staran 0.6.0__py3-none-any.whl → 1.0.0__py3-none-any.whl

staran/models/config.py

@@ -1,271 +0,0 @@
-"""
-模型配置管理模块
-
-定义模型的核心配置信息，包括模型类型、参数、特征配置等
-"""
-
-from enum import Enum
-from typing import Dict, Any, List, Optional
-from dataclasses import dataclass, field
-from datetime import datetime
-
-
-class ModelType(Enum):
-    """模型类型枚举"""
-    CLASSIFICATION = "classification"
-    REGRESSION = "regression" 
-    CLUSTERING = "clustering"
-    TIME_SERIES = "time_series"
-    ANOMALY_DETECTION = "anomaly_detection"
-    RECOMMENDATION = "recommendation"
-
-
-class ModelAlgorithm(Enum):
-    """模型算法枚举"""
-    # 分类算法
-    LOGISTIC_REGRESSION = "logistic_regression"
-    RANDOM_FOREST = "random_forest"
-    GRADIENT_BOOSTING = "gradient_boosting"
-    SVM = "svm"
-    NEURAL_NETWORK = "neural_network"
-    
-    # 回归算法
-    LINEAR_REGRESSION = "linear_regression"
-    RIDGE_REGRESSION = "ridge_regression"
-    LASSO_REGRESSION = "lasso_regression"
-    
-    # 聚类算法
-    KMEANS = "kmeans"
-    DBSCAN = "dbscan"
-    HIERARCHICAL = "hierarchical"
-    
-    # 时间序列
-    ARIMA = "arima"
-    LSTM = "lstm"
-    PROPHET = "prophet"
-
-
-@dataclass
-class FeatureConfig:
-    """特征配置"""
-    schema_name: str                    # 使用的schema名称 (如 'aum')
-    table_types: List[str]              # 使用的表类型列表 (如 ['behavior', 'asset_avg'])
-    feature_selection: bool = True      # 是否启用特征选择
-    feature_engineering: bool = True    # 是否启用特征工程
-    scaling: bool = True                # 是否启用特征缩放
-    encoding: Dict[str, str] = field(default_factory=dict)  # 编码配置
-
-
-@dataclass 
-class ModelConfig:
-    """模型配置类"""
-    # 基本信息
-    name: str                           # 模型名称
-    model_type: ModelType               # 模型类型
-    algorithm: ModelAlgorithm           # 使用的算法
-    version: str = "1.0.0"              # 模型版本
-    
-    # 特征配置
-    feature_config: FeatureConfig = None
-    
-    # 模型参数
-    hyperparameters: Dict[str, Any] = field(default_factory=dict)
-    
-    # 训练配置
-    training_config: Dict[str, Any] = field(default_factory=lambda: {
-        'test_size': 0.2,
-        'random_state': 42,
-        'cross_validation': True,
-        'cv_folds': 5
-    })
-    
-    # 评估配置
-    evaluation_metrics: List[str] = field(default_factory=list)
-    
-    # 银行特定配置
-    bank_code: str = "generic"          # 银行代码
-    business_domain: str = "generic"    # 业务领域
-    
-    # 元数据
-    description: str = ""               # 模型描述
-    created_at: datetime = field(default_factory=datetime.now)
-    created_by: str = "system"          # 创建者
-    tags: List[str] = field(default_factory=list)
-    
-    # 部署配置
-    deployment_config: Dict[str, Any] = field(default_factory=dict)
-    
-    def __post_init__(self):
-        """初始化后处理"""
-        if self.feature_config is None:
-            self.feature_config = FeatureConfig(
-                schema_name="generic",
-                table_types=["base"]
-            )
-        
-        # 根据模型类型设置默认评估指标
-        if not self.evaluation_metrics:
-            self.evaluation_metrics = self._get_default_metrics()
-    
-    def _get_default_metrics(self) -> List[str]:
-        """根据模型类型获取默认评估指标"""
-        if self.model_type == ModelType.CLASSIFICATION:
-            return ['accuracy', 'precision', 'recall', 'f1_score', 'auc']
-        elif self.model_type == ModelType.REGRESSION:
-            return ['mae', 'mse', 'rmse', 'r2_score']
-        elif self.model_type == ModelType.CLUSTERING:
-            return ['silhouette_score', 'calinski_harabasz_score']
-        else:
-            return ['custom_metric']
-    
-    def to_dict(self) -> Dict[str, Any]:
-        """转换为字典格式"""
-        return {
-            'name': self.name,
-            'model_type': self.model_type.value,
-            'algorithm': self.algorithm.value,
-            'version': self.version,
-            'feature_config': {
-                'schema_name': self.feature_config.schema_name,
-                'table_types': self.feature_config.table_types,
-                'feature_selection': self.feature_config.feature_selection,
-                'feature_engineering': self.feature_config.feature_engineering,
-                'scaling': self.feature_config.scaling,
-                'encoding': self.feature_config.encoding
-            },
-            'hyperparameters': self.hyperparameters,
-            'training_config': self.training_config,
-            'evaluation_metrics': self.evaluation_metrics,
-            'bank_code': self.bank_code,
-            'business_domain': self.business_domain,
-            'description': self.description,
-            'created_at': self.created_at.isoformat(),
-            'created_by': self.created_by,
-            'tags': self.tags,
-            'deployment_config': self.deployment_config
-        }
-    
-    @classmethod
-    def from_dict(cls, data: Dict[str, Any]) -> 'ModelConfig':
-        """从字典创建ModelConfig实例"""
-        feature_config_data = data.get('feature_config', {})
-        feature_config = FeatureConfig(
-            schema_name=feature_config_data.get('schema_name', 'generic'),
-            table_types=feature_config_data.get('table_types', ['base']),
-            feature_selection=feature_config_data.get('feature_selection', True),
-            feature_engineering=feature_config_data.get('feature_engineering', True),
-            scaling=feature_config_data.get('scaling', True),
-            encoding=feature_config_data.get('encoding', {})
-        )
-        
-        return cls(
-            name=data['name'],
-            model_type=ModelType(data['model_type']),
-            algorithm=ModelAlgorithm(data['algorithm']),
-            version=data.get('version', '1.0.0'),
-            feature_config=feature_config,
-            hyperparameters=data.get('hyperparameters', {}),
-            training_config=data.get('training_config', {}),
-            evaluation_metrics=data.get('evaluation_metrics', []),
-            bank_code=data.get('bank_code', 'generic'),
-            business_domain=data.get('business_domain', 'generic'),
-            description=data.get('description', ''),
-            created_by=data.get('created_by', 'system'),
-            tags=data.get('tags', []),
-            deployment_config=data.get('deployment_config', {})
-        )
-
-
-def create_model_config(
-    name: str,
-    model_type: str,
-    algorithm: str,
-    schema_name: str = "generic",
-    table_types: List[str] = None,
-    bank_code: str = "generic",
-    **kwargs
-) -> ModelConfig:
-    """
-    创建模型配置的便捷函数
-    
-    Args:
-        name: 模型名称
-        model_type: 模型类型 
-        algorithm: 算法名称
-        schema_name: 使用的schema名称
-        table_types: 使用的表类型列表
-        bank_code: 银行代码
-        **kwargs: 其他配置参数
-        
-    Returns:
-        ModelConfig实例
-    """
-    if table_types is None:
-        table_types = ["base"]
-    
-    feature_config = FeatureConfig(
-        schema_name=schema_name,
-        table_types=table_types
-    )
-    
-    return ModelConfig(
-        name=name,
-        model_type=ModelType(model_type),
-        algorithm=ModelAlgorithm(algorithm),
-        feature_config=feature_config,
-        bank_code=bank_code,
-        **kwargs
-    )
-
-
-# 预定义的模型配置模板
-PRESET_CONFIGS = {
-    "aum_longtail_classification": {
-        "model_type": "classification",
-        "algorithm": "random_forest",
-        "schema_name": "aum",
-        "table_types": ["behavior", "asset_avg", "asset_config", "monthly_stat"],
-        "hyperparameters": {
-            "n_estimators": 100,
-            "max_depth": 10,
-            "random_state": 42
-        },
-        "description": "AUM长尾客户分类模型"
-    },
-    
-    "customer_value_regression": {
-        "model_type": "regression", 
-        "algorithm": "gradient_boosting",
-        "schema_name": "aum",
-        "table_types": ["behavior", "asset_avg"],
-        "hyperparameters": {
-            "n_estimators": 150,
-            "learning_rate": 0.1,
-            "max_depth": 8
-        },
-        "description": "客户价值预测回归模型"
-    }
-}
-
-
-def create_preset_config(preset_name: str, **overrides) -> ModelConfig:
-    """
-    基于预设模板创建模型配置
-    
-    Args:
-        preset_name: 预设模板名称
-        **overrides: 覆盖的配置参数
-        
-    Returns:
-        ModelConfig实例
-    """
-    if preset_name not in PRESET_CONFIGS:
-        raise ValueError(f"未知的预设配置: {preset_name}")
-    
-    config = PRESET_CONFIGS[preset_name].copy()
-    config.update(overrides)
-    
-    return create_model_config(
-        name=preset_name,
-        **config
-    )

staran/models/daifa_models.py

@@ -1,361 +0,0 @@
-"""
-新疆工行代发长尾客户专用模型定义
-
-包含两个核心模型：
-1. 代发长尾客户提升3k预测模型
-2. 代发长尾客户防流失1.5k预测模型
-
-基于新疆工行代发长尾客户数据库和业务规则
-"""
-
-from typing import Dict, List
-from .config import create_model_config
-from .target import create_target_definition
-from .registry import ModelRegistry, register_model
-import os
-import json
-from datetime import datetime
-
-
-def save_model_registry(output_path: str):
-    """保存模型注册信息到文件"""
-    
-    def convert_to_serializable(obj):
-        """递归转换对象为可序列化格式"""
-        if isinstance(obj, datetime):
-            return obj.isoformat()
-        elif hasattr(obj, '__dict__'):
-            result = {}
-            for key, value in obj.__dict__.items():
-                result[key] = convert_to_serializable(value)
-            return result
-        elif hasattr(obj, 'value'):  # 枚举类型
-            return obj.value
-        elif isinstance(obj, (list, tuple)):
-            return [convert_to_serializable(item) for item in obj]
-        elif isinstance(obj, dict):
-            return {k: convert_to_serializable(v) for k, v in obj.items()}
-        else:
-            return obj
-    
-    data = {
-        "models": {},
-        "version_history": {},
-        "saved_at": str(datetime.now())
-    }
-    
-    # 获取所有注册的模型
-    for model_id, entry in ModelRegistry._models.items():
-        data["models"][model_id] = {
-            "model_config": convert_to_serializable(entry.model_config),
-            "target_definition": convert_to_serializable(entry.target_definition),
-            "registered_at": entry.registered_at.isoformat(),
-            "status": entry.status,
-            "performance_metrics": entry.performance_metrics
-        }
-    
-    data["version_history"] = ModelRegistry._version_history.copy()
-    
-    with open(output_path, 'w', encoding='utf-8') as f:
-        json.dump(data, f, indent=2, ensure_ascii=False)
-    
-    print(f"✅ 模型注册信息已保存到: {output_path}")
-    return output_path
-
-
-def create_daifa_longtail_upgrade_model() -> Dict:
-    """创建代发长尾客户提升3k预测模型"""
-    
-    # 模型配置
-    model_config = create_model_config(
-        name="xinjiang_icbc_daifa_longtail_upgrade_3k",
-        model_type="classification",
-        algorithm="gradient_boosting",
-        version="1.0.0",
-        schema_name="daifa_longtail",
-        table_types=["daifa_longtail_behavior", "daifa_longtail_asset_avg", 
-                    "daifa_longtail_asset_config", "daifa_longtail_monthly_stat"],
-        hyperparameters={
-            "n_estimators": 300,
-            "learning_rate": 0.05,
-            "max_depth": 12,
-            "min_samples_split": 20,
-            "min_samples_leaf": 10,
-            "subsample": 0.8,
-            "random_state": 42
-        },
-        bank_code="xinjiang_icbc",
-        business_domain="代发长尾客户",
-        description="新疆工行代发长尾客户下个月资产提升3k预测模型",
-        tags=["daifa", "longtail", "upgrade", "3k", "xinjiang_icbc"]
-    )
-    
-    # 目标定义 - 预测下个月提升3k
-    target_definition = create_target_definition(
-        name="daifa_longtail_upgrade_3k_target",
-        target_type="binary_classification",
-        description="新疆工行代发长尾客户下个月资产提升3000元预测目标",
-        sql_query="""
-        WITH customer_baseline AS (
-            -- 获取代发长尾客户基础信息（当月）
-            SELECT 
-                b.party_id,
-                b.asset_total_bal as current_asset,
-                b.salary_amount as current_salary,
-                b.longtail_score,
-                b.upgrade_potential,
-                CASE 
-                    WHEN b.asset_total_bal BETWEEN 10000 AND 100000 THEN 1 
-                    ELSE 0 
-                END as is_daifa_longtail
-            FROM xinjiang_icbc_daifa_hlwj_monthly_stat_wy b
-            WHERE b.data_dt = '{baseline_date}'
-        ),
-        
-        next_month_performance AS (
-            -- 计算下个月的资产变化
-            SELECT 
-                party_id,
-                asset_total_bal as next_month_asset,
-                salary_amount as next_month_salary,
-                monthly_deposit_amount,
-                monthly_withdraw_amount
-            FROM xinjiang_icbc_daifa_hlwj_monthly_stat_wy
-            WHERE data_dt = '{next_month_date}'
-        ),
-        
-        asset_change AS (
-            -- 计算资产变化情况
-            SELECT 
-                cb.party_id,
-                cb.current_asset,
-                nmp.next_month_asset,
-                (nmp.next_month_asset - cb.current_asset) as asset_change,
-                nmp.monthly_deposit_amount,
-                cb.upgrade_potential
-            FROM customer_baseline cb
-            INNER JOIN next_month_performance nmp ON cb.party_id = nmp.party_id
-            WHERE cb.is_daifa_longtail = 1  -- 只关注代发长尾客户
-        )
-        
-        SELECT 
-            party_id,
-            CASE 
-                -- 代发长尾客户资产提升3k的判断标准
-                WHEN asset_change >= 3000  -- 资产增长达到3000元
-                     AND monthly_deposit_amount > asset_change * 0.7  -- 主要通过存入实现
-                     AND upgrade_potential >= 0.6  -- 提升潜力评分较高
-                THEN 1 
-                ELSE 0 
-            END as upgrade_3k_target,
-            
-            -- 辅助分析字段
-            current_asset,
-            next_month_asset,
-            asset_change,
-            monthly_deposit_amount,
-            upgrade_potential
-            
-        FROM asset_change
-        """,
-        target_column="upgrade_3k_target",
-        class_labels=["no_upgrade", "upgrade_3k"],
-        class_weights={"no_upgrade": 1.0, "upgrade_3k": 2.5},  # 提升类样本权重更高
-        time_window="1_month",
-        prediction_horizon="1_month",
-        bank_code="xinjiang_icbc",
-        business_rules={
-            "min_asset_threshold": 10000,      # 代发长尾最小资产
-            "max_asset_threshold": 100000,     # 代发长尾最大资产
-            "upgrade_target_amount": 3000,     # 提升目标金额
-            "deposit_contribution_ratio": 0.7, # 存入贡献占比
-            "min_upgrade_potential": 0.6       # 最小提升潜力
-        }
-    )
-    
-    return {
-        "model_config": model_config,
-        "target_definition": target_definition,
-        "model_type": "upgrade_prediction"
-    }
-
-
-def create_daifa_longtail_churn_model() -> Dict:
-    """创建代发长尾客户防流失1.5k预测模型"""
-    
-    # 模型配置
-    model_config = create_model_config(
-        name="xinjiang_icbc_daifa_longtail_churn_1_5k",
-        model_type="classification",
-        algorithm="random_forest",  # 防流失模型使用随机森林
-        version="1.0.0",
-        schema_name="daifa_longtail",
-        table_types=["daifa_longtail_behavior", "daifa_longtail_asset_avg", 
-                    "daifa_longtail_asset_config", "daifa_longtail_monthly_stat"],
-        hyperparameters={
-            "n_estimators": 200,
-            "max_depth": 10,
-            "min_samples_split": 15,
-            "min_samples_leaf": 8,
-            "max_features": "sqrt",
-            "random_state": 42,
-            "class_weight": "balanced"  # 处理不平衡数据
-        },
-        bank_code="xinjiang_icbc",
-        business_domain="代发长尾客户",
-        description="新疆工行代发长尾客户下个月流失1.5k资产风险预测模型",
-        tags=["daifa", "longtail", "churn", "1_5k", "risk_prevention"]
-    )
-    
-    # 目标定义 - 预测下个月流失1.5k风险
-    target_definition = create_target_definition(
-        name="daifa_longtail_churn_1_5k_target",
-        target_type="binary_classification",
-        description="新疆工行代发长尾客户下个月流失1500元资产风险预测目标",
-        sql_query="""
-        WITH customer_baseline AS (
-            -- 获取代发长尾客户基础信息（当月）
-            SELECT 
-                b.party_id,
-                b.asset_total_bal as current_asset,
-                b.salary_amount as current_salary,
-                b.longtail_score,
-                b.churn_risk,
-                b.login_days,
-                CASE 
-                    WHEN b.asset_total_bal BETWEEN 10000 AND 100000 THEN 1 
-                    ELSE 0 
-                END as is_daifa_longtail
-            FROM xinjiang_icbc_daifa_hlwj_monthly_stat_wy b
-            WHERE b.data_dt = '{baseline_date}'
-        ),
-        
-        next_month_performance AS (
-            -- 计算下个月的资产变化和行为
-            SELECT 
-                party_id,
-                asset_total_bal as next_month_asset,
-                monthly_withdraw_amount,
-                login_days as next_month_login_days
-            FROM xinjiang_icbc_daifa_hlwj_monthly_stat_wy
-            WHERE data_dt = '{next_month_date}'
-        ),
-        
-        churn_analysis AS (
-            -- 分析流失风险情况
-            SELECT 
-                cb.party_id,
-                cb.current_asset,
-                nmp.next_month_asset,
-                (cb.current_asset - nmp.next_month_asset) as asset_decrease,
-                nmp.monthly_withdraw_amount,
-                cb.churn_risk,
-                cb.login_days,
-                nmp.next_month_login_days
-            FROM customer_baseline cb
-            INNER JOIN next_month_performance nmp ON cb.party_id = nmp.party_id
-            WHERE cb.is_daifa_longtail = 1  -- 只关注代发长尾客户
-        )
-        
-        SELECT 
-            party_id,
-            CASE 
-                -- 代发长尾客户流失1.5k的判断标准
-                WHEN asset_decrease >= 1500  -- 资产减少达到1500元
-                     AND monthly_withdraw_amount >= 1500  -- 主要通过取出导致
-                     AND (
-                         churn_risk >= 0.7  -- 流失风险评分高
-                         OR next_month_login_days <= login_days * 0.5  -- 活跃度大幅下降
-                     )
-                THEN 1 
-                ELSE 0 
-            END as churn_1_5k_target,
-            
-            -- 辅助分析字段
-            current_asset,
-            next_month_asset,
-            asset_decrease,
-            monthly_withdraw_amount,
-            churn_risk,
-            login_days,
-            next_month_login_days
-            
-        FROM churn_analysis
-        """,
-        target_column="churn_1_5k_target",
-        class_labels=["no_churn", "churn_1_5k"],
-        class_weights={"no_churn": 1.0, "churn_1_5k": 3.0},  # 流失类样本权重更高
-        time_window="1_month",
-        prediction_horizon="1_month",
-        bank_code="xinjiang_icbc",
-        business_rules={
-            "min_asset_threshold": 10000,        # 代发长尾最小资产
-            "max_asset_threshold": 100000,       # 代发长尾最大资产
-            "churn_threshold_amount": 1500,      # 流失阈值金额
-            "min_churn_risk": 0.7,               # 最小流失风险
-            "activity_decline_ratio": 0.5        # 活跃度下降比例
-        }
-    )
-    
-    return {
-        "model_config": model_config,
-        "target_definition": target_definition,
-        "model_type": "churn_prevention"
-    }
-
-
-def create_both_daifa_models(output_dir: str = "./xinjiang_models") -> Dict:
-    """创建两个代发长尾客户模型并注册"""
-    
-    # 确保输出目录存在
-    os.makedirs(output_dir, exist_ok=True)
-    
-    # 创建提升模型
-    upgrade_model = create_daifa_longtail_upgrade_model()
-    upgrade_id = register_model(
-        upgrade_model["model_config"], 
-        upgrade_model["target_definition"]
-    )
-    
-    # 创建防流失模型
-    churn_model = create_daifa_longtail_churn_model()
-    churn_id = register_model(
-        churn_model["model_config"], 
-        churn_model["target_definition"]
-    )
-    
-    # 保存注册信息到指定目录
-    registry_path = os.path.join(output_dir, "model_registry.json")
-    save_model_registry(registry_path)
-    
-    return {
-        "upgrade_model": {
-            "model_id": upgrade_id,
-            "config": upgrade_model["model_config"],
-            "target": upgrade_model["target_definition"]
-        },
-        "churn_model": {
-            "model_id": churn_id,
-            "config": churn_model["model_config"],
-            "target": churn_model["target_definition"]
-        },
-        "registry_path": registry_path,
-        "output_dir": output_dir
-    }
-
-
-def get_available_daifa_models() -> List[str]:
-    """获取所有可用的代发长尾客户模型"""
-    return [
-        "daifa_longtail_upgrade_3k",   # 代发长尾客户提升3k模型
-        "daifa_longtail_churn_1_5k"    # 代发长尾客户防流失1.5k模型
-    ]
-
-
-# 导出函数
-__all__ = [
-    'create_daifa_longtail_upgrade_model',
-    'create_daifa_longtail_churn_model', 
-    'create_both_daifa_models',
-    'get_available_daifa_models'
-]