Install any skill in seconds. Free to start, no credit card required.
Get Started Free →从带单位的字符串列中提取数值并清洗,生成包含直方图、饼图、条形图和累积分布图的多维度综合分布可视化图表,用于展示数据的集中趋势与分布特征。
.claude/skills/opensensenova-numeric-extraction-and-distribution-analysis/SKILL.md| Test case | Without → With | Effect | Δ tokens | Δ turns |
|---|---|---|---|---|
| case-03 | ✗→✓ | ▲ Improved | 3% | 0% |
| case-05 | ✗→✓ | ▲ Improved | 6% | 0% |
| case-08 | ✗→✓ | ▲ Improved | 14% | 0% |
| case-12 | ✗→✓ | ▲ Improved | 22% | 0% |
| case-14 | ✗→✓ | ▲ Improved | -11% | 0% |
Step1 从原始数据中提取目标列,清理无效和空值数据,并安全地将带单位的字符串转换为数值类型
pythonimport pandas as pd import numpy as np import matplotlib.pyplot as plt # 配置中英文字体,避免图表乱码 plt.rcParams['font.sans-serif'] = ['SimHei', 'DejaVu Sans'] plt.rcParams['axes.unicode_minus'] = False item_col = '项目名称' # 占位示例:分类或名称列 value_col = '带单位的数值' # 占位示例:需要提取数值的原始列 numeric_col = '提取数值' unit_str = 'g' # 占位示例:需要移除的单位字符串 def extract_numeric_value(val_str): """从带单位的字符串中提取数值""" if pd.isna(val_str): return None try: # 移除单位并转换为浮点数 return float(str(val_str).replace(unit_str, '').strip()) except ValueError: return None # 清理缺失值与异常占位符 df_clean = df.dropna(subset=[item_col, value_col]).copy() df_clean = df_clean[df_clean[item_col] != '...'] # 应用提取函数并过滤转换失败的行 df_clean[numeric_col] = df_clean[value_col].apply(extract_numeric_value) df_clean = df_clean.dropna(subset=[numeric_col])
Step2 创建基础分布直方图,并添加平均值和中位数的参考线以展示数据的集中趋势
pythonplt.figure(figsize=(12, 8)) # 绘制直方图 plt.hist(df_clean[numeric_col], bins=10, alpha=0.7, color='skyblue', edgecolor='black') # 计算并添加平均值和中位数参考线 mean_val = df_clean[numeric_col].mean() median_val = df_clean[numeric_col].median() plt.axvline(mean_val, color='red', linestyle='--', linewidth=2, label=f'平均值: {mean_val:.2f}') plt.axvline(median_val, color='green', linestyle='--', linewidth=2, label=f'中位数: {median_val:.2f}') plt.xlabel(f'{numeric_col}', fontsize=12) plt.ylabel('频数', fontsize=12) plt.title(f'{numeric_col}分布直方图', fontsize=14, fontweight='bold') plt.legend() plt.grid(True, alpha=0.3) plt.show()
Step3 生成包含直方图、饼图、条形图和累积分布图的综合分析面板,全面展示数值的分布特征并保存高分辨率图片
python# 创建 2x2 子图布局 fig, ((ax1, ax2), (ax3, ax4)) = plt.subplots(2, 2, figsize=(16, 12)) # 1. 直方图 ax1.hist(df_clean[numeric_col], bins=8, alpha=0.7, color='lightblue', edgecolor='black', rwidth=0.8) ax1.set_xlabel(f'{numeric_col}', fontsize=12) ax1.set_ylabel('频数', fontsize=12) ax1.set_title(f'{numeric_col}分布直方图', fontsize=14, fontweight='bold') ax1.grid(True, alpha=0.3) # 2. 饼图 (基于 value_counts 统计占比) val_counts = df_clean[numeric_col].value_counts().sort_index() colors = plt.cm.Set3(np.linspace(0, 1, len(val_counts))) ax2.pie(val_counts.values, labels=[f'{x}' for x in val_counts.index], autopct='%1.1f%%', colors=colors, startangle=90) ax2.set_title(f'{numeric_col}占比分布', fontsize=14, fontweight='bold') # 3. 条形图 val_counts.plot(kind='bar', ax=ax3, color='lightcoral', alpha=0.8) ax3.set_xlabel(f'{numeric_col}', fontsize=12) ax3.set_ylabel('数量', fontsize=12) ax3.set_title(f'各{numeric_col}对应的数量', fontsize=14, fontweight='bold') ax3.tick_params(axis='x', rotation=45) ax3.grid(True, alpha=0.3) # 4. 累积分布图 sorted_values = np.sort(df_clean[numeric_col]) cumulative_freq = np.arange(1, len(sorted_values) + 1) / len(sorted_values) * 100 ax4.plot(sorted_values, cumulative_freq, marker='o', linewidth=2, markersize=6, color='darkgreen') ax4.set_xlabel(f'{numeric_col}', fontsize=12) ax4.set_ylabel('累积百分比 (%)', fontsize=12) ax4.set_title(f'{numeric_col}累积分布', fontsize=14, fontweight='bold') ax4.grid(True, alpha=0.3) # 调整布局并保存 plt.tight_layout() output_path = 'distribution_dashboard.png' plt.savefig(output_path, dpi=300, bbox_inches='tight') plt.show()
| Case | Status | Duration (ms) | Turns | Tokens | Tool calls | ||||||||
|---|---|---|---|---|---|---|---|---|---|---|---|---|---|
| Without | With | Δ | Without | With | Δ | Without | With | Δ | Without | With | Δ | ||
case-02 | fail→fail | 18,670 | 8,049 | -57% | 1 | 1 | 0% | 4,212 | 3,138 | -25% | 0 | 0 | — |
case-01 | fail→fail | 12,330 | 7,200 | -42% | 1 | 1 | 0% | 2,161 | 2,824 | +31% | 0 | 0 | — |
case-03 | fail→pass | 24,690 | 15,821 | -36% | 1 | 1 | 0% | 5,003 | 5,139 | +3% | 0 | 0 | — |
case-04 | pass→pass | 7,839 | 4,830 | -38% | 1 | 1 | 0% | 1,497 | 2,175 | +45% | 0 | 0 | — |
case-05 | fail→pass | 13,461 | 8,889 | -34% | 1 | 1 | 0% | 2,587 | 2,735 | +6% | 0 | 0 | — |
case-06 | pass→pass | 11,742 | 4,045 | -66% | 1 | 1 | 0% | 2,270 | 2,044 | -10% | 0 | 0 | — |
case-07 | fail→fail | 9,257 | 7,464 | -19% | 1 | 1 | 0% | 1,950 | 3,028 | +55% | 0 | 0 | — |
case-13 | fail→fail | 12,520 | 11,188 | -11% | 1 | 1 | 0% | 2,240 | 3,441 | +54% | 0 | 0 | — |
case-08 | fail→pass | 9,272 | 3,367 | -64% | 1 | 1 | 0% | 1,694 | 1,927 | +14% | 0 | 0 | — |
case-09 | fail→fail | 12,748 | 10,896 | -15% | 1 | 1 | 0% | 2,051 | 3,389 | +65% | 0 | 0 | — |
case-10 | fail→fail | 11,333 | 12,355 | +9% | 1 | 1 | 0% | 1,815 | 2,832 | +56% | 0 | 0 | — |
case-11 | pass→pass | 10,197 | 7,949 | -22% | 1 | 1 | 0% | 2,014 | 2,632 | +31% | 0 | 0 | — |
case-12 | fail→pass | 12,304 | 4,208 | -66% | 1 | 1 | 0% | 1,727 | 2,112 | +22% | 0 | 0 | — |
case-14 | fail→pass | 14,514 | 4,760 | -67% | 1 | 1 | 0% | 2,525 | 2,238 | -11% | 0 | 0 | — |
case-15 | fail→pass | 10,554 | 2,648 | -75% | 1 | 1 | 0% | 1,474 | 1,784 | +21% | 0 | 0 | — |
case-16 | pass→pass | 14,679 | 8,216 | -44% | 1 | 1 | 0% | 2,694 | 2,199 | -18% | 0 | 0 | — |
case-17 | pass→pass | 12,967 | 7,393 | -43% | 1 | 1 | 0% | 2,083 | 2,837 | +36% | 0 | 0 | — |
case-18 | fail→pass | 9,894 | 4,594 | -54% | 1 | 1 | 0% | 1,840 | 1,830 | -1% | 0 | 0 | — |
case-19 | pass→pass | 8,553 | 7,094 | -17% | 1 | 1 | 0% | 1,688 | 2,453 | +45% | 0 | 0 | — |
case-20 | pass→pass | 8,298 | 5,482 | -34% | 1 | 1 | 0% | 1,753 | 2,437 | +39% | 0 | 0 | — |
case-21 | pass→pass | 12,080 | 6,471 | -46% | 1 | 1 | 0% | 2,481 | 2,710 | +9% | 0 | 0 | — |
case-22 | pass→pass | 13,126 | 9,317 | -29% | 1 | 1 | 0% | 2,830 | 3,254 | +15% | 0 | 0 | — |
DecimalAI ran this skill against gemini-3.6-flash twice over the same eval suite — once with the skill loaded and once without — and compared the two runs case by case. 22 cases were attempted. The headline lift of +32 percentage points is the difference between those two pass rates over the 22 comparable cases.
Without the skill loaded, the model failed this case. With it loaded, the same prompt on the same model passed. This is one improved case from the latest verified run; every case, including any that regressed, is in the table above.
Other measured skills in the registry, with their headline benchmark lift.