메뉴
Professional feature engineering for football prediction. Use when extracting features, selecting important features, transforming data, or implementing V25.1 adaptive extraction engine (48→12061 dimensions).
Codex 또는 Claude로 설치 이 Prompt를 복사해 Codex, Claude 또는 다른 어시스턴트에 붙여 넣으면 Skill 페이지를 검토하고 설치를 진행할 수 있습니다.
SOC 직업 분류 기준
| name | feature-engineering |
| description | Professional feature engineering for football prediction. Use when extracting features, selecting important features, transforming data, or implementing V25.1 adaptive extraction engine (48→12061 dimensions). |
专业的特征工程技能模块，专注于足球预测系统的特征提取、选择、变换和优化。
src/processors/v25_production_extractor.py - V25.1 自适应提取器src/processors/base_extractor.py - 基础提取器抽象类src/processors/exceptions.py - 异常定义# L2 特征解析（使用 V25.1 自适应引擎）
python main_production.py l2-parse --extractor-version V25.1
# 批量处理
python main_production.py l2-parse --batch --limit 50
# 单场比赛
python main_production.py l2-parse --match-id 123456
# 跳过验证
python main_production.py l2-parse --batch --skip-validation
# 输出特征
python main_production.py l2-parse --match-id 123456 --output-features
from src.processors import ExtractorRegistry
# 创建指定版本的提取器
extractor = ExtractorRegistry.create("V25.1")
# 列出可用版本
versions = ExtractorRegistry.list_versions()
# ['V25.1', 'V24.0', 'V23.0']
主队/客队各 8 维:
rolling_xg, rolling_xg_stdrolling_shots_on_target, rolling_shots_on_target_stdrolling_possession, rolling_possession_stdrolling_team_rating, rolling_team_rating_stdhome_table_position, away_table_position, table_position_diffhome_points, away_points, points_diffhome_recent_form_points, away_recent_form_pointsELO 相对差距:
raw_elo_gap, adjusted_elo_gap, fatigue_impact, schedule_impact疲劳度指数:
home_fatigue_index, away_fatigue_index, fatigue_diffhome_rest_days, away_rest_days保级战意:
home_relegation_incentive, away_relegation_incentiveincentive_diff, home_desperationtable_proximity: 积分榜接近度low_scoring_tendency: 低得分倾向elo_diff_cluster: ELO 差距聚类# 完整流程
python main_production.py --full-pipeline
# 分步执行
python main_production.py l1-harvest --season 2425 --target 100
python main_production.py l2-parse --extractor-version V25.1
python main_production.py train --train-size 600 --test-size 160
# 自动分批收割（推荐）
python scripts/auto_harvest_batches.py
# 全量收割
python scripts/run_v26_full_harvest.py --limit 100
from src.processors import ExtractorRegistry, ExtractionResult
from src.config_unified import get_settings
import json
# 创建 V25.1 提取器
extractor = ExtractorRegistry.create("V25.1")
# 加载数据
with open("match_data.json") as f:
raw_data = json.load(f)
# 提取特征
result: ExtractionResult = extractor.extract_with_validation(
raw_data,
skip_validation=False
)
# 检查结果
if result.is_success:
print(f"特征数量: {result.feature_count}")
print(f"状态: {result.status.value}")
print(f"特征: {result.features}")
else:
print(f"错误: {result.errors}")
import xgboost as xgb
from sklearn.datasets import load_iris
# 训练模型
model = xgb.XGBClassifier(
n_estimators=200,
max_depth=3,
learning_rate=0.01
)
model.fit(X_train, y_train)
# 获取特征重要性
importance = model.feature_importances_
# 排序
feature_importance = sorted(
zip(X.columns, importance),
key=lambda x: x[1],
reverse=True
)
# 打印 Top 10
for feature, score in feature_importance[:10]:
print(f"{feature}: {score:.4f}")
from sklearn.feature_selection import RFE
from xgboost import XGBClassifier
# 创建 RFE 选择器
rfe = RFE(
estimator=XGBClassifier(n_estimators=100),
n_features_to_select=50,
step=0.2
)
# 拟合
rfe.fit(X_train, y_train)
# 获取选中的特征
selected_features = X_train.columns[rfe.support_]
print(f"选中特征: {list(selected_features)}")
import shap
# 创建解释器
explainer = shap.TreeExplainer(model)
shap_values = explainer.shap_values(X_test)
# 可视化
shap.summary_plot(shap_values, X_test)
shap.dependence_plot("feature_name", shap_values, X_test)
from sklearn.preprocessing import StandardScaler
scaler = StandardScaler()
X_scaled = scaler.fit_transform(X)
from sklearn.preprocessing import MinMaxScaler
scaler = MinMaxScaler(feature_range=(0, 1))
X_normalized = scaler.fit_transform(X)
import numpy as np
# 对偏态数据应用对数变换
X_log = np.log1p(X) # log(1 + x) 避免 log(0)
from src.processors.v25_production_extractor import (
get_global_feature_keys,
register_feature_keys
)
# 获取全局特征键
all_features = get_global_feature_keys()
# 注册新特征
register_feature_keys({"new_feature_1", "new_feature_2"})
import pandas as pd
# 数值型特征：填充 0
df["numeric_feature"].fillna(0, inplace=True)
# 数值型特征：填充均值
df["numeric_feature"].fillna(df["numeric_feature"].mean(), inplace=True)
# 分类特征：填充众数
df["categorical_feature"].fillna(df["categorical_feature"].mode()[0], inplace=True)
# 前向填充
df.fillna(method="ffill", inplace=True)
import matplotlib.pyplot as plt
# 直方图
df["feature_name"].hist(bins=50)
plt.title("Feature Distribution")
plt.show()
# 箱线图
df.boxplot(column="feature_name")
plt.title("Feature Boxplot")
plt.show()
# 描述性统计
stats = df.describe()
# 缺失值统计
missing = df.isnull().sum()
# 特征类型统计
dtypes = df.dtypes
v26-harvest: V26.1 收割流水线machine-learning-engineering: ML 模型优化data-engineering: ETL 数据管道football-prediction: 足球预测系统Last updated: 2025-12-28 Target: 足球预测系统特征工程优化
Comprehensive API testing including unit tests, integration tests, performance testing, and automated test pipelines. Use when testing REST endpoints, validating API responses, load testing, or setting up test automation.
Manage code quality for Football Prediction project using Make commands. Use when running quality checks, formatting code, running tests, or preparing for commits. Supports black, flake8, mypy, bandit, pytest, and coverage tools.
Collect football data from external APIs including FotMob L2 data, odds information, and real-time match statistics. Use when gathering match data, collecting historical statistics, or updating prediction datasets.
Design and optimize data pipelines for football prediction system. Use when building ETL processes, data transformations, or optimizing data flow between PostgreSQL, Redis, and external APIs.
Manage PostgreSQL database operations including migrations, query optimization, connection pooling, and data backup/restore. Use when handling database tasks, optimizing queries, or managing database connections.
Manage application deployment including Docker containers, production environments, and service orchestration. Use when deploying to production, managing Docker services, or handling deployment rollbacks.