charts

Files (3)

• SKILL.md
• check_palette.py · scripts/
• make_chart.py · scripts/

SKILL.md (first lines)

## Which engine am I in?

Read the first line of the RunPython tool description:

- **PYODIDE** — Python in WebAssembly. Pure-Python packages install on import;
  no subprocess, no system binaries.
- **LOCAL SUBPROCESS** — the user's real Python. Everything works, including
  system tools — check with `shutil.which(...)` before relying on one.
- **PODMAN CONTAINER** — isolated container with python3, nodejs and tectonic.

Files persist between RunPython calls in the same chat, so build in steps and
inspect as you go rather than writing one long script.

matplotlib works in every engine. Save to PNG; the file persists in the working
directory and is attached to the conversation automatically, so it can be
embedded into a .docx, .pptx or .pdf in a later call.

## Rendering from a CSV — use the script

```
python scripts/make_chart.py data.csv --x region --y revenue --kind bar -o out.png
python scripts/make_chart.py data.csv --x month --y sales --kind line --keep-order
```

It applies the rules below instead of leaving them to be remembered: bar axes
start at zero, bars are sorted by value (`--keep-order` when the order carries
meaning, e.g. months), values are labelled directly on the bars rather than via
a legend, spines are trimmed, duplicate categories are summed, non-numeric rows
are reported rather than dropped in silence, and the font has Cyrillic glyphs.

Then LOOK at the file. Overlapping labels are only visible to the eye.

## Baseline (writing it yourself)

```python
import matplotlib
matplotlib.use('Agg')          # no display in any of these engines
import matplotlib.pyplot as plt

fig, ax = plt.subplots(figsize=(8, 4.5), dpi=150)
ax.bar(regions, revenue, color='#4472C4')
ax.set_title('Revenue by region')
ax.set_ylabel('₽')
ax.spines[['top', 'right']].set_visible(False)
ax.grid(axis='y', alpha=.3)
fig.tight_layout()
fig.savefig('revenue.png')
```

`matplotlib.use('Agg')` before importing pyplot — without it, the backend can
try to open a window and fail.

## Choosing the form

Comparing categories → bar, sorted by value unless the order carries meaning
(months stay in month order). Change over time → line. Part of a whole → bar,
not pie; a pie with more than four slices is unreadable. Relationship between
two variables → scatter. Distribution → histogram.

## Rules that decide whether it is readable

Start a bar axis at zero. A truncated bar axis misrepresents the difference —
that is the single most common misleading chart.

Label the axes with units. A number without a unit is not information.

Do not encode meaning only in colour — a colour-blind reader loses it. Order,
labels or patterns should carry the message too.

Sort bars by value. An unsorted bar chart makes the reader do the ranking.

Direct-label the important series instead of relying on a legend the eye has to
travel to.

## Check the palette — you cannot eyeball this

```
python scripts/check_palette.py "#4472C4" "#ED7D31" "#A5A5A5"
python scripts/check_palette.py --tab10 --bg "#1a1a1a"
```

Reports contrast against the background (below 3:1 a bar is hard to see) and
simulates deuteranopia, protanopia and tritanopia to catch series that collapse
into one colour. Around 8% of men cannot separate red from green: a palette that
looks varied to you can carry no information at all for them.

If it flags a pair, do not just swap hues — vary LIGHTNESS as well, direct-label
the series, or add markers/patterns. Hue alone is never a safe channel.

## Cyrillic labels

Set a font that has the glyphs, or every label renders as boxes:

```python
matplotlib.rcParams['font.family'] = 'DejaVu Sans'
```

## Verify

Reopen the file and confirm it is a real image, not a 0-byte stub:

```python
import os
from PIL import Image
assert os.path.getsize('revenue.png') > 5000
print(Image.open('revenue.png').size)
```

Then look at it. If the labels overlap or the bars are invisible, fix it before
delivering — a chart you never viewed is a guess.

From iaa2005/monet-directory, MIT.

More from iaa2005/monet-directory