feature-engineering

特征工程技术,包括特征提取、转换、选择和特征存储管理,用于机器学习系统。

📖 中文导读:特征工程技术,包括特征提取、转换、选择和特征存储管理,用于机器学习系统。 以下正文为该技能的原始 SKILL.md 内容(英文),上方安装命令可直接使用。

name: feature-engineering description: Feature engineering techniques including feature extraction, transformation, selection, and feature store management for ML systems.

Feature Engineering

Creating informative features for ML models.

Feature Types

Numerical Features

from sklearn.preprocessing import StandardScaler, RobustScaler

# Scaling
scaler = StandardScaler()  # Mean=0, Std=1
robust_scaler = RobustScaler()  # Robust to outliers

# Log transform (for skewed data)
df['log_income'] = np.log1p(df['income'])

# Polynomial features
from sklearn.preprocessing import PolynomialFeatures
poly = PolynomialFeatures(degree=2, include_bias=False)
X_poly = poly.fit_transform(X)

# Binning
df['age_group'] = pd.cut(df['age'], bins=[0, 18, 35, 55, 100],
                         labels=['youth', 'young_adult', 'middle', 'senior'])

Categorical Features

from sklearn.preprocessing import OneHotEncoder, LabelEncoder

# One-hot encoding
ohe = OneHotEncoder(sparse=False, handle_unknown='ignore')
encoded = ohe.fit_transform(df[['category']])

# Target encoding
def target_encode(df, col, target, smoothing=10):
    global_mean = df[target].mean()
    agg = df.groupby(col)[target].agg(['mean', 'count'])
    smooth = (agg['count'] * agg['mean'] + smoothing * global_mean) / (agg['count'] + smoothing)
    return df[col].map(smooth)

# Hash encoding (for high cardinality)
from sklearn.feature_extraction import FeatureHasher
hasher = FeatureHasher(n_features=100, input_type='string')

Text Features

from sklearn.feature_extraction.text import TfidfVectorizer

# TF-IDF
tfidf = TfidfVectorizer(max_features=5000, ngram_range=(1, 2))
text_features = tfidf.fit_transform(df['text'])

# Embeddings
from sentence_transformers import SentenceTransformer
model = SentenceTransformer('all-MiniLM-L6-v2')
embeddings = model.encode(df['text'].tolist())

# Text statistics
df['text_length'] = df['text'].str.len()
df['word_count'] = df['text'].str.split().str.len()
df['avg_word_length'] = df['text'].str.split().apply(lambda x: np.mean([len(w) for w in x]))

Temporal Features

# Datetime components
df['hour'] = df['timestamp'].dt.hour
df['day_of_week'] = df['timestamp'].dt.dayofweek
df['is_weekend'] = df['day_of_week'].isin([5, 6]).astype(int)

# Cyclical encoding
df['hour_sin'] = np.sin(2 * np.pi * df['hour'] / 24)
df['hour_cos'] = np.cos(2 * np.pi * df['hour'] / 24)

# Lag features
df['lag_1'] = df['value'].shift(1)
df['lag_7'] = df['value'].shift(7)
df['rolling_mean_7'] = df['value'].rolling(window=7).mean()

Feature Selection

from sklearn.feature_selection import SelectKBest, mutual_info_classif

# Filter method
selector = SelectKBest(mutual_info_classif, k=50)
X_selected = selector.fit_transform(X, y)

# Embedded method (tree importance)
from sklearn.ensemble import RandomForestClassifier
rf = RandomForestClassifier()
rf.fit(X, y)
importances = pd.Series(rf.feature_importances_, index=feature_names)

# Recursive Feature Elimination
from sklearn.feature_selection import RFE
rfe = RFE(estimator=LogisticRegression(), n_features_to_select=20)
X_rfe = rfe.fit_transform(X, y)

Feature Store

from feast import Entity, FeatureView, Feature, FileSource

# Define feature view
user_features = FeatureView(
    name="user_features",
    entities=["user_id"],
    features=[
        Feature(name="total_purchases", dtype=Float32),
        Feature(name="avg_order_value", dtype=Float32),
    ],
    ttl=timedelta(days=1),
    source=FileSource(path="data/user_features.parquet")
)

# Get features for training
training_df = store.get_historical_features(
    entity_df=entity_df,
    features=["user_features:total_purchases"]
).to_df()

# Get features for inference
online_features = store.get_online_features(
    entity_rows=[{"user_id": 123}],
    features=["user_features:total_purchases"]
)

Commands

  • /omgfeature:extract - Extract features
  • /omgfeature:select - Select features
  • /omgfeature:store - Feature store ops

Best Practices

  1. Start with simple features
  2. Use domain knowledge
  3. Validate feature distributions
  4. Document feature definitions
  5. Monitor feature drift