jiachengwang-punch
UserA reusable, multi-model, language-adaptive methodology for end-to-end machine learning analysis of tabular data.
Categories
Indexed Skills (2)
predictive-analytics
A complete, reusable methodology for end-to-end machine learning analysis of structured (tabular) data — data import and cleaning, exploratory analysis, feature engineering, classification/regression modeling, model diagnosis, clustering, interpretability, and final reporting. Use this skill WHENEVER the user wants to analyze a tabular dataset (CSV/Excel), build a predictive model (classification or regression), do EDA, engineer features, evaluate or diagnose a model, cluster/segment records, or produce a data-analysis report or thesis — even if they ask for just one piece (e.g. "build a churn model", "do EDA on this CSV", "what features should I use", "cluster my stores"). Trigger for any "I have a dataset and want to predict/understand Y" task: customer churn, sales/demand forecasting, risk/default prediction, site selection, sensor/monitoring data, and similar. Domain-agnostic; a city-noise dataset is the running example.
small-sample-analysis
End-to-end methodology for supervised machine learning on small datasets (typically 30-200 samples) where standard "throw XGBoost at it" approaches fail. Use this skill whenever the user is building a predictive model on a small dataset, especially when sample-to-feature ratios are tight, when interpretability matters as much as accuracy, when the user needs to justify model choices to non-technical stakeholders, or when they need a rigorous "diagnose-improve-verify" workflow rather than just a final model. Trigger this even if the user only asks for a specific piece (e.g. "help me pick features", "validate this model"), since small-sample problems require the full methodology to avoid silent overfitting. Also trigger for store-selection / site-selection problems, B2B sales analytics, biomedical studies, A/B test analysis with limited cohorts, and any "we only have N stores/patients/experiments and need to predict Y" scenario.
Bio shown is the top-scored skill's repo description as a fallback — real GitHub bios land in a future update.