ML-SHAP Skill

A reusable Codex skill for complete, leakage-aware machine-learning analysis of CSV and Excel tabular data.

该技能用于执行完整、可复现且防数据泄漏的表格机器学习流程，支持回归、分类、时间序列、分组数据与 SHAP 可解释性分析。

Capabilities

• Audit missing values, duplicates, data types, outliers, target balance, and leakage risks.
• Select validation strategies for IID, temporal, grouped, and spatial data.
• Compare naive, linear, tree-ensemble, boosting, and other justified models.
• Tune models without contaminating the final holdout set.
• Evaluate regression and classification with task-appropriate metrics and diagnostics.
• Explain predictions using SHAP; include SHAP beeswarm / 蜂群图（蜂巢图） to show feature importance, contribution direction, feature-value distribution, and sample heterogeneity.
• Fall back transparently to permutation importance and PDP/ALE when SHAP is unavailable.
• Analyze nonlinear effects, interactions, representative cases, subgroups, and temporal regimes.
• Run robustness and sensitivity checks.
• Deliver publication-ready figures, Excel tables, reports, reproducible code, and optional JSON summaries.

Skill structure

ml-shap/
├── SKILL.md
├── agents/
│   └── openai.yaml
└── references/
    ├── figures-and-tables.md
    └── quality-gates.md

Installation

Copy the ml-shap directory into your Codex skills directory:

~/.codex/skills/ml-shap

Restart or refresh Codex so the skill can be discovered.

Usage

Invoke the skill explicitly:

Use $ml-shap to run a complete machine-learning and SHAP analysis on my dataset.

Example specification:

File: data.xlsx
Worksheet: Sheet1
Output: carbon_emission
Inputs: GDP, population, temperature, urbanization
Task: regression
Time column: year
Group column: city
Research focus: nonlinear GDP effect and threshold

If task type or validation structure is uncertain, the skill audits the data and records defensible assumptions before modeling.

Expected outputs

• Data-quality and descriptive-statistics tables
• Validation design and model-comparison results
• Tuned model parameters and untouched-test predictions
• Regression or classification diagnostics
• SHAP importance bar chart and SHAP beeswarm / 蜂群图（蜂巢图）
• SHAP dependence, interaction, and local waterfall explanations
• Robustness and sensitivity analyses
• Publication-ready PNG plus PDF/SVG figures
• Excel workbook, written report, reproducible script, and optional JSON summary

Scientific safeguards

• Same-period estimation is not presented as future forecasting.
• Preprocessing and feature selection are fitted inside training folds.
• Temporal, grouped, and spatial structures receive matching validation designs.
• Predictive explanations are not described as causal effects.
• SHAP substitutes are labeled accurately rather than being presented as SHAP.

License

No license has been specified yet. Add a license before redistributing or incorporating the skill into another project.

ML-SHAP Skill

A reusable Codex skill for complete, leakage-aware machine-learning analysis of CSV and Excel tabular data.

该技能用于执行完整、可复现且防数据泄漏的表格机器学习流程，支持回归、分类、时间序列、分组数据与 SHAP 可解释性分析。

Capabilities

• Audit missing values, duplicates, data types, outliers, target balance, and leakage risks.
• Select validation strategies for IID, temporal, grouped, and spatial data.
• Compare naive, linear, tree-ensemble, boosting, and other justified models.
• Tune models without contaminating the final holdout set.
• Evaluate regression and classification with task-appropriate metrics and diagnostics.
• Explain predictions using SHAP; fall back transparently to permutation importance and PDP/ALE when necessary.
• Analyze nonlinear effects, interactions, representative cases, subgroups, and temporal regimes.
• Run robustness and sensitivity checks.
• Deliver publication-ready figures, Excel tables, reports, reproducible code, and machine-readable summaries.

Skill structure

ml-shap/
├── SKILL.md
├── agents/
│   └── openai.yaml
└── references/
    ├── figures-and-tables.md
    └── quality-gates.md

Installation

Copy the ml-shap directory into your Codex skills directory:

~/.codex/skills/ml-shap

Restart or refresh Codex so the skill can be discovered.

Usage

Invoke the skill explicitly:

Use $ml-shap to run a complete machine-learning and SHAP analysis on my dataset.

Example specification:

File: data.xlsx
Worksheet: Sheet1
Output: carbon_emission
Inputs: GDP, population, temperature, urbanization
Task: regression
Time column: year
Group column: city
Research focus: nonlinear GDP effect and threshold

If task type or validation structure is uncertain, the skill audits the data and records defensible assumptions before modeling.

Expected outputs

• Data-quality and descriptive-statistics tables
• Validation design and model-comparison results
• Tuned model parameters and untouched-test predictions
• Regression or classification diagnostics
• SHAP importance, beeswarm, dependence, interaction, and local explanations
• Robustness and sensitivity analyses
• Publication-ready PNG plus PDF/SVG figures
• Excel workbook, written report, reproducible script, and optional JSON summary

Scientific safeguards

• Same-period estimation is not presented as future forecasting.
• Preprocessing and feature selection are fitted inside training folds.
• Temporal, grouped, and spatial structures receive matching validation designs.
• Predictive explanations are not described as causal effects.
• SHAP substitutes are labeled accurately rather than being presented as SHAP.

License

No license has been specified yet. Add a license before redistributing or incorporating the skill into another project.

Example

上传时可以这样说明：

Excel文件：data.xlsx
工作表：Sheet1
输出变量：CarbonEmission
输入变量：GDP、Population、Temperature、Urbanization
任务类型：回归
时间变量：Year
地区变量：City
不参与建模：CityName、备注
研究重点：识别GDP的非线性影响和阈值