Skip to content

joshua-zyy/academic-paper-writer / skills/academic-figure

academic-figure

Verified

Create, revise, or audit academic data/result figures for CS/AI/ML papers. Data/result plots default to Python-generated editable SVG with CS/AI/ML-specific design rules for benchmarks, ablations, training dynamics, robustness, diagnostics, distributions, confusion matrices, and efficiency tradeoffs. Use when: generating plots from experiment results or numeric data, auditing publication figures, suggesting data-driven figure types, revising chart colors/layouts/labels, or preparing figure QA reports. Model framework diagrams, architecture diagrams, overview diagrams, and complex mechanism schematics are outside this skill's automatic drawing scope; provide only manual figure requirements or caption/blueprint notes when needed. Triggers on: 绘图, figure, chart, 画图, 实验图, 训练曲线, 消融实验, 对比图, 混淆矩阵, 结果图, 性能图, 鲁棒性图, 效率图, plot, publication figure, 数据可视化, generate plot, figure blueprint, 建议图表类型, figure audit, 审查图表, figure revision, 修改图表.

references/tutorials.md

View on GitHub
# Tutorials: 端到端示例教程

## 教程 1:训练曲线(chart-from-data)

### 用户输入
"我有三个方法在 ImageNet 上的训练 loss 数据,帮我画出训练曲线对比图。"

### 数据文件(train_loss.csv)
```csv
epoch,ours_mean,ours_std,baseline_a_mean,baseline_a_std,baseline_b_mean,baseline_b_std
1,2.85,0.12,3.12,0.15,3.45,0.18
5,1.72,0.08,2.01,0.10,2.33,0.12
10,0.95,0.05,1.22,0.07,1.51,0.09
15,0.62,0.04,0.85,0.06,1.08,0.07
20,0.45,0.03,0.62,0.05,0.82,0.06
30,0.32,0.02,0.48,0.04,0.61,0.05
40,0.28,0.02,0.42,0.03,0.55,0.04
50,0.25,0.02,0.39,0.03,0.51,0.04
```

### Figure Contract
```
Core Conclusion: 本方法的收敛速度更快且最终 loss 更低
Archetype: training-curve
Panels: 1 panel (全部方法在同一张图对比)
```

### 生成代码
```python
import pandas as pd
from api import make_training_curve, finalize_figure, PALETTE

data = pd.read_csv("train_loss.csv")

fig, ax = make_training_curve(
    data,
    x_col="epoch",
    y_cols=["ours_mean", "baseline_a_mean", "baseline_b_mean"],
    std_cols=["ours_std", "baseline_a_std", "baseline_b_std"],
    labels=["Ours", "Baseline A", "Baseline B"],
    colors=[PALETTE[0], PALETTE[2], PALETTE[3]],
    xlabel="Epoch",
    ylabel="Training Loss",
    figsize=(4, 3),
)

finalize_figure(fig, "training_curve")
```

### 输出
- training_curve.svg
- training_curve.pdf
- training_curve.tiff
- train_loss.csv(源数据)

---

## 教程 2:消融实验(chart-from-data)

### 用户输入
"我们做了消融实验,验证注意力机制和残差连接的贡献,帮我画柱状图。"

### 数据
```csv
variant,accuracy,std
Full Model,92.5,1.2
w/o Attention,88.3,1.5
w/o Residual,87.1,1.4
w/o Both,83.6,1.8
```

### 生成代码
```python
import pandas as pd
from api import make_ablation, finalize_figure

data = pd.read_csv("ablation.csv")

fig, ax = make_ablation(
    baseline_label="Full Model",
    ablation_cols=["w/o Attention", "w/o Residual", "w/o Both"],
    values=data["accuracy"].tolist(),
    errors=data["std"].tolist(),
    ylabel="Accuracy (%)",
    figsize=(4, 3.5),
)

finalize_figure(fig, "ablation_study")
```

---

## 教程 3:混淆矩阵(chart-from-data)

### 用户输入
"我有个 4 类分类的结果,画混淆矩阵。"

### 生成代码
```python
import numpy as np
from api import make_heatmap, finalize_figure

cm = np.array([[920, 35, 25, 20],
               [30, 890, 45, 35],
               [20, 40, 910, 30],
               [15, 25, 35, 925]])

classes = ["Class A", "Class B", "Class C", "Class D"]

fig, ax = make_heatmap(
    matrix=cm,
    xticklabels=classes,
    yticklabels=classes,
    annot=True,
    fmt="d",
    figsize=(4, 3.5),
    cbar_label="Count",
)

finalize_figure(fig, "confusion_matrix")
```

---

## Tutorial 4: Multi-panel 实验图(chart-from-data)

### 用户输入
"我们需要一个 2-panel 图,左边是训练曲线,右边是最后的性能对比柱状图。"

### Figure Contract
```
Core Conclusion: 本方法收敛更快且最终性能最优
Archetype: multi-panel (2 panels)
Panel (a): 训练曲线(ours vs 2 baselines)
Panel (b): 最终性能对比柱状图(3 方法 × 3 数据集)
```

### 生成代码框架
```python
import pandas as pd
import matplotlib.pyplot as plt
import seaborn as sns
import numpy as np
from api import apply_pub_style, finalize_figure, PALETTE

train_data = pd.read_csv("train_loss.csv")
perf_data = pd.read_csv("performance.csv")

fig = plt.figure(figsize=(7.5, 3))
gs = fig.add_gridspec(1, 2, width_ratios=[1, 1.2], wspace=0.4)

# Panel (a): 训练曲线(手动在 ax1 上绘制)
ax1 = fig.add_subplot(gs[0])
colors = [PALETTE[0], PALETTE[2], PALETTE[3]]
labels = ["Ours", "Baseline A", "Baseline B"]
for i, (col, std_col) in enumerate(zip(
    ["ours_mean", "baseline_a_mean", "baseline_b_mean"],
    ["ours_std", "baseline_a_std", "baseline_b_std"],
)):
    x = train_data["epoch"].values
    y = train_data[col].values
    s = train_data[std_col].values
    ax1.plot(x, y, color=colors[i], linewidth=1.2, label=labels[i])
    ax1.fill_between(x, y - s, y + s, color=colors[i], alpha=0.15, linewidth=0)
ax1.set_title("(a) Training Convergence", fontsize=9, fontweight="bold", loc="left")
ax1.set_xlabel("Epoch")
ax1.set_ylabel("Training Loss")
ax1.legend()

# Panel (b): 性能对比柱状图(sns 直接绘制)
ax2 = fig.add_subplot(gs[1])
sns.barplot(data=perf_data, x="dataset", y="score", hue="method",
            palette=[PALETTE[0], PALETTE[2], PALETTE[3]], ax=ax2,
            edgecolor="black", linewidth=0.5)
ax2.set_title("(b) Performance Comparison", fontsize=9, fontweight="bold", loc="left")
ax2.set_ylabel("Accuracy (%)")

apply_pub_style()
finalize_figure(fig, "multi_panel_experiment")
```
rea-analysis-optimizationmorluto/reaOptimize REA providers and shared analysis workflows for agent context, latency, CPU, and memory while preserving correctness and Evidence. Use for analysis bottlenecks and cache or query design; CI/test-suite optimization belongs to rea-ci-tests-audit.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
rea-ci-tests-auditmorluto/reaAudit REA CI and test suites for correctness, fixture drift, false passes, and measured performance; use for suite or workflow investigations and efficiency changes, rather than ordinary feature testing or a single test failure.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
rea-tool-designmorluto/reaAudit, investigate, design, or change REA investigation tools, CLI/MCP contracts, provider capabilities, and Evidence semantics.Stars: 472kUpdated Claude CodeCodexVerifiedInstall
ponytail-auditDietrichGebert/ponytailQuality audit of the whole repo: bugs, security, real load, missing tests, speed, and what to delete. Most important first.Stars: 194kUpdated Claude CodeCodexVerifiedInstall