菜单
Use when the target is rare (fraud, churn, disease, anomalies). Covers correct metrics, resampling (SMOTE/undersampling), class weights, threshold tuning, and avoiding the accuracy trap and resampling leakage.
用 Codex 或 Claude 帮你安装 复制这段 Prompt，粘贴到 Codex、Claude 或其他助手里，让它检查 Skill 页面并帮你完成安装。
基于 SOC 职业分类
| name | imbalanced-data |
| description | Use when the target is rare (fraud, churn, disease, anomalies). Covers correct metrics, resampling (SMOTE/undersampling), class weights, threshold tuning, and avoiding the accuracy trap and resampling leakage. |
When positives are rare, naive training and naive metrics both mislead. A model that always predicts "negative" can score 99% accuracy and catch zero fraud. Handle imbalance at three levels: metric, algorithm, and threshold — and resample inside cross-validation, never before.
Drop accuracy and ROC-AUC-as-sole-metric. Prefer:
| Technique | How | When |
|---|---|---|
| Class weights | class_weight="balanced" / scale_pos_weight | First choice — no data duplication |
| Undersample majority | RandomUnderSampler | Lots of data, majority redundant |
| Oversample minority | SMOTE / ADASYN | Limited minority samples |
| Combine | SMOTE + Tomek/ENN | Noisy boundaries |
Class weights are the cheapest, leak-free first move:
# sklearn
LogisticRegression(class_weight="balanced")
# XGBoost / LightGBM
scale_pos_weight = n_negative / n_positive
Resampling before CV leaks synthetic neighbors across the split and inflates scores. Use imblearn's pipeline so SMOTE fits on training folds only:
from imblearn.pipeline import Pipeline as ImbPipeline
from imblearn.over_sampling import SMOTE
from sklearn.ensemble import HistGradientBoostingClassifier
from sklearn.model_selection import cross_val_score, StratifiedKFold
pipe = ImbPipeline([
("smote", SMOTE(random_state=42)), # applied to train fold only
("clf", HistGradientBoostingClassifier(random_state=42)),
])
cv = StratifiedKFold(5, shuffle=True, random_state=42)
print(cross_val_score(pipe, X, y, cv=cv, scoring="average_precision").mean())
Default 0.5 is almost always wrong for imbalanced problems. Pick the threshold from the validation PR curve to hit your precision/recall target:
from sklearn.metrics import precision_recall_curve
import numpy as np
prec, rec, thr = precision_recall_curve(y_val, val_scores)
# smallest threshold achieving >= 0.90 precision
ok = np.where(prec[:-1] >= 0.90)[0]
chosen = thr[ok[0]] if len(ok) else 0.5
A model trained with leak-free resampling/weights, scored with PR-AUC and a tuned threshold via model-evaluation, ready for model-serving.
Use when preparing raw data for modeling — handling missing values, duplicates, inconsistent types, outliers, and bad categorical values. Emphasizes fitting all imputation on train-only to avoid leakage.
Use when running ML experiments that need to be compared, reproduced, or shared. Covers MLflow/Weights & Biases logging, what to track, run organization, and model registry basics.
Use when starting on a new dataset, before modeling, or when asked to "explore", "profile", "understand", or "summarize" data. Covers structured EDA, distributions, correlations, target leakage checks, and visualization.
Use when creating, encoding, scaling, or selecting features for ML models. Covers categorical encoding, numeric transforms, datetime/text/aggregation features, and leakage-safe target encoding.
Use when optimizing model hyperparameters. Covers search strategy (random vs Bayesian/Optuna), leakage-safe tuning inside CV, search-space design, early stopping, and budget management.
Use when fine-tuning a large language model. Covers choosing full vs LoRA/QLoRA, dataset formatting, the transformers/PEFT/TRL stack, key hyperparameters, and evaluating fine-tunes without overfitting.