maedoc/autotvb/concise-code

Concise Code for TVB Notebooks

Rules

1. One figure per scientific claim. If the goal asks for "PSD and FC", make exactly two figure blocks — not five.
2. No diagnostic prints. Remove print(x.shape), print(type(y)), print('Step done') from final notebooks. These are debugging artifacts.
3. No redundant plots. If raw and annotated heatmaps show the same data, keep only the annotated one.
4. Consolidate subplots. Use fig, axes = plt.subplots(1, N, ...) for side-by-side comparisons instead of N separate plt.figure() blocks.
5. No summary cells that repeat earlier output. A final markdown cell with 2–3 sentence takeaway is fine; do not re-print arrays or re-plot data.
6. Avoid boilerplate re-configuration. If Connectivity.from_file() + .speed = 4.0 + configure() appears once, don't repeat it in a later cell. Store the object and reuse.

Code patterns

Redundant (bad):

plt.figure(); plt.imshow(fc); plt.colorbar(); plt.title("FC")
plt.figure(); plt.imshow(sc); plt.colorbar(); plt.title("SC")  
plt.figure(); plt.imshow(fc - sc); plt.colorbar(); plt.title("Diff")

Concise (good):

fig, axes = plt.subplots(1, 3, figsize=(15, 4))
for ax, mat, title in zip(axes, [fc, sc, fc-sc], ["FC", "SC", "FC−SC"]):
    im = ax.imshow(mat, cmap="viridis"); ax.set_title(title)
    plt.colorbar(im, ax=ax)

Token budget

Every code cell costs tokens in the prompt. A 20-cell notebook with 3 redundant sections costs ~20% more tokens for zero scientific gain. Target ≤12 code cells for standard goals, ≤16 for complex multi-requirement goals.