learning-experiment-for-my-research

## Prompt

```text
Help me design a simulation or machine learning experiment for my research.

**Research question:** [WHAT I want to show or test]
**Field and problem:** [e.g. "underwater acoustic target classification", "traffic flow", "protein structure"]
**Data:** [real datasets, simulated data, or both - and how simulated data is generated, e.g. physics model, parameter ranges]
**Methods I'm considering:** [e.g. CNN on spectrograms, transformer, classical baseline]
**Tools:** [e.g. PyTorch, MATLAB, Python simulators]
**Compute and time limits:** [DETAILS]
**Where I'm stuck:** [OPTIONAL]

Produce an experiment plan:
1. **Hypotheses** - specific, testable claims and what result would count against each.
2. **Data and simulation setup** - parameters to vary, ranges, noise and realism conditions, train/validation/test splits that avoid leakage (e.g. splitting by recording session or environment, not random samples), and how to check simulated data resembles real data.
3. **Baselines** - simple and established methods to compare against.
4. **Ablations and sensitivity tests** - what to remove or vary to show why the method works.
5. **Metrics** - primary and secondary, and why they fit the question.
6. **Statistics and reproducibility** - multiple random seeds, confidence intervals, fixed configs, versioned data and code.
7. **Experiment matrix** - a table of runs, with a priority order if compute is limited.
8. **Threats to validity** - sim-to-real gap, overfitting to the simulator, and other limitations to report.

Base advice on established methodology. Don't invent citations - if you suggest a paper or method, give its name so I can look it up, and mark anything uncertain [VERIFY].
```

## What to fill in

- WHAT I want to show or test
- e.g. CNN on spectrograms, transformer, classical baseline
- e.g. PyTorch, MATLAB, Python simulators
- DETAILS
- OPTIONAL

Free to copy and use. From ThinkWithPrompt (https://www.thinkwithprompt.com/), a library of 2,663 curated AI prompts.