主题
可视化对比图表
可视化策略
模型评估的可视化需要传达 "谁更好？好多少？在哪些方面？" 这三个核心问题。我们用 Pandas 内置绑图 + Matplotlib 扩展来实现。
模型总体对比雷达图
python
import matplotlib.pyplot as plt
import matplotlib
matplotlib.rcParams['font.sans-serif'] = ['PingFang SC', 'Arial Unicode MS']
matplotlib.rcParams['axes.unicode_minus'] = False
def plot_radar_comparison(engine: ModelBenchmarkEngine):
basic = engine.compute_basic_metrics().set_index("model")
categories = ["平均分", "稳定性(CV)", "P10(鲁棒性)",
"P90(顶尖)", "胜率%", "性价比"]
def normalize(series, higher_better=True):
if higher_better:
return (series - series.min()) / (series.max() - series.min() + 1e-10)
else:
return (series.max() - series) / (series.max() - series.min() + 1e-10)
metrics_data = pd.DataFrame({
"平均分": normalize(basic["mean"]),
"稳定性(CV)": normalize(basic["cv"], higher_better=False),
"P10(鲁棒性)": normalize(basic["p10"]),
"P90(顶尖)": normalize(basic["p90"]),
})
win_rates = engine.compute_win_rate().set_index("model")
eff = engine.compute_efficiency_metrics().set_index("model")
metrics_data["胜率%"] = normalize(win_rates["win_rate"])
metrics_data["性价比"] = normalize(eff["score_per_dollar"])
angles = np.linspace(0, 2 * np.pi, len(categories), endpoint=False).tolist()
angles += angles[:1]
colors = ["#2196F3", "#FF5722", "#4CAF50", "#9C27B0", "#FF9800"]
fig, ax = plt.subplots(figsize=(8, 8), subplot_kw=dict(polar=True))
for i, model in enumerate(metrics_data.index):
values = metrics_data.loc[model].tolist()
values += values[:1]
ax.plot(angles, values, "o-", linewidth=2,
label=model, color=colors[i % len(colors)])
ax.fill(angles, values, alpha=0.15, color=colors[i % len(colors)])
ax.set_xticks(angles[:-1])
ax.set_xticklabels(categories, fontsize=11)
ax.set_ylim(0, 1.05)
ax.legend(loc="upper right", bbox_to_anchor=(1.3, 1.0))
ax.set_title("多维度模型能力雷达图", fontsize=14, fontweight="bold", pad=20)
plt.tight_layout()
return fig
fig_radar = plot_radar_comparison(engine)
plt.show()输出效果：一个五边形雷达图，每个顶点代表一个评估维度，5 个模型用不同颜色的多边形覆盖，面积越大表示综合能力越强。
分组柱状图：按任务类型对比
python
def plot_task_bar_chart(per_task_df: pd.DataFrame):
score_cols = [c for c in per_task_df.columns
if c not in ["best_model", "gap_best_second"]]
tasks = per_task_df.index.tolist()
x = np.arange(len(tasks))
width = 0.15
n_models = len(score_cols)
colors = ["#1f77b4","#ff7f0e","#2ca02c","#d62728","#9467bd"]
fig, ax = plt.subplots(figsize=(14, 7))
for i, col in enumerate(score_cols):
offset = (i - n_models / 2 + 0.5) * width
bars = ax.bar(x + offset, per_task_df[col], width,
label=col, color=colors[i % len(colors)], alpha=0.85)
for bar in bars:
height = bar.get_height()
ax.annotate(f"{height:.2f}",
xy=(bar.get_x() + bar.get_width() / 2, height),
xytext=(0, 2), textcoords="offset points",
ha="center", va="bottom", fontsize=7)
ax.set_xlabel("任务类型", fontsize=12)
ax.set_ylabel("平均得分", fontsize=12)
ax.set_title("各模型在不同任务上的表现对比", fontsize=14, fontweight="bold")
ax.set_xticks(x)
ax.set_xticklabels(tasks, rotation=30, ha="right")
ax.legend(title="模型", loc="upper right")
ax.set_ylim(0, 1.05)
ax.grid(axis="y", alpha=0.3)
ax.axhline(y=per_task_df[score_cols].values.mean(), color="red",
linestyle="--", alpha=0.5, label="全局均值")
plt.tight_layout()
return fig
fig_bar = plot_task_bar_chart(per_task)
plt.show()效果：分组柱状图，X 轴为任务类型（reasoning/coding/math 等），每组内各模型的柱子并排显示，可以直观看出哪个模型在哪个任务上领先。
分布对比：箱线图与密度图
python
def plot_distribution_comparison(df: pd.DataFrame):
models = sorted(df["model"].unique())
fig, axes = plt.subplots(1, 2, figsize=(14, 5))
ax1 = axes[0]
data_by_model = [df[df["model"] == m]["score"] for m in models]
bp = ax1.boxplot(data_by_model, labels=models, patch_artist=True)
colors = ["#3498db", "#e74c3c", "#2ecc71", "#9b59b6", "#f39c12"]
for patch, color in zip(bp["boxes"], colors):
patch.set_facecolor(color)
patch.set_alpha(0.6)
ax1.set_title("分数分布箱线图", fontweight="bold")
ax1.set_ylabel("Score")
ax1.grid(axis="y", alpha=0.3)
for i, mdata in enumerate(data_by_model):
y = mdata
x = np.random.normal(i + 1, 0.04, size=len(y))
ax1.scatter(x, y, alpha=0.15, s=8, color=colors[i])
ax2 = axes[1]
for i, model in enumerate(models):
scores = df[df["model"] == model]["score"]
scores.plot.kde(ax=ax2, label=model, linewidth=2,
color=colors[i % len(colors)])
mean_val = scores.mean()
ax2.axvline(mean_val, color=colors[i % len(colors)],
linestyle=":", alpha=0.7, linewidth=1)
ax2.set_title("分数密度分布", fontweight="bold")
ax2.set_xlabel("Score")
ax2.set_ylabel("Density")
ax2.legend()
ax2.grid(alpha=0.3)
ax2.set_xlim(0, 1.05)
plt.tight_layout()
return fig
fig_dist = plot_distribution_comparison(df_std)
plt.show()效果：左图为箱线图叠加散点（展示分布形态和离群点），右图为核密度估计曲线（展示分数分布的形状差异）。
散点矩阵：两两相关性分析
python
def plot_scatter_matrix(wide_df: pd.DataFrame, score_cols: list):
n = len(score_cols)
fig, axes = plt.subplots(n, n, figsize=(3 * n + 2, 3 * n + 2))
for i in range(n):
for j in range(n):
ax = axes[i][j] if n > 1 else axes
if i == j:
data = wide_df[score_cols[j]]
ax.hist(data, bins=25, color="#3498db", alpha=0.7, edgecolor="white")
ax.axvline(data.mean(), color="red", linestyle="--", linewidth=1.5)
else:
ax.scatter(
wide_df[score_cols[j]], wide_df[score_cols[i]],
alpha=0.15, s=8, c="#2ecc71", edgecolors="none"
)
corr = wide_df[score_cols[j]].corr(wide_df[score_cols[i]])
ax.set_title(f"r={corr:.2f}", fontsize=8, pad=2)
if j == 0:
ax.set_ylabel(score_cols[i], fontsize=8, rotation=0,
ha="right", va="center")
if i == n - 1:
ax.set_xlabel(score_cols[j], fontsize=8)
ax.tick_params(labelsize=6)
fig.suptitle("模型间分数相关性散点矩阵", fontsize=13, fontweight="bold", y=1.01)
plt.tight_layout()
return fig
score_cols = [c for c in wide_df.columns
if c not in ("sample_id","task","best_model","best_score",
"worst_score","score_gap","score_std",
"diff_best_second","is_close_competition")]
fig_scatter = plot_scatter_matrix(wide_df, score_cols[:4])
plt.show()效果：N×N 的散点矩阵，对角线为直方图，非对角线为两个模型间的散点图+相关系数。高相关系数意味着两个模型表现一致。
热力图：任务×模型矩阵
python
def plot_heatmap_matrix(per_task_df: pd.DataFrame):
score_cols = [c for c in per_task_df.columns
if c not in ["best_model", "gap_best_second"]]
matrix = per_task_df[score_cols].astype(float)
fig, ax = plt.subplots(figsize=(max(6, len(score_cols)), max(5, len(matrix))))
im = ax.imshow(matrix.values, cmap="YlGnBu", aspect="auto", vmin=0.5, vmax=0.85)
ax.set_xticks(np.arange(len(score_cols)))
ax.set_yticks(np.arange(len(matrix)))
ax.set_xticklabels(score_cols, rotation=35, ha="right", fontsize=9)
ax.set_yticklabels(matrix.index, fontsize=9)
for i in range(len(matrix)):
for j in range(len(score_cols)):
val = matrix.iloc[i, j]
text_color = "white" if val < 0.62 or val > 0.78 else "black"
ax.text(j, i, f"{val:.2f}", ha="center", va="center",
color=text_color, fontsize=9, fontweight="bold")
cbar = ax.figure.colorbar(im, ax=ax, shrink=0.8)
cbar.ax.set_ylabel("Score", rotation=-90, va="bottom")
best_per_row = matrix.idxmax(axis=1)
for i, best_m in enumerate(best_per_row.items()):
j = score_cols.index(best_m[1])
rect = plt.Rectangle((j - 0.5, i - 0.5), 1, 1,
fill=False, edgecolor="red", linewidth=2.5)
ax.add_patch(rect)
ax.set_title("任务 × 模型 得分热力图 (红框=最优)", fontweight="bold", pad=15)
plt.tight_layout()
return fig
fig_heatmap = plot_heatmap_matrix(per_task)
plt.show()效果：热力图中颜色越深表示分数越高，红色方框标记每行的最高分（即该任务的最优模型），一目了然地看到每个任务的赢家。