plot

Compare original and translation side by side

🇺🇸

Original

English

🇨🇳

Chinese

Help the user create publication-quality figures.

帮助用户创建符合出版标准的图表。

Use

plt.tight_layout()

constrained_layout=True

to avoid clipping

使用

plt.tight_layout()

或

constrained_layout=True

避免内容被裁剪

Readable font sizes — axis labels ~12pt, tick labels ~10pt, legend ~10pt. Scale up for posters/slides.
Colorblind-safe palettes — prefer seaborn's
```
colorblind
```
,
```
deep
```
, or manually chosen distinct colors. Avoid red/green as sole differentiator.
Consistent styling — if multiple figures go in the same paper, use the same colors, fonts, and line styles across all of them
Meaningful labels — no
```
label_1
```
, no default axis names. Every axis labeled with units where applicable.
Legends that help — place legends where they don't occlude data. Use labels that match the paper's terminology.
Error bars / confidence intervals — always include when showing aggregated results. State what they represent (std, SEM, 95% CI).
No chartjunk — remove unnecessary gridlines, borders, and decoration. Less is more.

易读的字体大小 — 坐标轴标签约12pt，刻度标签约10pt，图例约10pt。若用于海报/幻灯片可适当放大。
色盲友好调色板 — 优先使用seaborn的
```
colorblind
```
、
```
deep
```
调色板，或手动选择区分度高的颜色。避免仅用红/绿作为区分标识。
风格一致性 — 若多个图表将用于同一篇论文，所有图表需使用相同的颜色、字体和线条样式
有意义的标签 — 避免使用
```
label_1
```
这类默认命名，所有坐标轴需标注清晰，必要时附上单位。
实用的图例 — 将图例放置在不遮挡数据的位置。使用与论文术语一致的标签。
误差线/置信区间 — 展示聚合结果时务必包含误差线/置信区间，并说明其代表的含义（标准差、标准误、95%置信区间）。
去除冗余元素 — 移除不必要的网格线、边框和装饰。少即是多。

Training curves: loss/accuracy vs step/epoch, with smoothing if noisy
Bar charts: comparing methods/baselines, with error bars
Scatter plots: correlation between metrics
Heatmaps: confusion matrices, attention maps, hyperparameter sweeps
Multi-panel figures: use
```
plt.subplots()
```
with shared axes where appropriate

Read existing plotting code first — match the style if figures already exist in the project
Save figures to a sensible path — e.g.
```
figures/
```
or
```
plots/
```
Use
fig, ax
API — not
```
plt.plot()
```
directly. This makes multi-panel and customization easier.

$ARGUMENTS