9. Notebooks and figures¶
← Reproducibility · Contents · Next: Skills →
Notebooks are useful for exploring data and explaining an analysis. The main questions are whether you can review changes and rerun the work from a fresh session.
If you already use Jupyter¶
You do not need to change tools to complete this book. Restart the kernel and run all cells in order before trusting saved output. Move repeated calculations into functions that you can test.
A notebook file contains code, metadata and often embedded outputs. Ask the agent to use a notebook-aware reader or extract the relevant cells, rather than dumping the JSON into context. Inspect the code changes and newly generated outputs. Tools such as Jupytext can pair a notebook with a plain Python file for easier Git review.
Text-based options¶
| Tool | When it may help |
|---|---|
| marimo | Python notebooks stored as .py files, with reactive cell execution. |
Quarto (.qmd) |
Reports combining prose, code and figures. |
| Jupytext | A text representation alongside an existing Jupyter notebook. |
These are optional choices. Keep the project's existing format unless changing it solves a specific problem.
For results you will reuse, put the calculation in tested functions and provide a script or documented notebook execution command. A reader should be able to reproduce the analysis without guessing the order of your interactive work.
Use plots to check the analysis¶
Ask for diagnostics that answer a question:
- participant-level distributions: is an aggregate hiding an unusual case?
- observations with fitted curves: where does the model fit poorly?
- values before and after transformation: did the units or scale change as expected?
- residuals, where appropriate: is there structure the model does not capture?
Save useful diagnostics with the run record. Temporary plots can go in an ignored scratch directory.
Publication figures¶
Specify units, the quantity plotted, what intervals represent, and whether points represent trials or participants. Check labels, colour accessibility and text at the final printed size. Use axis limits that support an honest comparison and clearly indicate any truncation.
An agent can help write and revise the plotting code. Inspect the rendered figure yourself and compare the plotted values with the calculation that produced them. For raincloud plots, the lab's RainCloudPlots is an existing implementation to consider.
Try it¶
For the first exercise, plot both participant means and the group mean. Ask a lab mate to identify the unit represented by each mark without reading your code.